GIFSQ ID: FTN-0090
Industry 4.0 in Food Industry
Unit 1: What Industry 4.0 Means for a Food Factory
Note
Unit 1: What Industry 4.0 Means for a Food Factory
A question to start with
A biscuit factory runs three shifts a day. On Monday morning the ovens bake well, the biscuits are golden, and the packing line moves fast. On Tuesday afternoon the same ovens, the same recipe, and the same people produce biscuits that are pale on one side and dark on the other, and the packing line stops four times. Nobody changed the recipe. The flour came from the same supplier. The weather was a little warmer on Tuesday, the oven doors were opened more often, and a conveyor belt was running a little slower than usual. Each small change on its own seemed too small to matter, but together they changed the result. The factory manager asks a simple question: how can we see these small changes while they are happening, instead of finding out about them when the biscuits are already baked?
This unit is about that question. Industry 4.0 is a name for a way of running a factory in which machines, sensors, computers, and people share information continuously, so that small changes are seen early, causes are found quickly, and decisions are based on what is actually happening on the line. It is not one machine and it is not one piece of software. It is a set of connected ideas that change how a food factory measures, controls, records, and improves its work.
Food factories have always used machines. A mill grinds grain, a mixer blends dough, an oven bakes, a filler puts sauce into jars, a sealer closes the pack. For a long time each machine worked on its own. An operator watched a dial, turned a knob, wrote a number in a logbook, and passed the information on by talking to the next person. That way of working can produce good food, and many factories still work that way. Its weakness is that information moves slowly, it lives in separate places, and it depends on one person noticing a change and remembering to tell someone else. When a problem appears, the record of what happened is often incomplete, and the cause has to be guessed.
Industry 4.0 changes the flow of information. A temperature sensor in an oven sends its reading to a computer every few seconds. A flow meter on a milk line records how much milk passed through. A camera checks the colour of biscuits as they leave the oven. A weighing machine checks the weight of every pack. All of these readings are stored together, in the same place, with the time attached to each one. When something goes wrong, the factory can look back at the record and see the exact minute the temperature drifted, the exact batch of flour that was in use, and the exact speed of the conveyor. When something goes right, the factory can see the exact conditions that produced the good result and repeat them.
The name Industry 4.0 comes from the idea that industry has passed through earlier revolutions. The first used steam power to drive machines. The second used electricity and production lines, so that many workers each did one step in order. The third used computers and automation, so that machines could follow a fixed program without a person turning every knob. The fourth, the one this subject is about, connects those computers and machines together, so that they share data, learn from it, and adjust themselves. A food factory does not need to throw away its ovens and mixers to take part in this. It needs to connect what it already has, add measurement where it is missing, and use the information well.
Questions to think about
- If an oven temperature drifts by a few degrees and nobody notices for an hour, how much food is affected, and who finds out first: the operator, the laboratory, or the customer?
- A factory buys a new sensor that measures moisture in flour as it enters the mixer. What decisions could that one measurement improve?
- Two factories make the same jam. One keeps paper logbooks, the other keeps a shared digital record of every batch. When a customer asks a question about one jar, which factory can answer faster, and why?
Think about the biscuit question before you read on
The pale and dark biscuits came from a set of small changes acting together: warmer air, doors opened more often, and a slower belt. None of these alone would explain the result, and a paper logbook written once an hour would miss most of them. A connected factory would have a record of oven temperature minute by minute, belt speed, and door openings, so the manager could see the pattern and fix the belt speed or the door routine before the next batch. The point is not that the machines are clever: the point is that the information is visible while there is still time to act.
From steam to connection: a short history
The story of industry is a story of new sources of power and new ways of organising work. In the first industrial revolution, steam engines drove mills and looms. Work that had been done by hand or by water power could be done faster and in one place, so factories grew up around the engine. Food work changed too: milling, brewing, and baking could run at a larger scale than a small workshop allowed.
In the second revolution, electricity and the production line changed the shape of the factory. Each worker or machine did one step, and the product moved along the line from step to step. Canning lines, bottling lines, and packing lines are examples of this idea in food. The line made large volumes possible, and it made each step simpler to learn and to repeat. Its weakness was rigidity: changing the product or the pack size meant stopping the line and rebuilding parts of it.
