GIFSQ ID: FTN-0091
Internet of Things in Food Industry
Unit 1: What the Internet of Things Is
Note
Unit 1: What the Internet of Things Is
A question to start with
A cold store holds pallets of yoghurt. At two in the morning the cooling unit fails. Nobody is in the building. By six in the morning, when the first worker arrives, the yoghurt has warmed past its safe limit and the whole store must be thrown away. Now picture the same cold store with a small temperature sensor on the wall, a sensor that sends a reading to a phone every few minutes. At two in the morning the phone of the duty manager buzzes: the temperature is rising. She calls the engineer, the fault is fixed before dawn, and the yoghurt is saved. Nothing about the yoghurt changed. Nothing about the cooling unit changed. The difference is that the cold store could now speak, and somebody could hear it in time to act.
That is the plain idea behind the Internet of Things. Ordinary things, a cold store, a milk tank, a grain silo, a delivery truck, are given a way to measure what is happening to them and a way to send that measurement somewhere it can be seen and used. This unit explains what that means, how the parts fit together, and why it matters so much in food work, where temperature, time, moisture, and cleanliness decide whether food is safe to eat.
Questions to think about
- A fridge at home keeps food cold, but it cannot tell you when it stops working. What would change if it could send you a message?
- A truck carries fresh milk for hours. Which facts about that journey would you want to know while it is still on the road, and which facts could wait until it arrives?
- If every machine in a food plant reported its condition all day, what would a plant manager do with all that information?
Think about the cold store question before you read on
If the fridge could send a message, the failure would stop being a surprise found hours later. A warning at the moment the temperature starts to rise gives time to move the food, fix the fault, or call for help. The value is not the message itself: it is the time the message buys.
Things that can speak
The Internet of Things, often shortened to IoT, is a name for a simple arrangement: physical things fitted with sensors and a connection to a network, so that they can collect information about themselves and their surroundings and share it with people or with other machines. The "things" can be large, like a pasteuriser or a cold store, or small, like a single crate of tomatoes. The "internet" part means the information travels over a network, the same kind of network that carries email and web pages, so that a reading taken in a warehouse can be seen on a screen in an office far away, or on a phone in a pocket.
Three parts are always present. First, there is the thing itself and the sensor that measures something about it. Second, there is the connection that carries the measurement away. Third, there is the place where the measurement is received, stored, shown, and used to make a decision. If any one of these three parts is missing, the system is not complete. A sensor with no connection is a thermometer nobody reads. A connection with no sensor carries nothing. A screen with no sensor behind it shows nothing worth seeing.
It helps to compare an IoT system with the human body. A sensor is like a sense organ: it notices temperature, pressure, movement, or light. The network is like the nerves: it carries the signal to the brain. The software that receives the data is like the brain: it notices a pattern, judges whether something is wrong, and decides what to do. The action that follows, switching on a fan, sending an alert, stopping a line, is like a hand reaching out. The comparison is not perfect, but it shows why the parts must work together, and why a system that only collects data without acting on it is only half built.
Quick self-check
Question: A temperature sensor in a cold store records readings on a memory card, but nobody looks at the card for a week. Is that an IoT system?
No. The sensor measures, but the reading is not sent anywhere it can be seen and used in time. The connection and the place where the data is received and acted on are missing, so the store cannot warn anyone while the fault is still small.
Question: Why does the definition insist on three parts: sensor, connection, and receiver?
Because each part does a job the others cannot do. The sensor measures, the connection carries, and the receiver stores, shows, and supports a decision. Take one away and the chain breaks: measurement without a carrier stays local, and a carrier without measurement carries nothing.
A short history in plain terms
The idea did not arrive in one step. Long before the name existed, factories used instruments to measure temperature and pressure, and control rooms used dials and charts to watch them. Those instruments were wired to one place, and a person had to be there to read them. Then small computers became cheap enough to place inside machines, and wireless radio became cheap enough to carry short messages. When those two changes met, a sensor no longer needed a person standing beside it or a cable running to a control room. It could send its reading through the air to a computer that might be in the next room or on another continent.
The name itself came from work on connecting everyday objects to networks so that they could be tracked and identified. Early examples were simple: a pallet that could report where it was, a container that could report whether it had been opened. As sensors became smaller and cheaper, the same idea spread into farming, transport, warehouses, and food plants. The food industry adopted it for a plain reason: food spoils quietly, and spoilage is expensive and dangerous.
It is worth noting what IoT is not. It is not a single machine, and it is not a single brand of equipment. It is a way of arranging parts that come from many makers: sensors, radios, computers, and software. It is also not the same as automation, although the two often work together. Automation means a machine does a job by itself, such as a filler that fills bottles without a person holding each one. IoT means the machine reports what it is doing, so people can see, check, and improve it. A plant can be automated without IoT, and it can have IoT sensors without being automated. The best food plants often have both.
Quick self-check
Question: A filling machine fills bottles by itself, but it reports nothing. Is it automated, IoT, both, or neither?
It is automated, because it does the job by itself, but it is not IoT, because it does not measure and share information about its work. Adding sensors and a connection that report fill levels and stoppages would add the IoT part.
The four layers of an IoT system
Every IoT system, large or small, can be described in four layers. The layers are a way of thinking, not four boxes that must be bought separately. Keeping them in mind helps when a system goes wrong, because the fault is usually in one layer, and knowing which layer saves time.
The first layer is sensing. This is where the physical world is measured: temperature in a cold store, moisture in grain, pressure in a steam line, vibration in a motor, the position of a truck. Sensors sit in this layer, and so do the small devices that switch things on and off, called actuators, such as a valve that opens or a fan that starts.
The second layer is connectivity. This is how the measurement travels from the sensor to somewhere it can be used. The journey may be short, across a room on a wireless link, or long, across a mobile phone network or the internet. This layer also includes the small gateway device that often sits between many sensors and the wider network, collecting their messages and passing them on in one stream.
The third layer is data processing and storage. This is where readings are kept, cleaned, and turned into useful form. Raw readings arrive as numbers with a time attached: 4.2 degrees at 02:14, 4.5 degrees at 02:19. Processing turns that stream into answers: the average temperature over an hour, the highest reading in the night, the moment the temperature passed a limit. Storage keeps the history so that a question asked next week can still be answered.
The fourth layer is application. This is what the user sees and uses: a dashboard on a screen, an alert on a phone, a report for a quality manager, a command sent back to a machine. This layer is where decisions are made, by a person reading a chart or by software that follows a rule, such as "if the temperature rises above the limit for ten minutes, send an alert and start the backup cooler."
The diagram shows the four layers in order. The steps point to the diagram: a reading starts at the sensor, travels through the network, is stored and processed, and ends as something a person can act on.
Quick self-check
Question: A dashboard shows a flat line at 4 degrees all night, but the cold store actually warmed up. Which layer is most likely at fault?
The sensing layer or the connectivity layer. Either the sensor stopped measuring the real temperature, or its readings stopped reaching the dashboard. The application layer can only show what it receives, so a flat line that hides a real change points to a fault below it.
Question: Where does an alert belong in the four layers?
In the application layer, because it is the part the user sees and acts on. The rule that decides when to alert lives in the processing layer, and the sensor reading that triggers it starts in the sensing layer.
Why food needs IoT more than most industries
Food is different from most manufactured goods in three ways, and each one makes IoT valuable. First, food changes with time even when nothing is done to it. Milk sours, bread stales, fruit ripens and then rots, fat goes rancid. Time and temperature drive these changes, so knowing the temperature history of a food is often the same as knowing its condition. Second, food safety depends on conditions that cannot be seen. A crate of meat can look fine and still have spent hours too warm. A sensor sees what the eye cannot. Third, food passes through many hands: farm, collection centre, factory, warehouse, truck, shop. At each handover, information can be lost. A system that records conditions at every step keeps the chain visible from start to finish.
Consider a simple chain. Milk is collected from farms, cooled in a tank, carried by truck, received at a dairy, pasteurised, packed, stored, and delivered to shops. Without sensors, the dairy knows the milk temperature when the truck arrives, but not what happened during the journey. With sensors in the tank and the truck, the dairy knows the whole history. If a load arrives warm, the record shows whether the fault was at the farm, on the road, or at reception, and the right place can be fixed.
The same logic applies inside a plant. A pasteuriser must hold milk at a set temperature for a set time. An IoT system records readings every few seconds, stored automatically and checked against the limit at once, so a question about a batch made weeks ago can still be answered. The principle has not changed: hold the right condition, prove it was held, act at once when it is not.
There is also a waste argument. Food that is thrown away wastes the land, water, feed, fuel, and labour that produced it. Much food waste in cold chains comes from failures that are found too late: a door left open, a cooler that fails overnight, a truck delayed in heat. Early warning turns many of those losses into small fixes. This is not only a money argument. It is a food security argument: every load saved is food that reaches people instead of a bin.
Quick self-check
Question: Why is a temperature history more useful than a single temperature reading on arrival?
A single reading shows the condition at one moment. A history shows how long the food spent at each temperature, and when the change happened, so the cause can be found and the safety of the food can be judged fairly.
Question: Name one food loss that early warning could prevent.
A cold store that fails overnight. An alert in the first minutes lets someone fix the cooler or move the food before it warms past its safe limit, instead of finding the loss in the morning.
What an IoT system does all day
It helps to walk through a day in a food plant that uses IoT, and to name the jobs the system does. The jobs fall into four groups: watching, warning, recording, and improving.
Watching means showing the present condition. A dashboard shows the temperature of each cold store, the pressure in each steam line, the level in each tank, and the running state of each motor. A manager can look at one screen and see the whole plant, instead of walking from room to room reading dials. Watching also means seeing change over time: a slow rise in the temperature of a freezer across a week may mean ice is building up, or a seal is failing, long before the limit is crossed.
Warning means speaking up when a condition leaves its safe range. An alert is a message sent at once, by text, sound, or light, when a rule is broken. Good alerts are specific: they name the place, the value, and the time, such as "Cold store 2, 7.1 degrees, 02:14, limit 5 degrees." Good alerts are also rare enough to be taken seriously. A system that sends an alert every hour for a small, harmless change trains people to ignore it, and then the real warning is ignored too. Setting the limit, the delay before alerting, and the person who receives the alert are quality decisions, not only technical ones.
Recording means keeping a history that can be checked later. Every reading carries a time, so the history can answer questions such as: was this batch held at the right temperature, when did the cooler fail, how long was the door open? Records support traceability, the ability to follow a food from its origin to the shop shelf, and they support audits, where an inspector asks for proof that a process was controlled. Paper charts can do some of this, but paper is hard to search, easy to lose, and slow to check. Digital records can be searched in seconds and copied without loss.
Improving means using the history to make the plant better. If the data shows that one cold store warms up every afternoon, the cause may be sunlight on a wall, or deliveries at midday, and the fix may be shading, a change in delivery time, or a bigger cooler. If the data shows that one motor vibrates more each month, it can be serviced before it fails during production. This use of data, finding patterns and fixing causes, is where the long-term value lies. The sensor is only the start: the improvement is the point.
The diagram shows the four jobs in order. The steps point to the diagram: watching finds the present, warning catches the fault, recording keeps the proof, and improving stops the fault from coming back.
Quick self-check
Question: A plant receives 40 alerts a day, and staff have started ignoring them. What is wrong?
The alert rules are too sensitive, or the limits are set wrong. Alerts must be rare and specific enough that each one is worth acting on. The fix is to review the limits and the delay before alerting, not to add more alerts.
Question: How does recording help an audit?
An auditor asks for proof that a process was controlled on a given date. A stored, time-stamped history gives that proof in seconds, while a lost paper chart gives nothing.
Examples from across the food chain
A few plain examples show how wide the idea reaches. On a farm, soil moisture sensors tell a grower when a field needs water, so water is used only where it is needed. In a grain store, temperature and moisture sensors inside a silo warn when a hot spot forms, which is a sign that grain is heating and may spoil. At a dairy, a sensor in a milk tank watches the temperature from milking to collection, and a sensor on the truck watches it on the road. In a bakery, an oven sensor records the temperature curve of every bake, so a pale loaf can be traced to a cool oven rather than a wrong recipe. In a packing hall, a sensor counts packs, watches the seal temperature, and flags a sealing jaw that is cooling down before it starts making weak seals. In a shop, a sensor in a display chiller watches the temperature through the day and alerts staff before food warms.
Notice that each example follows the same pattern: something is measured, the reading travels, and somebody acts. The measured thing changes, but the pattern stays the same.
One more example shows the system working end to end. A crate of fresh fish is packed with ice and a small temperature sensor. The sensor records the temperature every few minutes and sends it to a dashboard. At the market, the buyer can see the whole journey: the temperature stayed low, with one short rise during loading. The buyer accepts the fish with confidence, and the record stays with the batch. The sensor turns a claim, "kept cold all the way," into a record that can be checked.
Quick self-check
Question: A silo sensor warns of a hot spot in stored grain. What should happen next?
Someone should check the grain at that spot, move air through it or move the grain, and find why it heated: moisture, insects, or poor airflow. The warning is the start of the work, not the end of it.
What IoT cannot do, and what can go wrong
A clear view of IoT includes its limits. First, a sensor measures only what it is placed to measure, and only where it is placed. A temperature sensor on one wall of a cold store does not know the temperature at the door, or inside a pallet in the middle. Placing sensors in the right places, and enough of them, is a skill in itself, and a system with too few sensors can give false comfort. Second, a sensor can drift: its readings slowly move away from the truth as it ages or gets dirty. A temperature probe coated in ice, a humidity sensor fouled with flour dust, a level sensor coated in product, will all report wrong values with complete confidence. Regular checking against a known reference, called calibration, is the guard against drift, and it is covered in detail in Unit 2.
Third, a network can fail. Wireless signals are blocked by metal walls, crowded by other radios, and lost when power fails. A system that depends on a live connection must have a plan for the gap: sensors that store readings and send them later, a local alarm that sounds even when the network is down, and a person who checks the plant when the dashboard goes blank. A blank dashboard must be treated as a warning, not as a sign that all is well.
Fourth, data can mislead when it is read without sense. A single high reading may be a door opened for loading, not a cooler failure. A flat line may be a stuck sensor, not a steady temperature. Software can help by showing trends and by waiting a few minutes before alerting, but judgement stays human. The person who reads the data must know the plant, the product, and the process, or the numbers will be misread.
Fifth, more data is not always better. A sensor that reports every second produces far more numbers than anyone can read, and storing and sorting them costs time and money. Good systems choose a sensible rate: fast enough to catch a real change, slow enough to stay readable. A cold store changes slowly, so a reading every few minutes is enough. A filling line changes fast, so a reading every second may be needed. The rate must match the process.
Quick self-check
Question: A sensor reports 4 degrees all day, every day, without ever changing by even a tenth. What should you suspect?
A stuck or failed sensor, or a reading that is not reaching the dashboard and is being repeated. A real cold store varies a little as doors open, goods arrive, and the cooler cycles. A perfectly flat line is a reason to check the sensor, not to relax.
Question: Why is calibration a food safety matter, not only a technical chore?
Because every safety decision rests on the sensor reading. If the sensor reads two degrees low, food held at a dangerous temperature looks safe on the dashboard. The error is invisible until the sensor is checked against a known reference.
The words you will meet
IoT brings a set of words that appear in every unit of this subject. A sensor measures a physical quantity and turns it into a signal. An actuator does the opposite: it takes a signal and makes a physical change, such as opening a valve. A gateway collects messages from many sensors and passes them to the wider network. A dashboard is a screen that shows the present state of a system in one view. An alert is a message sent when a rule is broken. Telemetry is the sending of measurements from a distant place. A digital twin is a computer model of a physical thing, kept up to date with live sensor readings, so that the model can be used to test "what if" questions without touching the real plant. The cloud is a name for computers reached over the internet that store data and run software for many users. Edge computing is the opposite idea: doing some of the processing close to the sensor, on a small computer at the plant, so that a decision can be made even when the internet is down. Both have a place, and Unit 6 explains when to use each.
The learner does not need to memorise these words as a list. They will return in each unit, attached to a real food example. The glossary at the end of this unit gives a first, plain definition for each one, with a food example.
Try it yourself
Walk through your kitchen and list five places where a sensor would help: the fridge, the freezer, a pantry shelf, a cooking pot, a delivery bag. For each place, write down four things: what you would measure, where exactly you would put the sensor, how often the reading should be sent, and who should get the alert. Then pick one and draw the four layers for it on paper. Mark one weak point in your sketch and write one sentence saying how you would guard it.
Key ideas to carry forward
- The Internet of Things connects physical things to a network so they can measure, report, and be acted on from a distance.
- Three parts are always present: a sensor that measures, a connection that carries, and a receiver that stores, shows, and supports decisions.
- Four layers describe every system: sensing, connectivity, data processing and storage, and application.
- Food needs IoT because food changes with time, safety depends on conditions the eye cannot see, and food passes through many hands.
- An IoT system watches, warns, records, and improves: the improvement, fixing causes found in the history, is where the long-term value lies.
- Sensors measure only where they are placed, they drift and need calibration, networks fail, and data needs human judgement.
- A blank dashboard is a warning, not a sign that all is well, and a perfectly flat reading is a reason to check the sensor.
Glossary
Actuator
A device that takes a signal and makes a physical change, such as opening a valve or starting a fan. Food tech example: an actuator opens a cooling valve when a sensor reports that a cold store is warming.
Alert
A message sent at once when a measured value passes a set limit. Food tech example: an alert tells the duty manager when a freezer rises above its safe temperature.
Cloud
Computers reached over the internet that store data and run software for many users. Food tech example: temperature records from many cold stores are stored in the cloud so a manager can check them from any office.
Connectivity
The network link that carries sensor readings to where they are stored and used. Food tech example: a wireless link carries readings from a grain silo to the farm office.
Dashboard
A screen that shows the present state of a system in one view, such as temperatures, levels, and running machines. Food tech example: a dairy dashboard shows the temperature of every milk tank at a glance.
Data processing
Turning raw readings into useful answers, such as averages, highest values, and times when a limit was passed. Food tech example: processing turns a day of temperature readings into the highest value and the time it happened.
Digital twin
A computer model of a physical thing, kept up to date with live sensor readings. Food tech example: a digital twin of a cold store lets an engineer test what would happen if a door were left open, without touching the real store.
Edge computing
Doing some data processing close to the sensor, on a small computer at the plant, instead of sending everything far away. Food tech example: a gateway at a bakery processes oven readings locally so an alarm still sounds when the internet is down.
Gateway
A device that collects messages from many sensors and passes them on to the wider network in one stream. Food tech example: a gateway in a packing hall gathers readings from a dozen sensors and sends them to the office system.
Internet of Things (IoT)
A system in which physical things fitted with sensors and a network connection can collect information and share it with people or other machines. Food tech example: a milk truck fitted with temperature sensors is part of an IoT system that reports the milk temperature during the journey.
Sensor
A device that measures a physical quantity, such as temperature or pressure, and turns it into a signal. Food tech example: a temperature sensor inside a freezer measures how cold the food is.
Telemetry
The sending of measurements from a distant place to a receiver. Food tech example: telemetry sends the temperature of a truckload of fish to the buyer while the truck is still on the road.
Traceability
The ability to follow a food and its history from origin to the shop shelf. Food tech example: temperature records linked to a batch number let a yoghurt maker trace one pot back to the milk it came from.
Calibration
Checking a sensor against a known reference and correcting it so its readings are true. Food tech example: a cold store thermometer is checked in iced water and corrected if it reads wrong.
Automation
A machine doing a job by itself without a person holding or guiding each step. Food tech example: a filling machine that fills bottles by itself is automated, even if it reports nothing about its work.
References
- Singh, R. P., Computer Applications in Food Technology, Academic Press, 1996.
- Irudayaraj, J., Food Processing Operations Modeling, Design, and Analysis, 2nd edition, 2002.
- Nielsen, S. S., Introduction to the Chemical Analysis of Foods, 1997.
Unit 2: Sensors: the Eyes and Ears of the Food Plant
Note
Unit 2: Sensors: the Eyes and Ears of the Food Plant
A question to start with
Two cold stores sit side by side in a dairy. Both hold the same yoghurt, both have a temperature sensor on the wall, and both dashboards read 4 degrees. In the first store the yoghurt is fine. In the second store, a pallet in the middle has warmed to 9 degrees and must be thrown away. How can both dashboards read 4 degrees when one store has a warm pallet? The answer is placement. The sensor in the second store sits beside the cooling unit, where the air is coldest. It measures the cold air near the cooler, not the warm air trapped inside a tightly packed pallet. The sensor was working, the network was working, and the dashboard was working. The measurement itself was in the wrong place.
This unit is about sensors: what they measure, how they work in plain terms, where to put them, and how to know whether to trust them. A food plant is full of things worth measuring, and each one needs the right sensor in the right place, checked often enough to stay true.
Questions to think about
- If you could measure only three things in a bakery, which three would you pick, and why those three?
- A sensor is cheap, so why not put a hundred of them in every room and measure everything?
- A thermometer reads 4 degrees, but the food beside it feels warm to your hand. Which one do you trust, and what do you do next?
Think about the two cold stores before you read on
The sensor measures the air at its own spot, not the whole store. Air near a cooler is colder than air inside a pallet, and air at the ceiling differs from air at the floor. One sensor can only speak for its own place, so a large store needs sensors in several places: near the door, in the middle, at the warmest corner, and inside a pallet as well as on the wall.
What a sensor does
A sensor takes a physical quantity from the world and turns it into a signal that a computer can read. Temperature is a physical quantity: a cold store is cold, a hot oven is hot. The sensor turns that fact into a number, such as 4.2 degrees. Pressure, humidity, light, movement, weight, flow, and vibration are all physical quantities, and each one has sensors built to measure it. Some sensors measure directly, touching the thing they measure: a probe pushed into a vat of milk reads the milk temperature. Others measure without touching: a camera watches the colour of biscuits on a belt, and a sensor above a tank measures the level by timing a sound pulse that bounces back from the surface.
Every sensor has four qualities that matter in food work. The first is range: the span of values it can measure. A sensor built for a freezer may read from -30 to 10 degrees, while a sensor built for an oven may read from 50 to 300 degrees. Using a freezer sensor in an oven will damage it or give a useless reading at the top of its range. The second is accuracy: how close its reading is to the true value. An accuracy of plus or minus 0.5 degrees means the true temperature may be half a degree above or below the reading. In food safety work, half a degree can matter, so the accuracy must match the decision the reading will support. The third is response time: how fast the sensor notices a change. A thick probe in a steel sheath is slow to warm up, so it may miss a short rise in temperature, while a thin probe reacts fast but breaks easily. The fourth is robustness: whether the sensor survives the place it works in. A food plant is wet, steamy, dusty with flour, washed down with hot water and cleaning chemicals, and full of vibration. A sensor that works well on a laboratory bench may fail in a week on a factory floor.
Quick self-check
Question: A sensor is accurate to plus or minus 2 degrees and is used to check a limit of 5 degrees. It reads 4 degrees. Is the food certainly safe?
No. The true temperature could be as high as 6 degrees, above the limit. The accuracy is too loose for a decision that close to the limit. A more accurate sensor is needed, or the working limit must be set lower to allow for the error.
Question: Why does response time matter for a sensor in a pasteuriser?
Because the temperature in a pasteuriser can change in seconds, and a slow sensor would still show the old temperature after the real one has risen or fallen. A slow reading can hide a short failure that leaves food undertreated.
Temperature sensors: the workhorse of food IoT
Temperature is the most measured quantity in food work, because temperature drives spoilage, safety, and quality. Several kinds of temperature sensor are used, and each suits a different place.
The resistance temperature detector, often called an RTD, uses a small piece of metal whose electrical resistance changes in a steady, known way as temperature changes. It is accurate and stable, so it is used where exact readings matter, such as in a pasteuriser or a cold store that holds food near its limit. It is also slower and more costly than some other kinds, and its thin wires need care.
The thermocouple uses two different metals joined together. The join produces a tiny voltage that changes with temperature. Thermocouples are tough, cheap, and can measure very high temperatures, so they are used in ovens, fryers, and steam lines. They are less accurate than an RTD at low temperatures, and they need a reference point to give a true reading, so they suit hot processes better than cold stores.
The thermistor is a small ceramic piece whose resistance changes sharply with temperature. It is cheap, small, and fast, so it is used in small devices, such as a sensor inside a delivery box or a probe in a small chiller. Its range is narrow, and its response is less steady over a wide span, so it suits a narrow job in a known range.
The infrared sensor measures temperature without touching, by reading the heat radiation from a surface. It is used to check the surface temperature of food on a belt, or the temperature of a seal, where touching would disturb the product or slow the line. It reads the surface only, not the inside, and it can be fooled by shiny surfaces or steam, so it is a surface tool, not a core temperature tool.
Choosing between them is a trade: accuracy against cost, speed against toughness, range against precision. The right choice starts with the question: what temperature, in what place, how exact must it be, and how fast must it react? The answer to that question picks the sensor, not the catalogue.
Quick self-check
Question: You need to check the core temperature of a cooked pie as it leaves the oven. Which sensor type is wrong for the job, and why?
An infrared sensor is wrong for the core, because it reads the surface only. The crust may be hot while the centre is still cool. A probe pushed into the centre is needed for the core temperature.
Question: Why would a pasteuriser use an RTD rather than a cheap thermistor?
Because the pasteuriser temperature decides safety, and the reading must be accurate and stable over time. An RTD gives that accuracy and stability, while a thermistor is built for narrow, less exact work.
Humidity, moisture, and gas sensors
Water matters in food as much as temperature. Humidity is the amount of water vapour in air, and it decides whether bread goes stale or soft, whether powder clumps, and whether mould grows on a stored product. A humidity sensor in a store or a packing hall measures the air, and the reading guides ventilation and dehumidifying. In a dry store for flour or milk powder, low humidity keeps the powder flowing and stops clumping. In a ripening room for cheese, controlled humidity keeps the rind from drying and cracking.
Moisture in a solid food is a different measurement. A moisture sensor in grain measures the water inside the kernels, often by passing a small electrical signal through the grain and reading how it responds, since water changes the electrical behaviour of the material. Grain that is too wet will heat and spoil in storage, while grain that is too dry loses weight and value. A sensor in a silo or at the intake of a dryer lets the operator dry the grain just enough, saving fuel and keeping quality.
Gas sensors measure the air around food. Oxygen, carbon dioxide, and other gases change as food respires, ripens, or spoils. In a sealed pack, a gas sensor can show whether the protective gas mixture is still correct. In a store, a carbon dioxide sensor can show that fruit is ripening fast, or that ventilation has failed. An ammonia sensor near a refrigeration plant warns of a leak from the cooling system, which is a safety matter for people as well as for food. Gas sensors are often cross-sensitive, meaning they react to more than one gas, so they must be chosen and placed with care, and their readings read as signs to check, not as proof by themselves.
Quick self-check
Question: A flour store has high humidity. What food problems could follow?
The flour can pick up moisture, clump, and grow mould, and its keeping quality falls. Insects also favour damp stores. The fix is ventilation or dehumidifying, guided by the humidity reading.
Pressure, flow, level, and vibration
Pressure sensors watch the force in a pipe or a vessel. Steam pressure in a cooker, air pressure in a packing machine, and the pressure drop across a filter all tell a story. A filter that slowly blocks shows a rising pressure drop across it, so the reading warns that cleaning is due before the flow falls and production stops. A steam line that loses pressure may mean a leak or a valve fault. Pressure is measured in units such as bar or pascal, and the sensor must be built for the fluid it touches: steam, milk, cleaning chemical, or air each need a sensor that resists them.
Flow sensors measure how much liquid or gas passes a point in a given time. A milk intake line, a water line to a mixer, and a cleaning solution line all need known flows. A flow meter can be a spinning wheel in the pipe, a sensor that times a sound pulse in the moving liquid, or a meter that reads the magnetic effect of a moving conductive liquid such as milk. The choice depends on the liquid, its thickness, and whether anything may be put inside the pipe. Flow readings support recipes: if the water flow to a dough mixer drifts, every batch changes.
Level sensors measure how full a tank or silo is. A float that rises with the liquid, a pressure sensor at the bottom that reads the weight of the liquid above it, or a sound pulse timed from the top to the surface all serve the same job. Level readings stop tanks from overflowing, stop pumps from running dry, and let a manager see stock without climbing a silo. In a milk tank, a level sensor also supports stock records: what came in, what was used, and what is left.
Vibration sensors measure the shaking of a motor, a pump, or a fan. A healthy machine vibrates a little in a steady pattern. A bearing that is wearing out vibrates more, and the pattern changes before the machine fails. A vibration sensor fixed to a motor casing reads that pattern day after day, and a rising trend warns that service is due. This is the basis of predictive maintenance: fix the machine when the data says it is wearing, not after it has broken during production, and not on a fixed calendar date that may be too early or too late.
The diagram shows the main sensor families and where each one is used. The steps point to the diagram: the job picks the sensor family, and the place picks the exact type within it.
Quick self-check
Question: A filter in a juice line is slowly blocking. Which sensor reading would warn you first?
The pressure drop across the filter. As the filter blocks, the pressure on the inlet side rises relative to the outlet side, so the difference grows before the flow falls.
Seeing, counting, and identifying: cameras, light, and tags
Not every sensor measures temperature or pressure. Some sensors see. A camera on a packing line can check the colour of biscuits, the fill level of a jar, the presence of a label, and the seal of a pack, many times a second. A simple light sensor can count packs passing a point, or detect whether a bottle is present before a filler starts. A colour sensor can sort fruit by ripeness, rejecting green fruit from a batch of ripe fruit. These optical sensors work fast, touch nothing, and are well suited to a line that moves quickly, but they need good light, a clean lens, and a clear view. Flour dust on a lens, steam in the air, or a flickering light can all fool them, so their housing and cleaning matter as much as their electronics.
Identification sensors answer a different question: which item is this? A barcode is a printed pattern read by light, cheap and widely used on packs and pallets. A radio tag, often called RFID, is a small chip that answers a radio signal with its identity number, without needing a clear line of sight, so a whole pallet of tagged crates can be read at once as it passes a reader at a doorway. A QR code is a square pattern that a phone or camera can read, carrying more information than a simple barcode, such as a batch number and a web link. These tools do not measure a condition, but they link a measurement to an identity: this temperature reading belongs to this batch, this pallet, this truck. Without that link, a temperature record is a number without an owner, and traceability fails.
Weight sensors, called load cells, measure the force of a load on a platform. They weigh ingredients into a mixer, check that each pack holds the right weight, and weigh a truck as it arrives and leaves. A checkweigher on a packing line rejects packs that are too light or too heavy, and the pattern of rejections tells the operator whether the filler is drifting. A load cell under a silo shows how much material is left without opening it. Weight is a direct, honest measurement: it is hard to fool, and easy to check with a known test weight.
Where to put the sensor
Placement decides whether a sensor tells the truth about the food or only about its own corner. Four rules guide placement. First, measure the food, not the equipment, wherever the decision is about the food. A probe in the milk, not on the tank wall, tells the milk temperature. A sensor inside a pallet, not only on the wall, tells the pallet temperature. Second, measure the worst place, not the best place. In a cold store, the warmest spot is usually near the door, at the top, or in the middle of a dense stack. A sensor at the worst place gives the earliest warning, while a sensor at the coldest place gives false comfort. Finding the worst place takes a survey: place several temporary sensors around the store for a period, find where the temperature is highest and most variable, and put the permanent sensors there.
Third, keep the sensor out of the way of the process it measures. A probe that blocks a flow, a sensor that catches product, or a cable that trips a worker is a bad installation, however good its readings. Sensors in a food plant must also be cleanable: smooth, sealed, and made of materials that stand up to washing and to contact with food. A sensor that cannot be cleaned becomes a source of contamination itself. Fourth, think about the air and the surroundings. A temperature sensor in direct sunlight, beside a hot motor, or in the draught from a door will read its surroundings, not the food. Shielding, shade, and distance from heat sources are part of placement, not afterthoughts.
The number of sensors is a judgement too. One sensor in a small chiller may be enough. A large warehouse needs several, and a long truck may need sensors at the front, the middle, and the back. The test is simple: if a fault could hide in a place no sensor watches, add a sensor or move one. If two sensors always read the same, one of them may be spare, and the money may be better spent elsewhere.
Quick self-check
Question: A cold store has sensors on the wall beside the cooler and nowhere else. What fault could hide?
A warm pallet in the middle, or warm air near the door. The wall beside the cooler is the coldest place, so it will read safe long after other parts of the store have warmed.
Question: Why survey a store with temporary sensors before fixing permanent ones?
Because the warmest and most variable spot is not always obvious. A survey finds where the temperature is highest, and the permanent sensor belongs there, where it gives the earliest warning.