In the third revolution, computers and programmable controllers entered the factory. A programmable logic controller, a small industrial computer, could follow a set of instructions: open a valve for ten seconds, start a mixer, stop it when the temperature reaches a set point. Machines could repeat a sequence exactly, day after day, without a person standing at each switch. Food factories gained better repeatability, and the first digital records appeared. Each controller, though, still worked mostly on its own. The mixer controller did not talk to the oven controller, and the packing machine did not talk to the laboratory.
Industry 4.0 builds on all three earlier steps and adds connection. The machines keep their programs, but they also send their data to shared systems, and they receive instructions back. A temperature reading from the oven can change the speed of the conveyor. A moisture reading from the flour can change the amount of water added to the dough. A weight reading from the packing line can stop the filler before a hundred underweight packs are made. The factory becomes a system in which information flows in all directions, not only from the operator to the machine.
The diagram shows the loop that runs through every unit in this subject. The steps point to the diagram: connection without recording is lost, recording without understanding is only storage, and understanding without action changes nothing.
Quick self-check
Question: A factory installs sensors everywhere but nobody looks at the data. Which part of the loop is missing?
Understanding and action. The connection and recording steps are done, but the data is not studied and no process change follows, so the factory is no better than before, only busier with data.
Question: Why does a temperature reading need a time and a batch number attached to it?
Because a number without context cannot be used. The time shows when the change happened, and the batch number shows which food was affected. Without those two facts, the reading cannot be linked to a cause or to a product.
The four ideas that hold Industry 4.0 together
Industry 4.0 is often described with a long list of technologies, and the list can feel confusing. Four ideas hold the list together, and every technology in the later units serves one or more of them.
The first idea is connection. Machines, sensors, computers, and people are linked so that information can move between them. A sensor on a pasteuriser sends its reading to a control system and to a record that managers can see. Connection is the foundation: without it, each device is an island, and its data stays where it was made.
The second idea is information. Connection is useless if the information is messy, late, or wrong. A temperature reading is useful only if the factory knows which oven it came from, which batch was inside, and what time it was taken. The third idea is insight. Once data is collected, it can be studied to find patterns. A factory can learn that biscuits darken when the belt slows below a certain speed, or that milk spoils faster when cooling after pasteurisation takes too long. The fourth idea is action. Insight that does not change anything is only a report. Industry 4.0 closes the loop: a finding leads to a change in the process, and the change is then measured to see whether it worked. The change may be automatic, such as a controller adjusting a valve, or it may be a decision by a person, such as changing a cleaning schedule. Either way, the cycle of measure, understand, act, and measure again keeps turning.
The diagram shows the loop that runs through every unit in this subject. The steps point to the diagram: connection without recording is lost, recording without understanding is only storage, and understanding without action changes nothing.
Quick self-check
Question: A factory installs sensors everywhere but nobody looks at the data. Which part of the loop is missing?
Understanding and action. The connection and recording steps are done, but the data is not studied and no process change follows, so the factory is no better than before, only busier with data.
Question: Why does a temperature reading need a time and a batch number attached to it?
Because a number without context cannot be used. The time shows when the change happened, and the batch number shows which food was affected. Without those two facts, the reading cannot be linked to a cause or to a product.
What changes on the factory floor
For the people who work in the plant, Industry 4.0 changes daily work in practical ways. The operator at the mixer no longer walks to the oven to ask whether the temperature is steady: the oven temperature is on a screen at the mixer, with a chart showing the last hour. The laboratory no longer waits for a paper sample form to arrive with a batch: the sample is logged when it is taken, and the result is linked to the batch record as soon as it is entered. The maintenance team no longer waits for a pump to fail: a vibration sensor shows the pump shaking more than usual, and the team can plan a repair during a planned stop instead of during a production run.
A supervisor who once collected logbooks at the end of the day now looks at a live picture of the whole plant and can send help to the line that needs it most.
Food safety work changes too. Traceability, the ability to follow a food from raw material to finished pack and back again, becomes faster and more complete when each step is recorded digitally with its time and batch. If a raw material is later found to be a problem, the factory can list every finished batch that used it within minutes, instead of searching through paper records for a day. That speed matters when food must be held or recalled, and it matters for the trust of buyers and customers.
Quality work changes in a similar way. Instead of testing a few samples at the end of a run and hoping they represent the whole run, the factory can watch measurements continuously through the run. A filler that weighs every pack can find a drift in filling weight within minutes. A colour sensor on a biscuit line can find a change in baking within seconds. Problems are caught while they are small, and good batches are not mixed with bad ones.