Calibration: knowing the reading is true
A sensor reading is a claim about the world, and like any claim it needs checking. Calibration is the checking of a sensor against a known reference, and the correction of it if it is wrong. A temperature sensor is checked in iced water, which is a known point, and in boiling water or against a reference thermometer at a second point. A weighing scale is checked with a known test weight. A humidity sensor is checked in air held at a known humidity by a standard salt solution. The result of a calibration is a record: this sensor read this value against this reference on this date, and it was corrected by this amount, or it was passed as fit to use.
Calibration matters because sensors drift. Heat, age, dirt, knocks, and washing all move a reading slowly away from the truth. The drift is silent: the sensor keeps reporting numbers with the same confidence, and nothing on the dashboard says the numbers are wrong. Only a check against a reference reveals it. The interval between checks is set by how fast the sensor drifts, how exact the decision is, and what happens if the reading is wrong. A sensor that guards a safety limit is checked more often than one that watches a store room. A sensor that has drifted once before is checked more often than one that has always held true.
Between calibrations, two cheap checks catch many faults. The first is the ice point or zero check: a quick test at a known point that takes minutes. The second is comparison: two sensors in the same place should read nearly the same. If one starts to differ from its neighbour, one of them is drifting, and the pair must be checked. A plant that keeps a spare calibrated sensor can swap a suspect sensor out and check it on the bench, without leaving the place unmeasured.
Records of calibration are part of food safety proof. An auditor may ask: when was this sensor last checked, against what, and by whom? A system that stores calibration dates with the sensor record can answer in seconds. A system that relies on memory cannot. The later units show how IoT software can store these records beside the readings themselves, so the proof and the data live together.
Quick self-check
Question: Two sensors side by side read 4 degrees and 6 degrees. What do you do?
Trust neither one until they are checked. Take a reference thermometer, check both at a known point, and find which one has drifted. Do not average them and carry on, because the average of a true and a false reading is still false.
Power, housing, and life in a food plant
A sensor needs power, protection, and a way to send its reading. Power may come from a wire, which is steady but ties the sensor to one place, or from a battery, which frees the sensor to travel with a pallet or a truck but must be changed or charged. Some sensors harvest a little power from light or movement, but those suit only low-power jobs. The choice is a trade between freedom and maintenance: a battery sensor that runs for years with one reading an hour is easy to live with, while a sensor that reports every second will drain a battery fast and needs a wire.
Housing is the case that protects the sensor from its surroundings. In a food plant, housings are often sealed against water and dust, rated by a code that states how well they resist them. A sensor washed down with a hose needs a high rating. The housing material must also suit food contact: smooth stainless steel or food-grade plastic, with no cracks where microbes can hide. Cables need the same care: they must be sealed where they enter the housing, kept clear of moving parts, and made of material that stands up to cleaning chemicals.
The sensor node is the small unit that holds the sensor, its power, and its radio together. A plant may have dozens of nodes, each sending readings on a schedule. The node must be easy to reach for battery change and cleaning, and it must be fixed firmly enough that vibration and washing do not move it. A node that falls into a product is a foreign body in the food, so fixing and checking are safety jobs, not only technical ones.
Try it yourself
Choose one room in your home where food is kept, such as a kitchen or a pantry. Draw a simple plan of the room on paper. Mark three places where you would put a temperature sensor and one place where you would put a humidity sensor, and write one sentence for each mark saying why that spot and not another. Then list the sensor type you would use at each mark, and name one thing that could make its reading wrong in that spot: sunlight, a hot cooker, a draught, steam, dust. Finally, write a calibration plan in three lines: what reference you would check it against, how often, and what you would do if it read wrong. Compare your plan with a classmate and ask: whose sensor would catch a warm spot first?
Key ideas to carry forward
- A sensor turns a physical quantity into a signal a computer can read: range, accuracy, response time, and robustness decide whether it suits a food job.
- Temperature sensors include RTDs for accuracy, thermocouples for high heat, thermistors for small cheap jobs, and infrared sensors for surfaces only.
- Humidity, moisture, and gas sensors watch water and air around food, and each one must be read with its limits in mind.
- Pressure, flow, level, and vibration sensors watch pipes, tanks, and machines, and vibration trends give early warning of wear.
- Cameras, light sensors, barcodes, radio tags, and load cells see, count, identify, and weigh, and identification links a reading to its batch.
- Placement decides truth: measure the food, measure the worst place, keep the sensor cleanable, and keep it clear of sun, draughts, and heat.
- Calibration checks a sensor against a known reference, and drift is silent, so regular checks and comparison between sensors are the guard.
- Power, housing, and fixing are food safety matters: a sensor must survive washing, stay clean, and never fall into the product.
Glossary
Accuracy
How close a sensor reading is to the true value. Food tech example: a thermometer accurate to plus or minus 0.5 degrees may read 4.0 degrees when the true temperature is 4.5 degrees.
Bar code
A printed pattern of lines read by light, carrying an identity number. Food tech example: a bar code on a carton of milk links the carton to its batch and date.
Calibration
Checking a sensor against a known reference and correcting it so its readings are true. Food tech example: a cold store thermometer is checked in iced water and corrected if it reads wrong.
Drift
The slow movement of a sensor reading away from the true value as the sensor ages or gets dirty. Food tech example: a humidity sensor fouled with flour dust slowly reads lower than the true humidity.
Flow sensor
A sensor that measures how much liquid or gas passes a point in a given time. Food tech example: a flow sensor on a water line to a dough mixer checks that each batch gets the right amount of water.
Humidity sensor
A sensor that measures the amount of water vapour in air. Food tech example: a humidity sensor in a flour store warns when damp air could make the flour clump.
Infrared sensor
A sensor that measures temperature without touching, by reading heat radiation from a surface. Food tech example: an infrared sensor checks the surface temperature of biscuits on a belt.
Level sensor
A sensor that measures how full a tank or silo is. Food tech example: a level sensor in a milk tank shows how much milk is left without opening the tank.
Load cell
A sensor that measures weight by reading the force of a load on a platform. Food tech example: a load cell under a mixer weighs ingredients as they are added.
Pressure sensor
A sensor that measures the force in a pipe or vessel. Food tech example: a pressure sensor across a juice filter shows when the filter is blocking.
Radio tag (RFID)
A small chip that answers a radio signal with its identity number, without needing a clear line of sight. Food tech example: a radio tag on a crate of tomatoes lets a reader at a doorway record the whole pallet at once.
Range
The span of values a sensor can measure. Food tech example: a freezer sensor with a range of -30 to 10 degrees cannot measure an oven at 200 degrees.
Response time
How fast a sensor notices a change in what it measures. Food tech example: a thin probe in a pasteuriser reacts fast enough to catch a short temperature dip.
RTD (resistance temperature detector)
A temperature sensor that uses a metal whose electrical resistance changes in a steady way with temperature. Food tech example: an RTD in a pasteuriser gives the accurate reading that a safety decision needs.
Thermocouple
A temperature sensor made of two different metals joined together, producing a tiny voltage that changes with temperature. Food tech example: a thermocouple in an oven measures high baking temperatures.
Thermistor
A small ceramic temperature sensor whose resistance changes sharply with temperature. Food tech example: a thermistor inside a delivery box of chilled food reports the box temperature.
Vibration sensor
A sensor that measures the shaking of a machine, used to spot wear before failure. Food tech example: a vibration sensor on a pump warns that a bearing is wearing out.
Sensor node
A small unit holding a sensor, its power supply, and its radio together. Food tech example: a sensor node on a grain silo sends temperature and moisture readings to the farm office.
References
- Singh, R. P., Computer Applications in Food Technology, Academic Press, 1996.
- Irudayaraj, J., Food Processing Operations Modeling, Design, and Analysis, 2nd edition, 2002.
- Nielsen, S. S., Introduction to the Chemical Analysis of Foods, 1997.
Unit 3: Cold Chain Monitoring with IoT
Note
Unit 3: Cold Chain Monitoring with IoT
A question to start with
A truck leaves a dairy at dawn carrying yoghurt. The journey takes six hours, with two stops to load more pallets. At the far end, the receiver checks the yoghurt temperature and finds it too warm. The dairy says the yoghurt left cold. The driver says the truck stayed cold. The receiver says it arrived warm. Somebody is wrong, but nobody can prove who, because nobody watched the temperature during the journey. Now give the truck a sensor that records the temperature every five minutes and sends it to a dashboard. The record shows the truth: the temperature stayed low for four hours, rose during the second stop when the door stood open for an hour, and never came back down. The fault is found in one look, and the fix is plain: shorten the stop, keep the door shut, or pre-cool the load before the stop.
This unit is about the cold chain: the unbroken series of cold steps that keeps chilled and frozen food safe from production to the shop shelf, and how IoT watches every link in it.
Questions to think about
- A cold chain is a chain: what happens to the whole chain when one link warms up for an hour?
- A truck, a warehouse, and a shop chiller all hold the same yoghurt. Which one is hardest to watch, and why?
- If a temperature record shows a short rise that then falls again, is the food safe? What else would you need to know before deciding?
Think about the truck journey before you read on
A short rise is not judged by the peak alone. What matters is how high the temperature went, how long it stayed high, and what food it was, because microbes grow faster at higher temperatures and need time to grow. A record that shows both the height and the length of the rise lets a food safety person judge the risk, while a single reading on arrival cannot.
What the cold chain is
The cold chain is the set of steps, equipment, and practices that keep chilled and frozen food at the right temperature from the place it is made to the place it is eaten. It includes cooling after production, cold storage, refrigerated transport, cold rooms at distribution centres, and display chillers and freezers in shops. At each step the food must stay within its safe range: chilled food stays cold but above freezing, frozen food stays frozen. The chain is only as strong as its weakest link, because a single warm step can undo the care taken in every cold step before it.
Two facts make the cold chain hard to manage. First, temperature is invisible: food that has warmed and cooled again can look, smell, and feel the same as food that stayed cold, while microbes inside it have grown. Second, the chain has many owners: the maker, the transporter, the warehouse, and the shop each control one part, and each one sees only their own part. IoT answers both problems by making temperature visible and by keeping one record that follows the food across owners.
The temperature limits themselves are set by food safety practice and by the nature of the food. Chilled food is kept cold enough to slow microbial growth, and frozen food is kept cold enough to stop it. Different foods have different limits: fresh meat, milk, fish, and ready meals each have their own safe range, and a cold chain system must know the limit for the food it carries, not only a single number for everything. A system that alerts at the same temperature for ice cream and for fresh milk would be wrong for one of them. The limit, the food, and the batch must travel together in the record.
Quick self-check
Question: Why is a cold chain called a chain?
Because it is a series of linked steps, and a failure in any one step breaks the protection for the whole journey. A warm truck undoes a cold warehouse, just as a weak link breaks a chain.
Question: Why can warmed food look safe when it is not?
Because microbial growth is invisible at first. Food that has warmed and cooled again can look and smell normal while microbes have grown to unsafe numbers during the warm time.
Temperature, time, and microbial growth
Microbes grow faster when food is warm, and slower when food is cold. That simple fact is the whole basis of chilled storage. Keeping food cold does not kill most microbes, it slows them, so time matters as much as temperature. A food held a little too warm for a short time may still be safe, while the same food held at the same temperature for many hours may not be. This is why a temperature history is worth more than a single reading: the history shows both the height of the rise and its length, and both are needed to judge the risk.
In practice, quality and safety staff use two ideas. The first is the limit: the highest temperature the food may reach. The second is the time above the limit: how long it stayed there. A system that records both can answer the question that matters: was the food above its limit, and for how long? A good dashboard shows both at once, and a good alert says both too: "Cold store 2, 7 degrees, above the 5 degree limit for 40 minutes." That message lets the receiver decide at once, instead of waiting for a laboratory test that takes days.
Freezing adds a second concern: temperature cycling. Frozen food that warms a little and refreezes may form large ice crystals that damage its texture, even if it stays safe. Ice cream that has melted and refrozen is grainy, and frozen vegetables that have thawed and refrozen lose texture and drip on cooking. A temperature record that shows repeated small rises and falls is a sign of poor control, even when the food never crosses the safety limit. Quality and safety are separate questions, and the same record answers both.
Quick self-check
Question: Two loads both reach 8 degrees. One stayed there for 10 minutes, the other for 6 hours. Are they the same risk?
No. The load that stayed warm for 6 hours gives microbes far more time to grow, so its risk is much higher. The peak temperature alone does not judge safety: time above the limit matters just as much.
Monitoring a cold store
A cold store is a room built to stay cold, with insulated walls, a cooling unit, a door, and racks or pallets of food. Monitoring it sounds simple, but the details decide whether the readings mean anything. The first detail is sensor number and placement, covered in Unit 2 and worth repeating here: a store needs sensors at the warmest places, near the door, at the top where warm air gathers, and in the middle of a dense stack, not only beside the cooler. A store that holds different foods in different zones may need a sensor per zone, because the zones may have different limits.
The second detail is the reading rate. A cold store changes slowly, so a reading every few minutes is enough to catch a failure early and to draw a useful trend. A reading every second would add noise and cost without adding knowledge. The rate is chosen so that a real change is caught within minutes, and so that a door opening for loading shows as a short, clear rise that then falls again.
The third detail is the door. Doors are the commonest cause of warming in a cold store: left open during loading, propped open for convenience, or sealed badly so warm air leaks in all day. A door sensor that records each opening, and how long the door stayed open, turns a vague complaint into a fact. A store that warms every afternoon may be warming because deliveries arrive at midday, and the fix may be a change in delivery time or a strip curtain on the door, not a bigger cooler. The data finds the cause, and the cause decides the fix.
The fourth detail is the cooler itself. A cooling unit that runs without stopping, or that cycles on and off too often, is a sign of a fault: low refrigerant, a blocked coil, or a failing fan. A sensor on the cooler, reading its running state or its power use, watches the machine as well as the room. When the room temperature and the cooler state are shown together, the picture is complete: a rising room temperature with a cooler that has stopped is a cooler fault, while a rising temperature with a cooler running hard is a load or door problem. The pair of readings points to the cause faster than either one alone.
The diagram shows the four measurements that together describe a cold store. The steps point to the diagram: air temperature alone can mislead, and the door, cooler, and product readings complete the picture.
Quick self-check
Question: A cold store warms every afternoon but is cold at night. What is the most likely cause?
Something that happens in the afternoon: deliveries, door openings, sunlight on a wall, or a cooler struggling in the heat of the day. The time pattern points to the cause, so check what changes in the afternoon before buying a bigger cooler.
Monitoring a refrigerated truck
A truck is a cold store that moves, and movement adds problems a fixed store does not have. The truck passes through areas with no mobile signal, its power supply changes between the engine, the battery, and the mains at a depot, and its door opens at every stop. The sensor system must keep working through all of that. The usual answer is a sensor that stores readings on the truck when no signal is available and sends them on when the signal returns. That way the record is complete even when the live view is not, and a gap in the live view is understood as a signal gap, not a temperature gap.
Sensor placement in a truck matters as much as in a store. The front near the cooling unit is coldest, the back near the door is warmest, and the middle of a tightly packed load warms slowest to cool down and fastest to warm up when the door opens. A truck with sensors at the front, middle, and back can show the spread across the load, and a receiver can see whether the whole load is at risk or only the pallets near the door. Some trucks also carry a sensor in the load itself, inside a pallet or a box, because the air temperature in a truck falls fast when the cooler runs, while the food inside a pallet takes much longer to cool. Air temperature can look safe while the food inside is still warm, and the product sensor catches that gap.
Pre-cooling is a practice the data makes visible. A truck that is loaded with warm food and then cooled on the road will show a slow fall in temperature over hours, and the food spends those hours warmer than it should. A truck that is pre-cooled before loading, and loaded with food that is already cold, shows a flat, low line from the start. The record proves which practice was followed, and a receiver who sees a slow cooling curve can ask why the food was loaded warm.
The driver is part of the system too. A small display in the cab that shows the load temperature lets the driver notice a rise and act: shut a door, check the cooler, call the depot. A system that only shows the temperature in a faraway office wastes the person who is closest to the fault. Good truck systems show the reading in the cab and in the office, so the nearest hands and the widest view work together.
Quick self-check
Question: A truck dashboard goes blank for an hour in a rural area, then comes back showing a full record. What happened?
The signal was lost, not the measurement. The sensor stored the readings on the truck and sent them when the signal returned. The record is complete, and the blank hour was a connection gap, not a temperature gap.
Question: Why put a sensor inside a pallet as well as in the truck air?
Because air cools fast and food cools slowly. The air can read cold while the food inside a dense pallet is still warm. The product sensor shows the food temperature, which is the one that decides safety.
Handover, traceability, and proof
Food changes hands many times, and each handover is a moment when the cold chain can break and when blame can be passed. The IoT record turns handover from an argument into a check. When a truck arrives, the receiver looks at the journey record before unloading: the highest temperature, the time above the limit, and any gaps. If the record is clean, the load is accepted with confidence. If the record shows a long warm period, the load is held for a decision by a quality person, not accepted on trust. The decision is written down with the record, so it can be checked later.
Traceability ties the temperature record to the batch. A batch number on a pallet, a tag on a crate, or a code on a delivery note links the readings to a specific load of a specific food made on a specific day. When a problem is found later, perhaps a customer complaint or a laboratory result, the batch number lets the maker find every place that batch went, and every temperature record along the way. Without the link, the record is a set of numbers without an owner, and a recall cannot be aimed at the right loads.
Proof is also needed for audits and for customers who ask how their food was kept. A stored, time-stamped history can be shown, searched, and copied without loss. A paper chart can be lost, smudged, or filled in from memory, and it cannot be searched. The digital record is not proof by itself: it is proof only when the sensors were calibrated, the placement was sound, and the system was working. That is why calibration records and sensor checks, covered in Unit 2, are kept beside the temperature records. The two together are the proof.
Quick self-check
Question: A receiver accepts a load without looking at the journey record. What is lost?
The chance to catch a warm load before it is unloaded and mixed with good stock. Once the load is mixed in, the warm food cannot be separated, and the whole store is at risk.
Alerts that work, and alerts that fail
An alert is only useful if it reaches the right person in time to act. Three choices decide that: the limit, the delay, and the receiver. The limit is the temperature that triggers the alert. It is usually set a little below the true safety limit, so that the warning comes while there is still time to act. A limit set at the safety line itself warns only when the food is already at risk. A limit set far below the safety line warns too often, and people stop listening.
The delay is the time the temperature must stay above the limit before the alert is sent. A short delay catches faults fast but also catches harmless rises, such as a door opened for loading. A longer delay, ten or fifteen minutes, filters out those short rises and alerts only when the warming lasts. The delay is chosen by watching real data: how long does a normal loading rise last, and how fast does a real fault rise? The answer sets the delay between the two.
The receiver is the person who can act. At night that may be a duty manager, in the day the store supervisor, and for a truck the driver and the depot together. An alert that goes to a phone nobody checks at night is useless, and an alert that goes to five people at once may be acted on by nobody, because each one assumes another will act. A clear rota, one named person at a time, with a second person if the first does not answer, is part of the system design, not an afterthought.
Alert fatigue is the failure to guard against. When alerts are too many, too vague, or too often false, people learn to ignore them, and then a real alert is ignored too. The signs of fatigue are plain: alerts deleted without reading, alerts answered late, staff saying the system "always cries wolf." The cure is to review the alerts each month: how many were sent, how many were real faults, how many were false, and what changed after each one. Limits and delays are then tuned until most alerts are real and each one leads to an action.
Quick self-check
Question: A cold store sends an alert every time the door opens for loading. What is wrong, and how would you fix it?
The delay is too short, or there is no delay at all. A door opening warms the air for a few minutes, which is normal. Adding a delay of ten or fifteen minutes filters out those short rises, so the alert fires only when the warming lasts.
When the power or the network fails
Cold chain systems must keep working when things go wrong, because faults happen at night, in storms, and in places with poor signal. Three kinds of backup matter. The first is sensor memory: a sensor that stores readings when it cannot send them, and sends them later when the connection returns, so the record stays complete. The second is local warning: a light or a buzzer in the store that sounds even when the network is down, so a person nearby is warned without needing the internet. The third is battery backup: a sensor and a gateway that keep running on battery when the mains power fails, at least long enough to send an alert and keep recording.
The cooling itself needs a plan too. A cold store that loses power warms slowly if its doors stay shut and its insulation is good, and fast if doors are opened. The alert should tell staff to keep doors shut, move food to a working store if one is available, and call the engineer. A plan written down before the failure, with phone numbers and a named person, is worth more than a clever sensor with no plan behind it. The sensor finds the fault: the plan fixes it.
Testing the backup is part of the work. A backup that has never been tested is a guess. A monthly test, switching off the network or the power for a short time and checking that the local alarm sounds and the readings are stored and sent later, proves the system works when it is needed. The test result is recorded, like a calibration, so that an auditor can see the backup was checked, not only installed.
Energy, cost, and the shop chiller
Cold chains use a large amount of energy, and IoT data can cut that use without risking food. A cold store that is cooled below its needed temperature wastes energy and can damage food by freezing chilled products. A store that runs its cooler all day at full power when the load is light wastes energy too. Temperature and cooler data show where the waste is: a cooler that cycles too often, a store that is overcooled at night, a door that leaks warm air all day. Fixing those faults saves energy and keeps food safer, because a cooler that is not strained is less likely to fail.
The shop chiller is the last link, and often the weakest. A display chiller in a shop is opened by customers all day, loaded by staff, and sometimes blocked by goods piled too high, which stops cold air from flowing. A sensor in a display chiller watches the temperature through the day, and the record shows the pattern: warm in the afternoon when the shop is busy, cold at night when it is quiet. Staff who see that pattern can change loading practice, keep air paths clear, and check the chiller before the warmest part of the day. The shop is also where the customer sees the food, so a chiller that looks full and cold builds trust, while a chiller that is warm and messy loses it.
Cost is a fair question. Sensors, gateways, software, and maintenance all cost money, and a small business must weigh that cost against the loss it prevents. The honest way to weigh it is to count the losses without the system: loads thrown away, customers lost, and time spent searching for faults. One saved load can pay for a year of sensors, and one prevented recall can pay for many years. The cost must also be counted fairly on the other side: batteries, calibration, and staff time to answer alerts are real costs, and a system that nobody maintains will fail quietly and give false comfort. The decision is a business one, made with numbers from the plant itself, not from a sales claim.
Quick self-check
Question: A display chiller is warmest in the afternoon. What should staff check first?
Loading practice and air flow: goods piled too high blocking cold air, doors opened often by customers, and warm goods loaded in the afternoon. The time pattern points to use, not to a broken cooler.
Try it yourself
Imagine a truck carrying chilled milk on a six-hour journey with two stops. Draw a timeline of the six hours on paper. Sketch the temperature line for three cases: a good journey that stays cold all the way, a journey where the door stands open for an hour at the second stop, and a journey where the cooler fails halfway and is fixed an hour later. For each sketch, write down the highest temperature, how long the milk stayed above its limit, and what you would do on arrival: accept, hold for a decision, or reject. Then write an alert rule for the truck in one sentence: the limit, the delay, and who receives the alert. Compare your sketches with a classmate: did you draw the same shape for the door-open case, and does your rule catch it in time?
Key ideas to carry forward
- The cold chain is an unbroken series of cold steps from production to the shop, and it is only as strong as its weakest link.
- Temperature and time together decide risk: how high the temperature went and how long it stayed there.
- A cold store needs sensors at the warmest places, a door record, a cooler state reading, and a product temperature, not only air near the cooler.
- A truck needs sensors at front, middle, and back, storage when the signal is lost, and a display in the cab as well as in the office.
- Handover is a check, not an argument: the journey record is read before unloading, and the batch number links the record to the food.
- Alerts need a limit set below the safety line, a delay that filters out short harmless rises, and one named receiver at a time.
- Backups must be tested: sensor memory, local alarms, and battery power keep the system working when the network or the mains fail.
- The same data that protects food can cut energy use, by finding overcooling, door leaks, and coolers working too hard.
Glossary
Alert fatigue
The habit of ignoring alerts because they are too many, too vague, or too often false. Food tech example: staff stop reading alerts after a cold store sends one every time its door opens.
Batch number
A code that identifies one batch of food made at one time, used to link records to that food. Food tech example: a batch number on a pallet of yoghurt links its temperature record to that day's production.
Chilled food
Food kept cold but above freezing, so microbial growth is slowed but not stopped. Food tech example: fresh milk and yoghurt are chilled foods.
Cold chain
The unbroken series of cold steps that keeps chilled and frozen food at the right temperature from production to eating. Food tech example: the cold chain for fish runs from the boat through trucks and cold stores to the shop chiller.
Cooling unit
The machine that removes heat from a cold store, truck, or chiller. Food tech example: the cooling unit on a truck keeps the load cold while the truck is on the road.
Door sensor
A sensor that records when a door opens and how long it stays open. Food tech example: a door sensor in a cold store shows that the door stood open for an hour during loading.
Frozen food
Food kept below freezing, so microbial growth is stopped and the food stays hard. Food tech example: ice cream and frozen vegetables are frozen foods.
Handover
The moment food passes from one owner to the next, such as from a truck to a warehouse. Food tech example: at handover the receiver checks the journey temperature record before unloading.
Limit
The highest temperature a food may reach before action is needed. Food tech example: the limit for a chilled store is the temperature above which an alert is sent.
Pre-cooling
Cooling a truck or a store before loading it with cold food, so the food does not warm on loading. Food tech example: a truck is pre-cooled before a load of chilled milk is put in.
Predictive maintenance
Servicing a machine when its data shows wear, before it fails. Food tech example: a cooler that runs without stopping is serviced before it breaks down.
Product temperature
The temperature inside the food itself, rather than the air around it. Food tech example: a probe inside a pallet of meat shows the product temperature, which decides safety.
Recall
The removal of a food from sale when it is found to be unsafe, aimed at the batches concerned. Food tech example: a batch number lets a maker recall only the yoghurt that travelled in a warm truck.
Temperature cycling
Repeated warming and cooling of frozen food, which damages texture even when the food stays safe. Food tech example: ice cream that has melted and refrozen is grainy because of temperature cycling.
Time above the limit
How long food stayed above its safe temperature, used with the peak temperature to judge risk. Food tech example: milk held above its limit for six hours is a higher risk than milk held above it for ten minutes.
References
- Singh, R. P., Computer Applications in Food Technology, Academic Press, 1996.
- Irudayaraj, J., Food Processing Operations Modeling, Design, and Analysis, 2nd edition, 2002.
- Nielsen, S. S., Introduction to the Chemical Analysis of Foods, 1997.
Unit 4: IoT in Food Processing and Packaging Lines
Note
Unit 4: IoT in Food Processing and Packaging Lines
A question to start with
A biscuit line runs all morning. The biscuits look right, the packs are sealed, and the line stops only twice, each time for a few minutes. At the end of the shift the quality team finds that one hour of production, from mid-morning, has weak seals, and the packs must be opened and repacked. Nobody noticed at the time, because the sealing temperature had drifted down slowly, a fraction of a degree at a time, while the line kept running. Now picture the same line with sensors on the sealing jaws, a counter that records each stop, and a dashboard that shows the seal temperature as a line across the morning. The drift shows as a slow fall on the chart long before the seals fail, and the operator sees it, adjusts the heater, and saves the hour of production.
This unit is about IoT on the processing and packaging line: how sensors watch each step of making and packing food, how they catch slow drift and sudden faults, and how the record of a line helps a plant make the same product the same way every day.
Questions to think about
- A line makes 200 packs a minute. How could a person check every pack, and what could a sensor check instead?
- A machine stops for two minutes, ten times a day. Is that a small problem or a large one, and how would you prove it?
- If a recipe says to add water at a set rate, what goes wrong when the flow drifts a little high all morning?
Think about the weak seals before you read on
A slow drift is hard to see by eye, because each pack looks nearly the same as the one before it. A chart of the sealing temperature across the morning makes the drift visible as a line falling step by step. The sensor does not need to be clever: it only needs to show the trend, and a person who looks at the trend can act before the limit is crossed.
A line as a series of steps
A food processing line is a series of steps in order: raw material comes in, is mixed, heated or cooled, formed, filled, sealed, labelled, and packed into cases. Each step has a condition that must be held: a mixing time, a heating temperature, a fill weight, a seal temperature. IoT watches each step by placing sensors at the points where the condition is set and where it can be checked. The result is a picture of the whole line, step by step, instead of a single check at the end.
It helps to think of the line as a river with gates. At each gate a measurement is taken, and the record shows whether the product passed the gate in the right condition. If a fault is found at the end, the record shows which gate let it through. Without that record, the fault is found only at the end, and the whole morning of production is in doubt. With it, only the packs made after the last good gate are in doubt, and the rest can be released with confidence.
The steps differ by product. A dairy line pasteurises, homogenises, cools, and fills. A bakery line mixes, proves, bakes, cools, and packs. A canning line fills, seals, heats, and cools. The sensors differ too, but the pattern is the same: measure the condition at each step, record it with the time, and link it to the batch. The later sections walk through the main steps and the sensors that watch them.
Quick self-check
Question: A fault is found in finished packs. Why does a step-by-step record help?
Because it shows which step let the fault through and when. Only the packs made after that step failed are in doubt, instead of the whole day of production.
Mixing, heating, and cooking
Mixing is the first step where IoT earns its place. A mixer that runs too short leaves a dough uneven, with pockets of dry flour. A mixer that runs too long warms the dough and changes its texture. Sensors on a mixer watch the motor load, which rises as the dough thickens, the mixing time, and the dough temperature. The motor load curve tells an experienced operator when the dough is ready, and the record proves the same mixing was given to each batch. A recipe that says "mix for eight minutes" is a clock rule. A record that shows time, temperature, and motor load is proof of what actually happened.
Heating and cooking are watched by temperature sensors at several points: the heating medium, the product inlet, the product outlet, and, where possible, the product itself. In a pasteuriser, the holding temperature and the holding time are the safety conditions, and the sensors that prove them are checked and calibrated with special care. In a continuous cooker, the temperature along the length of the cooker shows whether the product is heated evenly, and a cold spot in one zone can be found before it leaves food undercooked. In an oven, sensors watch the temperature in different zones, the humidity of the air, and the speed of the belt, because a loaf that spends too little time in a cool oven comes out pale and underbaked, while a loaf that spends too long in a hot oven comes out dark and dry.
Cooling after heating matters as much as heating. Cooked food that cools slowly spends a long time in the temperature range where microbes grow fast. Sensors that record the cooling curve, the fall in temperature over time, show whether the food passed through that range fast enough. A cooling tunnel that is blocked, or a chiller that is overloaded, shows as a cooling curve that flattens too high, and the operator can act before the food is packed.
Steam, water, and air are the helpers of heating and cooling, and their sensors matter too. Steam pressure, water flow, and air flow each have sensors, because a change in any one of them changes the heating or cooling. A steam line that loses pressure heats less, a water line that runs low cools less, and a fan that slows moves less air. Watching the helpers is often the fastest way to find the cause of a product fault.
Quick self-check
Question: A pasteuriser outlet temperature is correct, but the holding time is short. Is the milk safe?
Not on the temperature alone. Pasteurisation needs both the temperature and the time at that temperature. A correct temperature held for too short a time may leave the milk undertreated, so both readings must be right.
Question: Why record the cooling curve, not only the final cooled temperature?
Because the time spent cooling through the warm range decides microbial growth. Food that reaches the right final temperature slowly has spent longer in the risky range than food that cools fast.
Filling, weighing, and checking the pack
Filling is a step where small errors become large losses. A filler that puts a little too much in each pack gives product away, pack after pack, all day. A filler that puts a little too little in each pack breaks the weight on the label and risks the law and the customer. A checkweigher, a fast weighing machine on the line, weighs each pack as it passes and records the weight. The record shows the average weight, the spread of weights, and the trend: a slow drift upward means product is being given away, a slow drift downward means the filler needs adjusting. The checkweigher can also reject packs outside the allowed range, so a wrong pack does not reach the case.
Level sensors and flow sensors support filling too. A flow meter on a liquid line measures how much has passed, and a level sensor in a supply tank warns when the tank is running low, so the line is not starved halfway through a batch. A sensor that counts packs, using light or a camera, records the line speed and the number made, and the count supports stock records and the calculation of waste: packs made minus packs packed is the loss, and the loss should be small and known.
Cameras and light sensors check the pack itself: is the label present, is it straight, is the print readable, is the seal closed, is the cap on? A camera can check hundreds of packs a minute, far faster than a person, and it never tires. The camera does not judge taste or smell, and it cannot check the inside of a sealed pack, but for the outside of the pack it is fast and steady. When a camera starts rejecting many packs, the record shows when the rejections started, and the operator can look at the machine that feeds that fault, such as a labeller running low on labels or a sealer cooling down.
The diagram shows the checks along a filling and packing step. The steps point to the diagram: each check catches a different fault, and the record ties them to the batch.