Quick self-check
Question: How does a vibration sensor on a pump save production time?
It shows the pump wearing out before it fails. The repair can then be planned for a time when the line is already stopped, instead of the pump failing in the middle of a run and stopping production without warning.
Question: Why is continuous measurement better than end-of-run sampling for finding a filling problem?
Because end-of-run sampling only tests a few packs after the whole run is finished. Continuous weighing tests every pack as it is filled, so a drift is seen within minutes and only a small number of packs are affected.
What does not change
Some things in a food factory do not change, and it is worth naming them plainly, because new technology can make people forget them. Food still spoils, microbes still grow when conditions suit them, heat still kills microbes only when the right temperature is held for the right time, and cold still only slows growth, it does not stop it forever. A sensor cannot make a dirty line clean, and a computer cannot make a wrong recipe right. The science of food, the chemistry, the microbiology, and the engineering of heat, cold, flow, and mixing, stays at the centre of the work. Industry 4.0 gives better eyes and a better memory to the factory, but it does not change the food itself.
The need for honest records does not change either. A digital record can be wrong just as a paper record can be wrong, and a wrong sensor can record a wrong temperature with great confidence. Checking sensors against a known standard, checking that a reading makes sense, and keeping a habit of questioning surprising numbers are as necessary in a connected factory as they were in a paper one. Technology that is trusted without checking becomes a new way to make the same old mistakes, faster.
The need for clear responsibility does not change. When a machine adjusts itself, someone must still be responsible for the limits it works within, for the recipe it follows, and for the decision to release a batch. A screen that says a batch is good does not replace the person who signs for it. The factory needs to know who set the limits, who changed them last, and why, and that record is part of the system.
The building blocks, in plain words
The later units of this subject take each building block in turn. This section gives a short plain-language map, so that the names are familiar before they are studied in detail.
Sensors are devices that measure a physical quantity, such as temperature, pressure, flow, weight, colour, or moisture, and turn it into a signal a computer can read. They are the eyes and hands of the connected factory, and Unit 3 studies them.
Cyber-physical systems are machines that combine a physical part, such as a pump or an oven, with a computer part that senses, decides, and acts. A pasteuriser that measures its own temperature and adjusts its own heating is a cyber-physical system, and Unit 2 studies it.
The digital thread is the unbroken chain of information that follows a product from raw material through processing and packing to the customer. Each step adds its record to the same chain, so that the whole history of one batch can be followed in order. Unit 3 studies this too.
Cloud computing is the use of computers held elsewhere, reached through a network, to store and study data, instead of keeping all the computing inside the factory building. Big data is the name for the large volumes of data that connected factories produce, and for the methods used to study them. Unit 4 studies both.
A digital twin is a computer model of a factory, a line, or a single machine, kept up to date with live data, so that the factory can test a change in the model before making it in the real plant. Unit 5 studies digital twins.
The connected worker is the person in the factory who uses these tools: screens, mobile devices, and clear instructions that arrive where the work is done. Unit 6 studies how people and machines work together, and what skills the connected worker needs.
The road map in Unit 7 brings the pieces together into a practical order for a real food plant: what to do first, what to do later, and how to know whether each step is working.
The diagram shows how the blocks fit together. The steps point to the diagram: measurement feeds control, control feeds the record, the record feeds study, and study feeds the people who run the plant.
Quick self-check
Question: What is the difference between a sensor and a cyber-physical system?
A sensor only measures and reports. A cyber-physical system combines measurement with a computer and a physical machine, so it can sense, decide, and act, such as a pasteuriser that measures temperature and adjusts heating by itself.
Question: Where does a digital twin get its information?
From live data sent by sensors and machines in the real factory. Without that live link, the model is only a drawing. With it, the model stays in step with the real plant.
Why a food factory would want this
A food factory has four plain reasons to connect its work: less waste, steadier quality, faster answers, and safer food. Each reason is worth spelling out, because a factory that cannot name the reason will buy equipment without knowing what it should improve.
Less waste comes from seeing problems early. A filling machine that drifts out of adjustment wastes product in every pack it overfills, and it creates underweight packs that must be reworked or thrown away. A dryer that runs too hot wastes energy and damages the product. A cold store whose temperature rises for an hour can spoil a whole room of food. In each case, an early warning turns a large loss into a small correction. The saving is not in the sensor itself: it is in the food and energy that are not wasted after the warning.