Quick self-check
Question: A checkweigher shows the average pack weight drifting slowly upward all morning. What is being lost?
Product. Each pack holds a little too much, and the extra amount, multiplied by thousands of packs, is product given away. The filler needs adjusting back to the target weight.
Sealing, and why it fails quietly
Sealing is one of the quietest failures on a packing line. A seal can look closed and still leak, letting air and microbes in and letting freshness out. The seal is made by heat and pressure applied for a set time, and all three must be right. A sealing jaw that is too cool, a jaw that is dirty with product, a jaw that applies uneven pressure, or a film that has changed will all make weak seals. Sensors on the sealer watch the jaw temperature, the pressure, and the time the jaws are closed. The readings show drift long before the seals fail, as in the biscuit story at the start of this unit.
Seal testing is also part of the line. A sample of packs is tested at intervals, by squeezing, by dipping in water and watching for bubbles, or by measuring the force needed to peel the seal open. The test results are recorded with the time, so a failed test can be tied to the seal temperature at that time. The combination of continuous sensor readings and regular sample tests gives a full picture: the sensor shows the trend, and the test proves the seal strength.
Modified atmosphere packing adds a gas measurement. Packs are filled with a mixture of gases, often with less oxygen and more carbon dioxide or nitrogen, to slow spoilage. A gas sensor checks the mixture in the pack or in the gas supply line, and a wrong mixture warns that the gas supply has changed or a valve has stuck. A pack with the wrong gas looks the same as a good pack, so only the sensor can catch it. The gas reading is recorded with the batch, so a later question about keeping quality can be answered from the record.
Quick self-check
Question: A seal looks closed but the pack leaks. Which sensor reading would you check first?
The sealing jaw temperature and pressure at the time the pack was made. A jaw that was too cool or applied too little pressure makes a seal that looks closed but does not hold.
Stops, speed, and the true cost of small delays
A line that stops for two minutes, ten times a day, loses twenty minutes of production, and the cost is larger than the clock shows, because each stop and restart wastes product, needs checking, and disturbs the steady state of the machines. IoT counts stops automatically: each stop is recorded with its start, its end, and its cause, where the cause is known. The record shows the pattern: which machine stops most, at what time, and for what reason. A machine that stops briefly many times is often worse than a machine that stops once for a long repair, because the short stops are accepted as normal and never fixed.
Line speed is also recorded: packs per minute, cases per hour, and the time the line spent running against the time it was meant to run. The ratio of running time to planned time is a plain measure of how well the line used its day, and it is a fair way to compare one day with another or one line with another. A line that runs at full speed for half the day and stands still for the other half makes less than a line that runs a little slower all day without stopping. The data makes that visible, and the fix is often not a faster machine but fewer stops.
Downtime causes fall into groups: waiting for material, a machine fault, a changeover to a new product, cleaning, and a quality hold. Recording the cause each time, even in a few plain words, builds a picture over weeks. A line that waits for material every morning has a supply problem, not a machine problem. A line that stops for cleaning longer each week has a cleaning problem. The record points the fix at the right place, and the fix can be tested by watching the record after it is made: did the stops fall?
Quick self-check
Question: A line stops for two minutes, ten times a day. How much running time is lost?
Twenty minutes a day, plus the product wasted at each stop and restart, and the disturbance to steady running. Ten small stops are a real loss, not a small one, and the record makes the total visible.
Cleaning, changeover, and the record of a batch
Cleaning is a step in production, not a break from it, and IoT watches it too. A cleaning cycle, often called clean-in-place, runs cleaning solution through pipes and vessels without taking them apart. Sensors watch the solution temperature, its flow, its concentration, and the time it runs, because a cleaning cycle that is too cool, too short, or too weak leaves product behind, and a cycle that runs too long wastes water, chemical, and time. The record proves the cleaning was done as specified, and it shows when a cycle was skipped or shortened.
Changeover is the switch from one product to another: the line is cleaned, the recipe is changed, the pack size is changed, and the labels are changed. Changeover is a time when mistakes are easy: the wrong label on the right pack, the wrong recipe in the mixer, the old product left in a pipe. Sensors and checks help: a recipe loaded from the system rather than typed in by hand, a label checked by camera against the product code, a first pack checked and recorded before the line runs at speed. The record of the changeover, what was cleaned, what was loaded, and who checked it, is part of the batch record.
The batch record is the sum of all these readings for one batch: the ingredients used, the mixing time, the heating temperature and time, the cooling curve, the fill weights, the seal temperatures, the cleaning before and after, and the counts of packs made and rejected. In older plants the batch record was paper, filled in by hand. In an IoT plant much of it is filled in automatically, with times attached by the sensors themselves. The automatic record is faster, harder to forget, and easier to search, but it still needs a person to check it: a record that nobody reads is a store of numbers, not a proof.
Quick self-check
Question: Why record a cleaning cycle in the batch record?
Because the next batch can be affected by the cleaning before it. A cleaning cycle that was too short or too cool can leave old product behind, and the record proves the cleaning was done as specified.
Maintenance: fixing machines before they break
Machines wear, and wear shows in data before it shows as a breakdown. A motor that is wearing vibrates more, runs hotter, and draws more power. A pump that is wearing delivers less flow for the same power. A fan that is wearing runs noisier and slower. Sensors that watch vibration, temperature, and power draw on a machine build a trend over weeks and months, and a rising trend is a warning that service is due. This is predictive maintenance: the machine is serviced when its data says it needs it, not after it has broken during production, and not on a fixed date that may be too early, wasting a good part, or too late, missing the wear.
The alternative, fixing only after a breakdown, is the most expensive way to run a plant. A breakdown during production stops the line, wastes the product in the machines, and needs an urgent repair at any hour. A planned service, done between batches, costs less and wastes nothing. The data does not replace the engineer: it tells the engineer where to look, and when. The engineer still opens the machine, sees the worn part, and judges it. The sensor gives the timing, the person gives the judgement.
Spare parts and service records belong with the machine data. A motor that has had three bearings in a year is a different machine from one that has had none, and the record shows it. A plant that keeps machine data, service dates, and parts used in one place can see which machines cost most to keep, and whether a machine should be repaired, rebuilt, or replaced. Those are business decisions, and the data makes them honest.
Quick self-check
Question: A motor vibrates a little more each month. When should it be serviced?
Before the vibration reaches the point where the motor fails, and before it damages the product or the line. The rising trend is the warning: plan the service between batches, while the motor still runs.
People on the line, and the dashboard they need
A line is run by people, and the dashboard is their tool. A good dashboard shows the present state of the line in one view: the speed, the temperature at each key point, the count made, the count rejected, and any alert that needs action. It uses plain numbers and simple charts, with the safe range shown so that a reading outside it stands out at once. It does not show every sensor at once, because a screen crowded with numbers hides the one that matters. The operator sees the line, the supervisor sees several lines, and the manager sees the plant: each view shows what that person needs, and no more.
Alarms on a line must be treated with the same care as cold store alerts. An alarm that sounds for every small change will be ignored, and an alarm that sounds only when the product is already wrong comes too late. The alarm limit is set below the fault limit, the delay filters out short harmless changes, and the alarm names the machine and the value. When an alarm sounds, the operator should know what to do: the action is written down beside the alarm, not left to memory. After the fault is fixed, the alarm is recorded with its cause, so the pattern of alarms can be reviewed and the limits tuned.
Training matters as much as the screen. An operator who understands what a sensor measures, and what its reading means for the product, will act on it. An operator who sees only numbers will wait for somebody else to decide. The best plants teach the line with the data: a new operator watches the charts with an experienced one, sees what a normal day looks like, and learns what a drift looks like before it becomes a fault. The sensor is a tool, and like any tool it is only as good as the hand that uses it.
Try it yourself
Choose a food you know well, such as bread, yoghurt, or jam. Draw the line that makes it as a series of five steps on paper, from raw material to packed product. For each step, write down one condition that must be held, one sensor that would watch it, and one fault that the sensor would catch. Then add a checkweigher or a camera check at the end, and write down what it would reject. Finally, mark one step where a slow drift would be the hardest to see by eye, and write one sentence saying how a chart of that sensor across the morning would show the drift. Compare your line with a classmate who chose a different food: which steps are the same, and which are different?
Key ideas to carry forward
- A line is a series of steps, and IoT watches the condition at each step, so a fault can be traced to the gate that let it through.
- Mixing, heating, and cooling are watched by time, temperature, motor load, and flow, and pasteurisation needs both temperature and time.
- Filling is watched by flow, level, and checkweighing: a slow drift in average weight gives product away or breaks the label weight.
- Sealing needs heat, pressure, and time all right, and a weak seal can look closed while it leaks.
- Short stops add up: counting stops and their causes shows whether the loss is a machine, a supply, a cleaning, or a changeover problem.
- Cleaning and changeover are production steps, and their records belong in the batch record with the rest.
- Vibration, temperature, and power trends warn of wear before a breakdown, so service can be planned between batches.
- A dashboard must show what the operator needs in one view, and alarms must name the machine, the value, and the action.
Glossary
Batch record
The set of records for one batch: ingredients, process readings, checks, cleaning, and counts. Food tech example: the batch record for a day of yoghurt shows its pasteurisation temperature, fill weights, and seal checks.
Changeover
The switch of a line from one product to another, including cleaning, recipe change, and label change. Food tech example: changing a line from strawberry to plain yoghurt needs a full clean and a label check.
Checkweigher
A fast weighing machine on a line that weighs each pack and rejects packs outside the allowed range. Food tech example: a checkweigher rejects biscuit packs that are too light.
Clean-in-place
Cleaning pipes and vessels by running cleaning solution through them without taking them apart. Food tech example: a milk line is cleaned in place between batches with hot cleaning solution.
Cooling curve
The fall in temperature of a food over time as it cools. Food tech example: the cooling curve of cooked soup shows whether it passed through the warm range fast enough.
Downtime
Time when a line is meant to run but is stopped. Food tech example: a line stopped for a broken filler is in downtime until the filler is fixed.
Fill weight
The weight of product put into each pack. Food tech example: the fill weight of a jam jar is checked so each jar holds the weight printed on its label.
Holding time
The time a food is held at a set temperature, such as in pasteurisation. Food tech example: milk is held at its pasteurising temperature for a set time before it is cooled.
Line speed
How fast a line runs, often in packs per minute. Food tech example: a biscuit line running at 200 packs a minute makes far more in an hour than one running at 100.
Modified atmosphere packing
Packing food in a mixture of gases that slows spoilage, often with less oxygen. Food tech example: salad leaves packed in a gas mixture stay fresh longer than leaves packed in plain air.
Motor load
How hard a motor is working, shown by the power it draws. Food tech example: the motor load on a dough mixer rises as the dough thickens.
Predictive maintenance
Servicing a machine when its data shows wear, before it fails. Food tech example: a pump is serviced when its vibration trend rises, before it breaks down.
Seal temperature
The temperature of the sealing jaws that close a pack. Food tech example: a sealing jaw that cools down makes weak seals that leak.
Trend
The direction of readings over time, such as a slow rise or fall. Food tech example: a trend of rising seal failures across a morning points to a cooling sealing jaw.
Zone
One section of a long machine, such as an oven, with its own temperature control. Food tech example: the first zone of a bakery oven is cooler than the last zone, so bread bakes evenly as it moves through.
References
- Singh, R. P., Computer Applications in Food Technology, Academic Press, 1996.
- Irudayaraj, J., Food Processing Operations Modeling, Design, and Analysis, 2nd edition, 2002.
- Nielsen, S. S., Introduction to the Chemical Analysis of Foods, 1997.
Unit 5: IoT in Farming and Raw Material Supply
Note
Unit 5: IoT in Farming and Raw Material Supply
A question to start with
A farmer grows tomatoes for a sauce factory. In a dry week, she waters the whole field every day, because she cannot tell which parts are dry and which are still damp. The water bill is high, and some parts of the field are waterlogged while others are thirsty. The next season she puts soil moisture sensors in several parts of the field. The sensors show that the low ground stays damp for days after watering, while the high ground dries in a day. She waters only the high ground, and only when its sensor says it is dry. She uses less water, the plants grow more evenly, and the tomatoes arrive at the factory more uniform in size and ripeness. Nothing about the tomatoes changed. What changed is that the farmer could see the water in the soil, and water only where it was needed.
This unit is about IoT before the factory gate: on the farm, in the field, in the animal shed, and on the road that brings raw material to the factory. The quality of a food is fixed long before it reaches the line, and sensors that watch the growing and the journey help the factory receive material that is right for the process.
Questions to think about
- A field is not uniform: some parts are wet, some dry, some shady, some sunny. How many sensors would you need to know a whole field, and where would you put them?
- Milk is collected from many farms. How could a factory know, before the truck arrives, whether the milk from one farm is warm or sour?
- A tomato is picked green and ripens on the road. What would you measure on that journey to know when it will be ready?
Think about the tomato field before you read on
One sensor in one place can only speak for that place. A field needs sensors in its different parts: high ground and low ground, sunny and shady, near the water source and far from it. The pattern across those sensors is the picture of the field, and watering follows the pattern, not a fixed calendar.
The farm as a place to measure
A farm is a large, open place, and that changes how IoT works there. Inside a factory, sensors are close together, power is on the wall, and the network is in the building. On a farm, sensors may be spread over a wide area, power may be a battery or a small solar panel, and the network may be a long-range radio link to a gateway at the farmhouse. The sensors must survive sun, rain, dust, mud, and animals, and they must keep working for a season without daily attention. Those conditions shape every choice in farm IoT.
The quantities worth measuring on a farm follow the needs of the crop or the animal. For a crop, the main ones are soil moisture, soil temperature, air temperature, air humidity, light, and rainfall. For an animal, the main ones are body temperature, movement, feeding, and the temperature and humidity of the shed. For stored grain, the main ones are the temperature and moisture inside the store. Each measurement answers a plain question: does this plant need water, is this animal well, is this grain safe in store? The sensor is only useful when the question is clear, and the action that follows the answer is known.
A weather station on a farm is a small group of sensors together: air temperature, humidity, wind, rainfall, and light. Its readings guide planting, spraying, irrigation, and harvest. A frost warning from a weather station lets a grower protect a tender crop for one night, and a rain record shows whether a field needs water or has had enough. Weather data from the farm itself is more useful than a forecast for a distant town, because rain can fall on one field and miss the next. The farm station measures the farm, not the region.
Quick self-check
Question: Why is a farm weather station more useful than a forecast for a distant town?
Because weather varies over short distances. Rain can fall on one field and miss the next, and frost can touch a low field while a high field stays clear. The farm station measures the farm itself.
Soil, water, and irrigation
Soil moisture is the amount of water held in the soil, and it decides when a crop needs watering. A soil moisture sensor is buried in the root zone, where the plant takes up water, and it reads how wet the soil is there. A single sensor at one depth tells only part of the story, because the top soil dries first while deeper soil stays damp. Sensors at two depths, one shallow and one deep, show whether water has reached the roots or only wet the surface. A reading that falls slowly over days shows the crop using water steadily, while a reading that falls fast shows a thirsty crop or a hot wind.
Irrigation guided by sensors is called precision irrigation: water is given only where the soil is dry, and only as much as the soil needs. The water may be delivered by drip lines that put water at the roots, or by sprinklers that cover a field. A sensor that opens a valve when the soil is dry, and shuts it when the soil is wet enough, does the job without a person walking the field each day. The saving is water, and the gain is a crop that is neither thirsty nor waterlogged. Waterlogging matters as much as drought: roots without air rot, and a wet crop is open to disease. A sensor that stops watering in time prevents both.
Soil temperature matters too. Seed sown into cold soil germinates slowly or not at all, and a soil temperature sensor tells the grower when the soil is warm enough to sow. Soil temperature also guides the timing of harvest for root crops, and the storage of seed. A sensor that reads the soil at sowing depth is a small tool with a large effect: sowing a week too early can lose a crop, and sowing a week late can lose the season.
Water quality is a further measurement where irrigation water is reused or drawn from a store. Sensors that read the salt content or the acidity of irrigation water warn when the water would harm the crop or the soil. Salt builds up in soil over years, and a reading that shows rising salt lets the grower change the water source or flush the soil before the damage is done. The measurement is simple, but the damage it prevents is slow and hard to undo.
Quick self-check
Question: A soil sensor at the surface reads dry, but a sensor deeper down reads wet. Should you water?
Probably not yet, or only a little. The surface dries first, while the roots may still have water deeper down. The deep reading decides, because the roots drink from the deep soil.
Question: Why can too much water harm a crop as much as too little?
Because waterlogged soil has no air for the roots, and roots without air rot. Wet leaves and wet soil also favour disease. Water must be enough for the plant, and no more.
Animals, sheds, and the milking parlour
Animals tell their condition through movement, temperature, and feeding, and sensors can read those signs day and night. A movement sensor on a cow, often worn on a leg or a collar, records how much the animal walks, lies down, and feeds. A cow that moves less, lies longer, and eats less may be unwell, and the sensor notices the change a day before a person would see it by eye. Early notice means early treatment, and early treatment means a smaller loss and a healthier animal.
Body temperature is a direct sign of health. A temperature sensor can read an animal's temperature without handling it, and a rise can warn of infection. In a milking herd, a sensor in the milking system can read the milk temperature and the milk yield of each animal, and a fall in yield from one animal, or a rise in the milk temperature, can point to a problem with that animal alone. The data is per animal, not per herd, so the animal that needs attention is named, and the rest of the herd is left alone.
The shed itself is measured too. Air temperature, humidity, and air flow in a shed decide whether animals are comfortable or stressed. Heat stress in hot weather cuts milk yield and growth, and a sensor that warns when a shed is too hot lets the farmer start fans, open vents, or spray water to cool the air. Ammonia in the air, from animal waste, is a further measure: high ammonia harms animal health and the people who work in the shed, and a sensor that shows rising ammonia points to a ventilation or cleaning problem.
Feed and water are the last part of the animal picture. Sensors in feed stores and water troughs show how much is left and how fast it is used. A water trough that empties too fast may have a leak, and a trough that is not drunk from may have stale water or a sick group of animals. A feed bin that empties faster than the animals should eat may mean waste or theft, and a bin that empties slower may mean the feed is unpalatable or the animals are unwell. The readings are simple, but they watch the daily routine without a person standing there.
The diagram shows the main farm measurements in order. The steps point to the diagram: soil and weather guide the crop, animal sensors guide care, and shed and store sensors guard the harvest.
Quick self-check
Question: A cow moves less, lies longer, and eats less than usual. What should the farmer do?
Check the animal at once. The change in movement and feeding is an early sign of illness, and early treatment gives the best chance of a quick recovery.
Grain storage on the farm
Grain in store is a living material: it breathes, gives off heat and moisture, and can be damaged by insects and mould. A grain store that is not watched can lose a harvest quietly. Temperature sensors in a silo or a store watch for hot spots, places where the grain is heating. A hot spot is a warning: the grain there is damp, or insects are active, or air is not moving through it. The sensor finds the hot spot while it is still small, and the farmer can move air through the grain, move the grain, or dry it before the damage spreads.
Moisture is the partner of temperature in grain storage. Grain that is too wet will heat and spoil, while grain that is dry enough will keep. A moisture sensor in the store, and a sensor at the intake when grain arrives, tells the farmer whether the grain is safe to store or needs drying first. Drying uses fuel and time, and the sensor lets the farmer dry just enough: grain dried too far loses weight and value, and grain dried too little spoils. The reading sets the drying, not a guess.
Insects are a further risk, and sensors can help indirectly. Insect activity raises the temperature and the carbon dioxide in the store, so a rise in either one can be a sign to check the grain. A sensor does not see the insect, but it sees the heat and gas the insects make, and that is enough to prompt a check. The check, opening the store and looking, is still a human job, and the sensor is the prompt that says when to look.
Quick self-check
Question: A temperature sensor in a grain silo shows one spot warming while the rest stays cool. What should happen next?
Check that spot at once. A hot spot means the grain there is heating, from dampness or insects, and moving air or moving the grain early can stop the damage from spreading.
From farm to factory: collection, transport, and intake
Raw material travels from the farm to the factory, and the journey decides its condition on arrival. Milk is the clearest case. Milk leaves the farm warm from the animal and must be cooled fast, held cold in the farm tank, and carried cold to the factory. Sensors in the farm tank watch the cooling: did the milk cool fast enough, and did it stay cold until collection? Sensors on the truck watch the journey, as in Unit 3. At the factory gate, the intake test checks the milk, but the sensor record from the farm and the truck already tells the story: the factory can see which farm cooled slowly, which truck ran warm, and which load is the one to hold.
Fruit and vegetables travel differently. They are picked, packed, and carried, and they keep ripening on the road. Temperature, humidity, and the gas ethylene, which fruit gives off as it ripens, all shape the journey. A sensor in a crate or a truck that reads temperature and ethylene can show whether the fruit is ripening too fast, and whether the truck needs more cooling or more ventilation. A load that arrives overripe is a loss, and a load that arrives underripe can be ripened under control at the factory. The sensor record lets the factory plan: which load to use first, which to hold, and which to reject.
Grain, oilseeds, and other dry materials travel in bulk, and their journey is watched by temperature and moisture sensors in the store and the truck, and by weight sensors that record the load. A load that gains moisture on the road, from rain through a torn cover or from damp storage, arrives heavier and worse, and the record shows when the moisture rose. A load that loses weight on the road may have leaked, and the weight record shows it. The intake at the factory checks the material, and the journey record supports the check.
Quick self-check
Question: Milk from one farm always arrives warm, while milk from other farms arrives cold. Where is the fault?
Most likely at that farm: its cooling is slow, its tank is faulty, or collection comes too late after milking. The farm tank record, not the truck record, will show it, because the other farms on the same truck arrive cold.
Intake at the factory: the first quality decision
Intake is the factory gate, and the sensor record from the farm and the journey meets the factory test there. The intake check measures what the factory can test on arrival: temperature, weight, and for milk, acidity and composition, for grain, moisture and purity, for fruit, ripeness and damage. The sensor history adds what the intake test cannot: how the material was kept before it arrived. A load that tests well on arrival but spent hours warm on the road is a different risk from a load that stayed cold all the way, and the record lets the quality person tell them apart.
Acceptance rules are written down before the load arrives: the limits for temperature, moisture, and other measures, and the action for each result: accept, hold, use first, or reject. The sensor record feeds the rule, so the decision is made the same way each time, and the decision is recorded with the reason. A supplier whose loads repeatedly arrive warm can be shown the records, and the conversation is about facts, not about blame. The supplier can then fix the cooling at the farm or the transport, and the next records will show whether the fix worked.
Payment can also follow the record, where the arrangement allows it. Milk paid by quality, grain paid by moisture and purity, and fruit paid by ripeness grade all need measures that are fair and checkable. Sensors that measure those qualities at the farm and at intake give both sides the same numbers, and the payment follows the numbers. The details of payment differ by arrangement and are not part of this unit, but the principle is plain: a shared, checkable record makes a fairer trade than a guess.
Traceability from the field
Traceability from the field means that a finished product can be followed back to the farm, the field, and the day it was grown or collected. The link is made by identity: a batch code on a crate, a tag on a pallet, a code on a delivery note. The sensor record is attached to that identity, so the temperature, moisture, and other readings for a load stay with the load through the factory and into the finished product. If a problem is found later, the batch code lets the maker find which farm, which field, and which day the material came from, and which other loads from the same source may share the problem.
This matters most when a problem is found after the food has been sold. A recall that can aim at the loads from one farm on one day is smaller, faster, and cheaper than a recall that must take back everything made in a week because nobody knows which material went where. The sensor record does not prevent the problem, but it limits its reach, and it shows the maker took care. The record is also useful when the question is not safety but quality: a batch of sauce that tastes different can be traced back to the tomatoes, the field, and the weather that week, and the cause can be found in the record rather than guessed.
Field records also support good practice over seasons. A farm that keeps soil, water, and yield records for several years can see which fields give the best crop, which watering pattern works, and which variety suits the soil. The records are a memory for the farm, and a new season starts with the lessons of the last one, not from scratch. The factory gains too: a supplier whose records show steady quality over seasons is a supplier worth keeping, and the records prove it.
Quick self-check
Question: A sauce tastes different in one batch. How does field traceability help?
The batch code links the sauce back to the tomatoes, the farm, the field, and the week they were grown. The field record for that week, rain, heat, watering, can then explain the difference, instead of guessing.
Power and connection on a farm
Farm sensors live far from power and networks, so their design is shaped by that fact. A sensor in a field may run on a battery that must last a season, or on a small solar panel that charges the battery by day. The sensor sends its reading rarely, perhaps once an hour, because each sending uses power, and a sensor that sends every minute will drain its battery in days. The reading rate is chosen to match the change: soil moisture changes slowly, so an hourly reading is enough, while a frost warning needs a reading often enough to catch a sudden fall in temperature at night.
The connection is often a long-range, low-power radio link. Such a link can carry a small message over a long distance, from a field sensor to a gateway at the farmhouse, using very little power. It cannot carry a large amount of data, such as a photograph, but a temperature or moisture reading is a small message, and the link suits it well. The gateway collects messages from many field sensors and passes them on over the internet to the farm office or the factory. Where no internet reaches the farmhouse, the gateway may store the readings and send them when a connection is available, or send them over a mobile network.
The weather is the enemy of farm sensors. Rain can enter a badly sealed housing, sun can heat a sensor that should read the air, mud can cover a solar panel, and animals can knock a sensor over. Housings are sealed and shaded, sensors are fixed firmly, and the farmer checks them as part of the daily round. A sensor that has fallen into the mud reads the mud, not the field, and the data will show a strange flat line or a sudden jump that prompts a visit. The human round and the sensor watch each other: the sensor watches the field, and the farmer watches the sensor.
Quick self-check
Question: A field sensor sends a reading once an hour. Is that often enough to catch a frost?
Often yes for a slow fall, but a sudden frost needs more frequent readings at night, or an alert set to trigger when the temperature nears the frost point. The rate must match how fast the danger can arrive.
Try it yourself
Choose a crop you know, such as wheat, rice, tomatoes, or potatoes. Draw a simple map of a field on paper and mark four places where you would put a soil moisture sensor, with one sentence for each mark saying why that spot. Then list three other sensors you would add: one for weather, one for the crop or the store, and one for the journey to the factory. For each one, write down what it measures, how often it should send a reading, and what action its reading would lead to. Finally, write a short note saying how the factory would know, from your sensors, whether to accept the load, hold it, or use it first. Compare your map with a classmate who chose a different crop: which sensors are the same, and which are different?
Key ideas to carry forward
- The quality of a food is fixed long before it reaches the factory gate, so sensors on the farm and on the journey matter as much as sensors in the plant.
- Soil moisture at root depth, at two depths, guides watering: water only where the soil is dry, and stop before the soil is waterlogged.
- A farm weather station measures the farm itself, and its readings guide sowing, watering, and frost protection.
- Animal sensors read movement, temperature, and yield per animal, so the animal that needs care is named early.
- Grain in store is watched for hot spots and moisture, because a hot spot is the first sign of dampness or insects.
- Milk, fruit, and grain each travel differently, and the sensor record from the farm and the truck tells the factory the history the intake test cannot.
- Intake is a decision made by written rules, fed by the sensor record, and recorded with its reason.
- Traceability links the finished product back to the farm, the field, and the day, so a problem can be aimed at the right loads.
- Farm sensors run on batteries or solar power and send small messages rarely, over long-range radio to a gateway, and they must survive sun, rain, mud, and animals.
Glossary
Ethylene
A gas given off by ripening fruit, which speeds ripening in nearby fruit. Food tech example: a sensor that reads ethylene in a truck shows whether a load of fruit is ripening too fast.
Farm weather station
A small group of sensors on a farm that reads air temperature, humidity, wind, rainfall, and light. Food tech example: a farm weather station warns of frost on a low field before dawn.
Gateway
A device that collects messages from many sensors and passes them on to the wider network. Food tech example: a gateway at a farmhouse gathers readings from field sensors and sends them to the farm office.
Heat stress
Harm to animals from hot conditions, which cuts yield and growth. Food tech example: a shed sensor warns when the air is hot enough to stress the animals inside.
Hot spot
A place in stored grain that is warmer than the grain around it, a sign of dampness or insect activity. Food tech example: a hot spot in a silo prompts the farmer to move air through the grain.
Intake
The receiving of raw material at a factory, with checks on its condition. Food tech example: at intake, milk is checked for temperature and acidity before it is accepted.
Irrigation
Supplying water to a crop by pipes, drip lines, or sprinklers. Food tech example: drip irrigation puts water at the roots of tomato plants, guided by soil moisture sensors.
Precision irrigation
Watering only where the soil is dry, and only as much as the soil needs, guided by sensor readings. Food tech example: precision irrigation waters the high dry ground and leaves the low damp ground alone.
Root zone
The depth of soil where a plant's roots take up water and nutrients. Food tech example: a soil moisture sensor is buried in the root zone, where the plant drinks.
Soil moisture sensor
A sensor buried in soil that reads how wet the soil is. Food tech example: a soil moisture sensor tells a grower when a field needs water.
Soil temperature
The temperature of the soil, which decides when seed can be sown and how it grows. Food tech example: a soil temperature sensor shows when the soil is warm enough to sow seed.
Solar panel
A panel that turns sunlight into electricity, used to charge a sensor battery on a farm. Food tech example: a small solar panel keeps a field sensor running through the season.
Traceability
The ability to follow a food and its history from origin to the shop shelf. Food tech example: a batch code links a sauce back to the field where its tomatoes grew.
Waterlogging
Soil so wet that it has no air for plant roots, which then rot. Food tech example: a soil sensor that stops watering in time prevents waterlogging.
Yield
The amount of crop or milk produced, often per animal or per area. Food tech example: a sensor in a milking system records the milk yield of each cow.
References
- Singh, R. P., Computer Applications in Food Technology, Academic Press, 1996.
- Irudayaraj, J., Food Processing Operations Modeling, Design, and Analysis, 2nd edition, 2002.
Unit 6: Data, Connectivity and Security in IoT Systems
Note
Unit 6: Data, Connectivity and Security in IoT Systems
A question to start with
A dairy has sensors in twenty cold stores, each sending a temperature reading every five minutes. That is 5,760 readings a day, and more than two million readings a year. The quality manager does not want two million numbers. She wants to know three things: did any store go too warm, for how long, and which batch was inside it at the time? The numbers are not the answer. The answer is what the numbers are turned into: a short list of faults, a history for each batch, and a chart that shows the pattern across a week. The hard part of IoT is not collecting data. It is carrying it, storing it, turning it into answers, and keeping it safe from people who should not see it or change it.
This unit is about the middle of the system: how readings travel, where they are stored, how they become useful, and how the whole system is protected.
Questions to think about
- A sensor sends a reading every minute, but the temperature changes slowly. Is that reading rate wasteful, useful, or harmful?
- A cold store loses its internet connection for a night. Should the sensor keep measuring, keep sending, or wait? What is lost in each case?
- Somebody changes a temperature record from 9 degrees to 4 degrees after a fault. How would you ever know, and why does it matter?
Think about the two million readings before you read on
Most readings say the same thing: all is well. The value is in the few readings that differ, the ones that show a rise, a fall, or a gap. A good system keeps all the readings for proof, but shows the manager only the exceptions, the trends, and the answers to her three questions. Showing everything hides the fault in a flood of normal numbers.
How readings travel: wires and wireless
A reading travels from a sensor to a gateway, and from the gateway to the place where it is stored. The first part of the journey, from sensor to gateway, is often wireless, because wires across a factory floor or a field are costly and easily damaged. Several kinds of wireless link are used, and each suits a different distance and amount of data.
Short-range wireless links carry messages across a room or a building. They use little power and are cheap, so they suit sensors inside a plant, where a gateway is never far away. Their weakness is range and walls: metal racks, thick walls, and steam can block them, so the gateway must be placed with care, and the signal strength must be checked after the plant is full of goods, not only when it is empty.
Long-range, low-power links carry small messages over a long distance, from a field sensor to a farmhouse gateway, or from a sensor in a large warehouse to a gateway at one end. They use very little power, so a battery can last a season or longer, but they carry only small messages, slowly. A temperature reading is a small message, so it suits them well. A photograph or a video would not.
Mobile phone networks carry readings from moving things, such as trucks, because the network follows the road. A truck sensor sends its readings over the mobile network when a signal is available, and stores them when it is not, sending them later when the signal returns. The cost is the mobile subscription and the power to send, so the reading rate is kept modest.
Wired links are still used where they suit: a sensor fixed to a machine that never moves, a sensor in a place full of radio interference, or a sensor that must report every second. A wire is steady, fast, and hard to block, but it ties the sensor to one place and costs time to install. The choice is a trade between freedom and steadiness, and a large plant often uses several kinds at once: wires where the machine is fixed, short wireless where the sensor moves a little, and mobile where the sensor travels.