Steadier quality comes from repeatability. Customers expect the same biscuit, the same sauce, the same yoghurt every time they buy it. When the conditions that produced a good batch are recorded in detail, the factory can repeat them, and when the conditions drift, the factory can bring them back before the product changes. Over many batches, the variation between batches shrinks, and the factory spends less time sorting good packs from bad ones.
Faster answers come from a single shared record. When a buyer asks which farm a batch of milk came from, when a laboratory reports a high count, or when a manager asks why Tuesday was slower than Monday, the answer is in one place, with times and batch numbers attached. The factory does not need to send people to search filing cabinets or to phone three departments. That speed is useful every day, and it is essential when food must be held back quickly.
What Industry 4.0 cannot do, and what it costs
A clear-eyed view of Industry 4.0 also names its limits. Connection does not fix a bad recipe, a worn machine, or a dirty line. If a mixer cannot blend evenly, a sensor will only record the uneven result more precisely. If a process is not understood, collecting more data about it will not by itself explain it. The factory still needs food science, engineering, and careful observation, and data is a tool for those, not a replacement.
There are real costs. Sensors, networks, computers, and software cost money to buy and to keep running. Sensors need calibration, networks need care, software needs updates, and records need safe storage. A factory that installs equipment and then leaves it unmaintained will soon have sensors that read wrong, screens that show old data, and staff who stop trusting the system. The cost of keeping the system true is part of the decision, not an afterthought.
A small food plant faces a different question from a large one. A large plant can afford a full team and a wide system. A small plant must choose a few measurements that matter most, such as the temperature of a pasteuriser, the weight of packs, or the temperature of a cold store, and do those well. Starting small is not a failure: it is the sensible way to learn what the data is worth before spending more. Unit 7 gives that small start a practical order.
Quick self-check
Question: A factory buys sensors but never calibrates them. What happens to trust in the system?
Trust falls away. The sensors slowly read wrong, the operators learn that the screen does not match what they can see and feel, and they stop using the screen. The system is then worse than useless, because it gives wrong numbers with the appearance of accuracy.
Question: Why does a small plant start with a few measurements instead of connecting everything at once?
Because each measurement has a cost in buying, installing, calibrating, and watching it. A few well-chosen measurements teach the plant what data is worth, and the plant can add more once the first ones are paying for themselves.
Try it yourself
Choose a food you know well, such as bread, yoghurt, jam, or bottled juice. Draw the path of that food through a factory on one page, from raw material to packed product, in five to eight steps. For each step, write one measurement that would tell you whether the step is working well, and write who would need to see that measurement and how quickly. Then mark the two measurements that matter most for safety and the two that matter most for quality. Compare your list with a classmate who chose a different food: which measurements are the same in both factories, and which are different, and why?
Key ideas to carry forward
- Industry 4.0 connects machines, sensors, computers, and people so that information flows continuously and decisions use what is actually happening on the line.
- It follows three earlier revolutions based on steam, electricity and the production line, and computers and automation, and it adds connection to those earlier gains.
- Four ideas hold it together: connection, information, insight, and action, in a loop that keeps turning.
- Operators, laboratories, maintenance teams, and managers all gain a shared, live picture of the plant, and routine watching takes less time while judging and fixing take more.
- Traceability and continuous measurement make food safety and quality answers faster and more complete.
- The science of food does not change: heat, cold, time, and cleanliness still decide safety, and a sensor cannot fix a bad recipe or a dirty line.
- Connection costs money and needs care: calibration, backup, protection, and a plan for working when the network is down.
- A small plant should start with a few measurements that matter most and add more once the first ones prove their worth.
- The building blocks are sensors, cyber-physical systems, the digital thread, cloud computing and big data, digital twins, and the connected worker, studied in turn in the units that follow.
Glossary
Automation
The use of machines and controllers to carry out a process without a person doing each step by hand. Food tech example: a filler that fills and seals jars by itself, following a set program, is automation.
Big data
Very large volumes of data, collected fast and from many sources, studied with methods that can find patterns across them. Food tech example: temperature, weight, and laboratory readings from a year of biscuit production form big data for that factory.
Cloud computing
Using computers held elsewhere, reached through a network, to store and study data, instead of keeping all the computing inside the factory. Food tech example: a bakery stores its batch records on a cloud system and reads them from the office and the laboratory.
Control limit
A set value that marks the edge of acceptable working, such as a highest or lowest temperature. Food tech example: a pasteuriser has a control limit for its lowest holding temperature, below which the milk must not be passed on.