Quick self-check
Question: A sensor in a field sends a reading once an hour over a long-range radio link. Why not send every minute?
Because each sending uses battery power, and the soil moisture changes slowly, so an hourly reading is enough. Sending every minute would drain the battery in days without adding knowledge.
Question: Why does a truck sensor store readings when there is no mobile signal?
So the record stays complete. The readings are sent when the signal returns, and a gap in the live view is understood as a signal gap, not a gap in the measurement.
Gateways, the cloud, and the edge
A gateway is the meeting point between many sensors and the wider network. It collects messages from sensors nearby, checks them, adds the time if the sensor did not, and passes them on in one stream. A gateway can also do a little thinking: it can average readings, spot a value outside the limit, and send an alert at once, without waiting for a distant computer. That local thinking is useful when the internet is down, because the gateway can still warn the people nearby.
The cloud is a name for computers reached over the internet that store data and run software for many users. A cloud system can store years of readings from many plants, show them on a dashboard reached from any office, and run the checks that turn readings into answers. Its strength is scale and reach: the same dashboard can be seen in the plant, in the head office, and on a phone. Its weakness is dependence: when the internet fails, the cloud cannot be reached, and the plant must still be able to see its own readings and sound its own alarms.
Edge computing is the opposite idea: doing some of the processing close to the sensor, on a small computer at the plant, such as the gateway itself. Edge processing keeps the essential jobs running when the internet is down: the local dashboard, the local alarm, and the storage of readings until the connection returns. It also cuts the amount of data sent: instead of sending every reading, the gateway can send a summary each hour and keep the full detail locally. The best systems use both: the edge keeps the plant safe and running, and the cloud keeps the long history, the wide view across plants, and the deep analysis.
The diagram shows the journey of a reading. The steps point to the diagram: the reading starts at the sensor, passes the gateway, is used locally at the edge and stored in the cloud, and ends as an answer a person can act on.
Quick self-check
Question: The internet fails at a plant. What should still work?
The local dashboard, the local alarm, and the recording of readings. Those are edge jobs, and they must not depend on the internet. The cloud history can wait until the connection returns.
What data is, and what makes it good
Data is the readings, with their times, their sensor names, and their units. A reading without a time is a number without a place in history. A reading without a sensor name cannot be traced to a place. A reading without a unit, degrees or percent or bar, cannot be read at all. Good data carries all three, every time, so that any reading can be understood on its own, months later, by a person who was not there when it was taken.
Good data also has four qualities. The first is completeness: the readings that should be there are there, and gaps are marked as gaps, not filled in silently. A gap filled with a guessed value is a lie in the record, and a later reader cannot tell the guess from a real reading. The second is correctness: the sensor was calibrated, the placement was sound, and the reading is true within its stated accuracy. The third is timeliness: the reading arrives soon enough to act on. A warning that arrives a day late is a history note, not a warning. The fourth is order: the readings are stored in a way that lets a question be answered quickly, such as all readings for one store, or all readings for one batch, in time order.
Bad data has plain signs. Readings that never change, readings that jump wildly, readings with impossible values, such as a freezer at 40 degrees, and readings that arrive with no time attached are all signs to check. The check is not to delete the bad readings quietly, but to mark them, find the cause, and keep a note of what was done. Deleting a bad reading without a note hides the fault that made it, and the fault will come back.
Quick self-check
Question: A gap in a temperature record is filled in with the average of the readings on either side. Is that good practice?
No. A filled gap looks like a real reading, and a later reader cannot tell it was guessed. The gap should stay marked as a gap, with a note saying why it happened.
From readings to answers: processing and dashboards
Processing turns a stream of readings into answers. The simplest processing is a check against a limit: is this reading above the limit, and for how long? The next is a summary: the average, the highest, and the lowest reading in an hour or a day. The next is a trend: is the temperature rising slowly across a week, is the vibration growing month by month, is one store always warmer than the others? Each step answers a question a manager actually asks, and each step throws away detail that is not needed for that question, while keeping the full record for the day a deeper question is asked.
A dashboard is where the answers are shown. A good dashboard follows three rules. First, it shows the present state first: what is wrong now, and what needs action now. Second, it shows trends second: a chart of the last day or week, so the present can be judged against the past. Third, it hides detail until it is asked for: the full list of readings is one click away, not on the first screen. A dashboard that shows everything at once shows nothing, because the eye cannot find the fault in a crowd of numbers.
Reports are the dashboard in written form: a daily summary for the supervisor, a weekly summary for the manager, a batch report for the quality file. A report names the period, the place, the faults found, the actions taken, and the readings that support them. A report that says "all was well" with no readings behind it is a claim, not a proof. A report that lists the faults and the fixes is a record the plant can learn from.
Quick self-check
Question: A dashboard shows 200 sensor readings on one screen. What is wrong with it?
It shows everything and highlights nothing. The eye cannot find the fault in a crowd of numbers. The dashboard should show first what needs action, then the trends, and keep the detail one click away.
Who can see, who can change, and who can act
Data is valuable, and valuable things are protected. Three questions decide protection: who can see the data, who can change it, and who can act on it? Seeing is the widest right: an operator sees their own line, a supervisor sees several lines, a manager sees the plant. Changing is narrower: only a named person can change a limit, a calibration, or a sensor setting, and the change is recorded with who made it and when. Acting is the narrowest in some ways: only a trained person can release a held batch, or restart a line after a safety stop.
Passwords and accounts are the basic tools. Each person has their own account, and shared accounts, where several people use one login, hide who did what. When a fault is found, the record should show who changed the limit, who accepted the batch, and who answered the alert. A shared account cannot answer those questions. Passwords must be strong enough to resist guessing, and they must be changed when a person leaves. These are plain habits, but they are the foundation, and without them the cleverer protections are built on sand.
Access from outside the plant needs extra care. A dashboard that can be seen from any phone is convenient, and it is also a door. That door must be locked with a strong login, and it should only open to the people who need it. A system that lets anyone on the internet reach the plant network is a risk to the plant and to the food. The rule is simple: the fewest doors, the fewest people through each door, and a record of who passed through.
Quick self-check
Question: Three operators share one login for the dashboard. What is lost?
The record of who did what. When a limit is changed or a batch is released, nobody can tell which operator did it. Each person needs their own account, so the record names the person.
Security: keeping the system honest
Security in IoT has two sides: keeping outsiders out, and keeping the record honest. Keeping outsiders out means the sensors, the gateway, and the software cannot be reached by people who have no business there. The plain steps are: change every default password, because default passwords are widely known; keep the software updated, because updates fix faults that attackers use; and close every connection that is not needed, because each open connection is a door. A sensor that was installed years ago and never updated is a door left open, and a plant that does not know how many sensors it has cannot protect them.
Keeping the record honest means a reading cannot be changed quietly after the event. A temperature record that can be edited without a trace is not proof: a warm load can be made to look cold by changing the numbers. Honest records are protected in two ways. First, the system records who changed what and when, so any change leaves a trace. Second, the readings are stored in a way that makes silent change hard, such as keeping a copy that cannot be edited, or keeping the readings in two places so a change in one can be spotted against the other. The detail of the method matters less than the principle: a record that can be changed without a trace is not a record, it is a story.
The most serious risk is to control: a system that can start a machine, open a valve, or change a temperature set point can be misused if it is reached by the wrong person. Control actions must be limited to the people who need them, they must need a clear login, and they must be recorded. A plant should also be able to run without the network: if the connection is lost or the system is attacked, the plant must fall back to local control, with a person at the machine, rather than standing still or running blind. That fallback is planned before it is needed, and tested, like any other backup.
Human error is a larger risk than attack in most plants. A limit typed wrong, a sensor plugged into the wrong gateway, a calibration entered for the wrong sensor, a batch code typed wrong: each of these corrupts the record without any attacker. The guards are plain: limits are checked by a second person, sensors are labelled, batch codes are scanned rather than typed where possible, and changes are reviewed each week. A system that makes the right action easy and the wrong action hard is a secure system, even before any lock is added.
Quick self-check
Question: A temperature record is changed from 9 degrees to 4 degrees after a fault, with no trace of the change. Why does that matter?
Because the record is the proof that the food was kept safe. A record that can be changed silently cannot prove anything, and a warm load can be passed off as cold. The change must leave a trace, with who made it and when.
Question: Why is changing a default password a security job, not a tidy-up job?
Because default passwords are widely known, so a device left with its default password can be reached by anyone who knows it. Changing it closes a door that was open to the whole world.
Privacy, ownership, and the law of the record
Data about a plant belongs to the plant, and data about a farm belongs to the farm, but IoT data often travels across owners: from a farm sensor to a factory system, from a truck to a customer. Each handover should be agreed: who can see the data, what they can do with it, and how long they can keep it. A farmer who shares soil data with a buyer should know what the buyer will do with it. A factory that shares batch records with a customer should know the customer will use them to judge the batch, not to judge the factory's other business. These agreements are written down, in plain words, before the data flows.
Personal data needs care too. A sensor that records who was on shift, a camera that watches a line, or a movement sensor worn by a worker all touch people, not only machines. The people concerned should know what is recorded, why, and who can see it. Data that is collected for food safety should be used for food safety, not for watching workers for other reasons. The principle is the same as for food data: collect only what is needed, use it only for the purpose agreed, and keep it only as long as it is needed.
Keeping data has a cost and a limit. Years of readings take storage space, and old readings are rarely needed. A retention rule says how long each kind of record is kept: batch records for the life of the product and beyond, calibration records for the life of the sensor, and raw readings for a set period after which summaries are kept and detail is removed. The rule is written down, and the removal is recorded, so that a missing old record is understood as a record removed by rule, not a record lost by accident.
When the system fails: what to do
Every system fails, and the plan for failure is part of the system. The plan names what happens when the network fails, when the gateway fails, when a sensor fails, and when the cloud cannot be reached. In each case the plan says three things: what the people on site do, what the local system does, and who is called. The plan is short, written in plain words, and kept where the people who need it can find it, not only in a file on a computer that needs the network to open.
A failure drill tests the plan. The network is switched off for an hour, and the staff check: does the local alarm sound, do the sensors keep recording, does the dashboard in the plant still work, and do the readings arrive when the network returns? The drill is recorded, and any gap found is fixed. A plan that has never been tested is a guess, and a guess is a poor guard for food safety.
After a failure, the record is reviewed. What failed, why, how long it lasted, what was lost, and what was done to fix it? The review is written down, and the fix is checked a month later to see if it held. This is the same habit as for any quality fault: find the cause, fix it, and check that the fix worked. A system that is never reviewed will fail the same way twice.
Quick self-check
Question: The network fails for a night. What should the morning review ask?
What failed, why, how long it lasted, whether the local alarms worked, whether the readings were stored and sent later, and what must change so the same failure does less harm next time.
Try it yourself
Pick a cold store you know, real or imagined, with three sensors sending a reading every five minutes. Work out how many readings that store makes in one day, and in one week. Then write down, in three short lines, the three answers the manager wants each morning: the highest temperature, the longest time above the limit, and the batch that was inside. Next, write a security rule for that store in three lines: who can see the readings, who can change the limit, and what happens to the record if somebody tries to change a reading after the event. Finally, write a failure plan in three lines: what happens when the internet fails, what happens when the gateway fails, and who is called. Compare your plan with a classmate: whose plan keeps the store safe when the network is down?
Key ideas to carry forward
- A reading is a value plus a time, a sensor name, and a unit: without those, it cannot be understood months later.
- Wireless links suit different jobs: short range in a plant, long range and low power in a field, mobile networks for trucks, wires where steadiness matters.
- The gateway collects and checks, the edge keeps the local dashboard and alarm working when the internet is down, and the cloud keeps the long history and the wide view.
- Good data is complete, correct, timely, and ordered, and gaps are marked as gaps, never filled in silently.
- Processing turns readings into answers: limit checks, summaries, and trends, and a dashboard shows first what needs action.
- Access is layered: many can see, few can change, and every change leaves a trace with who made it and when.
- Security means changing default passwords, keeping software updated, closing unused connections, and keeping control actions limited and recorded.
- A record that can be changed without a trace is not proof, and a failure plan that has never been tested is a guess.
Glossary
Access control
Rules that decide who can see, change, or act on data in a system. Food tech example: only a quality manager can change the temperature limit for a cold store.
Cloud
Computers reached over the internet that store data and run software for many users. Food tech example: temperature records from many cold stores are stored in the cloud so a manager can check them from any office.
Dashboard
A screen that shows the present state of a system in one view. Food tech example: a dairy dashboard shows the temperature of every milk tank at a glance.
Data
Readings with their times, sensor names, and units, stored so questions can be answered from them. Food tech example: a day of temperature readings is the data that proves a cold store stayed cold.
Edge computing
Doing some data processing close to the sensor, on a small computer at the plant. Food tech example: a gateway at a bakery processes oven readings locally so an alarm still sounds when the internet is down.
Gateway
A device that collects messages from many sensors and passes them on to the wider network. Food tech example: a gateway in a packing hall gathers readings from a dozen sensors and sends them to the office system.
Mobile network
A phone network that carries data from moving things such as trucks. Food tech example: a truck sensor sends its temperature readings over the mobile network while on the road.
Processing
Turning raw readings into useful answers, such as averages, highest values, and times when a limit was passed. Food tech example: processing turns a day of temperature readings into the highest value and the time it happened.
Retention rule
A written rule saying how long each kind of record is kept before it is removed. Food tech example: batch records are kept for the life of the product and beyond, while raw readings are kept for a set period.
Security
Protecting a system and its records from people who should not reach them or change them. Food tech example: changing a default password keeps an outsider from reaching a sensor.
Sensor name
The identity of a sensor, showing which place it measures. Food tech example: a reading labelled "Cold store 2, door" can be traced to that exact place.
Summary
A short form of many readings, such as an average, a highest value, or a count of faults. Food tech example: a daily summary shows the highest temperature in each cold store.
Trend
The direction of readings over time, such as a slow rise or fall. Food tech example: a trend of rising temperature across a week shows a cooler losing power.
Wireless link
A radio connection that carries sensor messages without a wire. Food tech example: a wireless link carries readings from a grain silo to the farm office.
Wired link
A cable connection that carries sensor messages, steady and fast but tied to one place. Food tech example: a wired link carries readings from a pasteuriser sensor to the control room.
References
- Singh, R. P., Computer Applications in Food Technology, Academic Press, 1996.
- Irudayaraj, J., Food Processing Operations Modeling, Design, and Analysis, 2nd edition, 2002.
Unit 7: Building an IoT System: a Worked Example
Note
Unit 7: Building an IoT System: a Worked Example
A question to start with
A small dairy makes yoghurt in one plant. It has four cold stores, two pasteurisers, one filling line, and three delivery trucks. The owner has lost two loads this year to cold store failures found too late, and the quality manager spends an hour each morning walking the plant with a clipboard, reading dials and writing numbers down. The owner asks a plain question: can a small IoT system watch the cold stores, the pasteurisers, and the trucks, warn us before food is lost, and give us a record we can show an auditor? This unit answers that question by building the system step by step, on paper, so the learner can see every decision, every trade, and every check. No system is bought, no brand is named, and the numbers are working examples to show the method, not claims about any real plant.
This unit pulls together everything from the earlier units: sensors, placement, cold chain, processing lines, farming and supply, and data, connectivity, and security. Read it as a worked example to copy in shape, not in detail, because every plant is different, and the shape is what travels.
Questions to think about
- Before buying a single sensor, what would you need to know about the dairy: its foods, its limits, its people, and its worst losses?
- A small dairy cannot watch everything at once. What would you watch first, and what would you leave for later?
- How would you prove, after a month, that the system had earned its keep?
Think about the dairy before you read on
Start with the losses, not the catalogue. The dairy lost loads to cold store failures, and the manager spends an hour a day reading dials. The first job is therefore cold store watching and warning, with an automatic record that replaces the clipboard round. The pasteurisers and the trucks follow, because their failures are rarer but their proof matters. The order is set by the loss, not by the gadget.
Step 1: name the problem and the food
Every build starts on paper, with the problem written in plain words. For this dairy, the problem has three parts. First, cold store failures are found too late, and loads are lost. Second, the pasteuriser record is on paper charts, which are hard to search when a batch is questioned weeks later. Third, the trucks have no record of the journey, so a warm load on arrival starts an argument about where it warmed. The food is yoghurt, a chilled food that must stay cold from making to the shop, and the safety limit for the cold stores is a set temperature that the dairy's quality plan already states. The system must watch that limit, warn when it is passed, and keep the proof.
The scope is written down too: four cold stores, two pasteurisers, one filling line, and three trucks. Nothing else is in scope for the first build. The farm tanks are left out, because the dairy buys milk from farms it does not own, and the farm record is a later step. The shop chillers are left out, because they belong to the customers. A small scope that is finished is better than a large scope that is half built. The later steps can add to it, and the system is designed so that adding a sensor later is easy.
The people are named at the same time. The duty manager receives alerts at night. The store supervisor receives alerts in the day. The quality manager owns the records and the calibration. The maintenance engineer owns the sensors and the gateway. Each person has one job in the system, and the jobs are written down, so that when an alert sounds, the person who must act knows it is theirs.
Quick self-check
Question: Why write the scope down before buying anything?
Because a written scope stops the build from growing while it is being built. A small scope that is finished and working is better than a large scope that is half built and never trusted.
Step 2: choose what to measure, and where
The measurements follow the problem. For each cold store, the dairy measures air temperature at two places: near the door, and in the middle of the stacks, the warmest places found by a survey with temporary sensors. It also measures the door, with a sensor that records each opening and how long the door stays open, and the cooler state, running or stopped. One store also gets a product temperature sensor inside a pallet, to check that the air reading matches the food. That is three or four sensors per store, and the dairy accepts that more sensors would give a finer picture, but the cost and the work of maintaining them would grow too.
For each pasteuriser, the dairy measures the holding temperature with an accurate sensor, the holding time, and the outlet temperature. The holding temperature sensor is the safety sensor, so it is the most accurate type the dairy can afford, it is calibrated often, and a second sensor sits beside it for comparison. The pair must agree, and a difference between them is itself an alert, because one of them has drifted.
For the filling line, the dairy measures the fill weight with a checkweigher, the seal temperature, and the count of packs made and rejected. The seal temperature is watched as a trend across the morning, so a slow drift is seen before seals fail. The count supports the stock record and the waste count.
For each truck, the dairy measures the load temperature at three places, front, middle, and back, with a sensor that stores readings when the mobile signal is lost and sends them when it returns. The driver sees the temperature on a small display in the cab, and the office sees it on the dashboard. The truck is pre-cooled before loading, and the record must show a flat, low line from the start, not a slow cooling curve.
The reading rates are set to match each process. Cold stores send a reading every five minutes. Pasteurisers send a reading every few seconds while running, because their temperature can change fast. The filling line sends a weight for each pack and a seal temperature every minute. Trucks send a reading every five minutes, storing when the signal is lost. The rates are written down with the reason for each, so that a later change is a decision, not a guess.
The diagram shows the four parts of the dairy system. The steps point to the diagram: each part measures the condition that decides its food, at the places where a fault would hide first.
Quick self-check
Question: Why does the pasteuriser get two temperature sensors side by side?
Because its temperature decides safety, and a difference between two sensors is itself a warning that one has drifted. The pair checks each other, so a silent drift is caught.
Step 3: connectivity, gateway, edge, and cloud
The dairy plant is one building, so short-range wireless links carry readings from the cold store and line sensors to a gateway in the plant office. The gateway is placed where its signal reaches every sensor, checked after the stores are full of pallets, not only when they are empty, because full pallets block radio signals. The pasteuriser sensors, which must report every few seconds and never miss a reading, are wired to the gateway, because a wire is steadier than a radio link for a fast, safety-critical reading. The trucks use the mobile network, with storage when the signal is lost.
The gateway does the local thinking. It checks each reading against its limit, sends a local alarm in the plant when a limit is passed, and shows the plant dashboard, so the plant keeps working when the internet is down. It stores the readings locally, so a night without internet loses no data, and it sends a summary to the cloud each hour, with the full detail following when the connection is good. The cloud keeps the long history, shows the dashboard to the manager from any office, and makes the weekly and batch reports. The split is deliberate: the edge keeps the plant safe, and the cloud keeps the history and the wide view.
Power is planned too. The plant sensors run on mains power where a wire is already there, and on batteries where a wire would be a trip hazard. Battery sensors are checked on a monthly round, and their battery level is itself a reading on the dashboard, so a dying battery warns before the sensor goes silent. The gateway and the local dashboard run on mains power with a battery backup that keeps them alive long enough to send an alert and keep recording through a short power cut.
Quick self-check
Question: Why are the pasteuriser sensors wired, while the cold store sensors are wireless?
Because the pasteuriser reading is fast and safety-critical, and a wire is steadier than a radio link. The cold store changes slowly, so a wireless link that sends every five minutes is enough, and it avoids wires across a busy floor.
Step 4: alerts, dashboards, and the people who answer them
The alert rules are written before the system goes live, and they are tuned in the first month. The cold store limit is set a little below the true safety limit, so the warning comes while there is still time to act. The delay is ten minutes, so a short rise from a door opening for loading does not send an alert, but a lasting rise does. The alert names the store, the temperature, the time, and how long it has been above the limit. At night it goes to the duty manager, in the day to the store supervisor, and if the first person does not answer within fifteen minutes, it goes to the second person. That rota is written down and pinned beside the dashboard.
The pasteuriser alert is stricter. Any fall below the holding temperature, even for a short time, sends an alert at once, with no delay, because the safety condition has been broken. The line stops, the batch is held, and the quality manager decides what happens to it. The rule is strict because the consequence is safety, not only quality, and the dairy accepts the occasional false alert as the price of a strict rule on a safety step.
The dashboard has three views. The operator sees one line at a time, with its speed, temperatures, counts, and any alert that needs action. The supervisor sees the whole plant: four stores, two pasteurisers, one line, and three trucks, each as a simple tile that is green when all is well and red when it needs action. The manager sees the week: faults found, actions taken, loads held, and the trend of each store. Each view shows what that person needs, and no more. The dashboard is checked each morning, and the morning check is a job, not a glance: the supervisor looks at the overnight chart, notes any fault, and writes the action taken.
Alert fatigue is watched from the first week. Each alert is counted, and each month the quality manager reviews how many alerts were sent, how many were real faults, and how many were false. If false alerts are common, the limit or the delay is tuned. If real alerts are missed, the limit is tightened. The review is recorded, so the tuning is a decision with a reason, not a quiet change.
Quick self-check
Question: Why does the pasteuriser alert have no delay, while the cold store alert waits ten minutes?
Because the pasteuriser temperature decides safety, and any fall below it breaks the safety condition at once. The cold store warms slowly, and a short rise from a door opening is normal, so a delay filters out those harmless rises.
Step 5: testing, calibration, and going live
The system is tested before it is trusted. First, each sensor is checked against a known reference: the cold store sensors in iced water, the pasteuriser sensors against a reference thermometer, the checkweigher with a known test weight, the truck sensors in a cold room. The calibration is recorded with the date, the reference used, and who did it. A sensor that fails its check is not installed, and a sensor that passes is labelled with its name, its place, and its calibration date.
Second, the whole chain is tested end to end. A cold store door is left open, and the team watches: does the temperature rise on the dashboard, does the alert arrive after ten minutes, does the right person receive it, and does the record show the rise and the fall when the door is shut? A pasteuriser temperature is lowered in a test, and the team watches: does the alert arrive at once, does the line stop, and is the batch held? A truck drives through an area with no signal, and the team watches: do the readings arrive later, complete, when the signal returns? Each test is recorded, and any gap found is fixed before the system goes live.
Third, the backup is tested. The internet is switched off for an hour, and the team checks: does the local dashboard still work, does the local alarm sound, do the sensors keep recording, and do the readings arrive when the internet returns? The mains power is switched off for a short time, and the team checks the battery backup. A backup that has never been tested is a guess, and the dairy tests it before trusting it.
Going live is done one part at a time, not all at once. The cold stores go live first, because they are the reason for the build, and they run for two weeks while the team watches the alerts and tunes the limits. The pasteurisers follow, then the filling line, then the trucks. Each part goes live only when the part before it is working and trusted. The clipboard round is not stopped at once: for the first month, the manager still walks the plant each morning and compares the clipboard readings with the dashboard. When the two agree for a month, the clipboard round stops, and the dashboard becomes the record.
Step 6: the first month, and what the data shows
The first month of data is a lesson in the plant itself. The cold store charts show a pattern the dairy had not seen: one store warms every afternoon. The door record shows the cause: deliveries arrive at midday, and the door stands open while pallets are moved. The fix is not a bigger cooler, but a change in delivery time and a strip curtain on the door. After the fix, the afternoon rise disappears from the chart, and the dairy can see the fix worked, because the chart before and after the fix is in the record.
The pasteuriser record shows a second lesson: one pasteuriser holds its temperature well, while the other drifts down by a small amount each week. The paired sensors show the drift, and a check finds a valve that is sticking. The valve is serviced between batches, and the drift stops. Without the record, the drift would have grown until the pasteuriser failed during production.
The filling line shows a third lesson: the average fill weight drifts upward across each morning, giving product away. The filler is adjusted, and the drift stops. The checkweigher record shows the saving: the average weight returns to target, and the product given away falls to almost nothing. The truck records show a fourth lesson: one truck cools slowly after loading, because it is not pre-cooled. The driver is reminded to pre-cool, and the next records show a flat, low line from the start.
None of these lessons needed a new sensor. They needed the data to be looked at, the pattern to be seen, and the cause to be fixed. The dairy sets a weekly habit: the supervisor and the quality manager sit together for half an hour, look at the week's charts, name the faults found, and agree the fixes. The habit is the system, as much as the sensors are.
Quick self-check
Question: A cold store warms every afternoon, and the door record shows long openings at midday. What is the fix?
Change the cause, not the cooler: move deliveries to a cooler time, shorten the door opening, or fit a strip curtain. Then watch the chart to prove the fix worked.
Question: Why keep the clipboard round for the first month after going live?
So the dashboard readings can be checked against a hand reading the team already trusts. When the two agree for a month, the dashboard has earned trust, and the clipboard round can stop.
Step 7: security, records, and the audit
Security is built in from the start, not added later. Each person has their own account, and shared accounts are not used, so the record shows who changed a limit, who released a batch, and who answered an alert. Default passwords are changed on every device. The gateway and the software are updated on a monthly round, and the update is recorded. Connections that are not needed are closed, and the dashboard is reached only through a login, from the plant and from the office, not from the open internet without protection.
The record is protected too. Readings cannot be changed without a trace: any change is recorded with who made it and when, and a copy of the readings is kept that cannot be edited. When the quality manager holds a batch, the hold and the reason are recorded with the batch number, and when the batch is released or rejected, that decision is recorded too. The batch record is therefore a full story: the ingredients, the process readings, the checks, the faults, the holds, and the decisions, all linked by the batch number and the time.
The audit is the test of the whole system. An auditor asks for a batch made three weeks ago, and the quality manager finds it in minutes: the pasteuriser record shows the holding temperature and time, the cold store record shows the storage, the truck record shows the journey, and the filling record shows the weights and seals. The calibration records sit beside the readings, showing that the sensors were checked and true. The alert records show the faults that were found and the actions taken. The auditor does not need to trust a claim: the record is there to be read. That is the point of the system, and the reason it was built.
Cost, judgement, and what comes next
The cost of the system is counted honestly: sensors, gateway, software, installation, calibration, batteries, and the time staff spend answering alerts and reviewing data. The gain is counted honestly too: loads saved, product given away stopped, time saved on the morning round, and the audit passed without a search through paper charts. In the first year, the dairy counts the loads saved and the product saved, and compares them with the cost. If the gain is larger than the cost, the system has earned its keep. If not, the dairy asks what is wrong: too many sensors, alerts that waste time, or a part of the system that is not used.
What comes next is a choice, not a rush. The farm tanks are the natural next step, because the milk record would then start at the farm, not at the factory gate. The shop chillers are a step after that, in agreement with the customers. A second plant, if the dairy grows, would copy the same shape: the same layers, the same alert rules, and the same habits, fitted to the new plant's foods and limits. The shape travels, and the detail is fitted each time.
The final lesson is the simplest one in this subject. A sensor does not save food. A person who reads the sensor, acts on it, fixes the cause, and keeps the record saves food. The system is a tool for that person, and like any tool it is only as good as the hand that uses it, the habit that checks it, and the record that proves it. Build the tool well, but build the habit too, and the food will be safer for it.
Quick self-check
Question: An auditor asks for the record of a batch made three weeks ago. What should the dairy be able to show in minutes?
The pasteuriser holding temperature and time, the cold store history, the truck journey, the fill weights and seal checks, the calibration records for the sensors, and any holds and decisions, all linked by the batch number.
Try it yourself
Choose a small food business you know: a bakery, a dairy, a juice maker, or a cold store. Write a one-page plan for its first IoT system, following the seven steps in this unit. Name the problem in two sentences, list the foods and their limits, choose what to measure and where, name the connection and the gateway, write one alert rule with its limit, delay, and receiver, and name the person who owns the records and the calibration. Then write a test in three lines: one test for a sensor, one test for an alert, and one test for the backup when the internet fails. Finally, write one sentence saying how you would prove, after a month, that the system had earned its keep. Bring the plan to your next session and compare it with a classmate: whose plan watches the biggest loss first?
Key ideas to carry forward
- Start with the problem and the food, not the catalogue: name the losses, the limits, and the people before buying a sensor.
- Measure what decides the food, at the places where a fault would hide first, and set the reading rate to match how fast the process changes.
- Wire the fast, safety-critical sensors, use wireless where the process is slow, and use the mobile network with storage for trucks.
- The edge keeps the plant safe when the internet is down, and the cloud keeps the long history and the wide view.
- Alerts need a limit below the safety line, a delay that filters harmless rises, a named receiver, and a second person if the first does not answer.
- Test every sensor against a reference, test the whole chain end to end, test the backup, and go live one part at a time.
- Keep the old check running beside the new system until the new system has earned trust.
- Look at the data every week with a habit: name the faults, fix the causes, and check the chart to prove the fix worked.
- Security means own accounts, changed passwords, updated software, and a record that cannot be changed without a trace.
- A sensor does not save food: a person who reads it, acts on it, and keeps the record does.
Glossary
Audit
A check by an inspector that a plant followed its own rules and kept the proof. Food tech example: an auditor asks for the pasteuriser record of a batch made three weeks ago.
Batch number
A code that identifies one batch of food made at one time, used to link records to that food. Food tech example: a batch number on a pallet of yoghurt links its temperature record to that day's production.
Calibration record
The written result of checking a sensor against a known reference, with the date and who did it. Food tech example: a calibration record shows a cold store sensor was checked in iced water last month.
Dashboard view
A screen layout fitted to one job, such as the operator view, the supervisor view, or the manager view. Food tech example: the operator view shows one line, while the supervisor view shows the whole plant.
Delay
The time a value must stay above a limit before an alert is sent, used to filter out short harmless rises. Food tech example: a ten minute delay stops a cold store alert when the door opens briefly for loading.
Edge
Processing done close to the sensor, on a small computer at the plant. Food tech example: a gateway at a dairy processes readings locally so an alarm still sounds when the internet is down.
Failure drill
A planned test of what happens when part of a system fails, such as switching off the internet for an hour. Food tech example: a failure drill checks that the local alarm sounds when the network is down.
Gateway
A device that collects messages from many sensors and passes them on to the wider network. Food tech example: a gateway in a dairy office gathers readings from the cold stores and sends them to the cloud.
Going live
Starting to use a new system for real work, after testing. Food tech example: the cold stores went live first, and the trucks followed two weeks later.
Pre-cooling
Cooling a truck or a store before loading it with cold food, so the food does not warm on loading. Food tech example: a truck is pre-cooled before a load of chilled yoghurt is put in.
Scope
The written boundary of a build: what is included, and what is left for later. Food tech example: the first build covers the cold stores, pasteurisers, filling line, and trucks, and leaves the farm tanks for later.
Strip curtain
A curtain of hanging strips on a cold store door that lets people and pallets pass while cutting the flow of warm air. Food tech example: a strip curtain on a cold store door stops the afternoon warming from long door openings.
Working example
A full case worked through step by step to show the method, using example numbers to illustrate it. Food tech example: this unit is a working example of building a small dairy IoT system.
Rota
A written list saying which named person receives alerts at each time, with a second person if the first does not answer. Food tech example: the night rota sends cold store alerts to the duty manager, and to the supervisor if there is no answer in fifteen minutes.
References
- Singh, R. P., Computer Applications in Food Technology, Academic Press, 1996.
- Irudayaraj, J., Food Processing Operations Modeling, Design, and Analysis, 2nd edition, 2002.