Cyber-physical system
A machine that joins a physical part, such as a pump or an oven, with a computer that senses, decides, and acts. Food tech example: a pasteuriser that measures its own temperature and adjusts its own heating is a cyber-physical system.
Data
Recorded facts, such as numbers, times, and readings, that can be stored, searched, and compared. Food tech example: a list of oven temperatures taken every minute is data.
Digital thread
The unbroken chain of information that follows one product from raw material through processing and packing. Food tech example: the thread for one batch of jam links the fruit delivery, the cooking record, the filling record, and the pack code.
Digital twin
A computer model of a factory, line, or machine, kept up to date with live data, used to test changes before making them in the real plant. Food tech example: a model of a dryer lets engineers test a new temperature setting before changing the real dryer.
Industry 4.0
A way of running a factory in which machines, sensors, computers, and people share information continuously, so that changes are seen early and decisions use live data. Food tech example: a biscuit factory whose oven, conveyor, and packing line share readings on one screen is working in the Industry 4.0 way.
Insight
Understanding gained from studying data, such as a pattern that explains why quality changes. Food tech example: studying a year of records shows that biscuits darken when the conveyor slows, which is an insight.
Programmable logic controller
A small industrial computer that follows a set of instructions to control a machine, such as opening a valve or starting a mixer. Food tech example: a controller starts a dough mixer, runs it for a set time, and stops it when the dough reaches a set temperature.
Production line
A sequence of machines and work steps arranged in order, along which a product moves from raw material to finished pack. Food tech example: a bottling line washes, fills, caps, and labels bottles in order.
Sensor
A device that measures a physical quantity, such as temperature, pressure, flow, weight, or colour, and turns it into a signal a computer can read. Food tech example: a temperature sensor inside an oven sends its reading to a control screen every few seconds.
Traceability
The ability to follow a food from its raw materials through every step to the finished pack, and back again. Food tech example: a pack code lets a jam maker find the fruit delivery and the cooking batch that produced that jar.
Trend
The direction in which a measurement is moving over time, such as rising, falling, or staying steady. Food tech example: a chart showing oven temperature slowly falling across a morning shows a trend that warns of a problem.
References
- Singh, R. P., Computer Applications in Food Technology, 1996.
- Irudayaraj, J., Food Processing Operations Modeling, Design, and Analysis, 2002.
Unit 2: Cyber-Physical Systems and the Smart Factory
Note
Unit 2: Cyber-Physical Systems and the Smart Factory
A question to start with
A milk pasteuriser holds milk at a set temperature for a set time, then cools it fast. One morning the steam supply weakens for three minutes. In an old plant, the operator watches a dial, sees the temperature fall, and opens the steam valve by hand. If the operator is busy at the next machine, the milk passes through too cool, and the batch must be held, tested, and perhaps reprocessed. In a newer plant, a sensor reads the temperature every second, a controller sees the fall at once, opens the valve by itself, and writes the whole event into the batch record with the time. Nobody had to notice, nobody had to run, and the record shows exactly what happened. The machine did not become clever in the human sense. It became a system in which the physical part, the pipes, steam, and milk, and the computer part, the sensor, controller, and record, work as one thing.
That joined thing has a name: a cyber-physical system. This unit explains what it is, how its parts fit together, how it controls a food process, and what makes a whole factory built from such systems a smart factory.
Questions to think about
- A pasteuriser can hold its own temperature without a person turning a valve. What should the person still be responsible for?
- A filler counts every pack and stops when the count reaches the order size. Is that a cyber-physical system, or only a counting machine?
- If the computer part of a system fails, what should the physical part do: keep running, stop safely, or wait for a person?
Think about the pasteuriser question before you read on
The person stays responsible for the limits, the recipe, and the release of the batch. The system holds the temperature inside limits that a person set, using a method that a person approved. If the temperature falls outside those limits, the system should warn, hold the product, or stop, and a person decides what happens to the milk. Automatic control takes over routine adjustment: it does not take over responsibility for safety or for the decision to sell the food.
What a cyber-physical system is
A cyber-physical system joins two worlds. The physical world is the world of pipes, tanks, ovens, conveyors, steam, water, flour, milk, and heat. Things in this world have temperature, pressure, flow, weight, and speed, and they change over time. The cyber world is the world of measurement, calculation, and communication: sensors turn a physical quantity into a number, software compares that number with a target, and a signal goes back to a valve, a motor, or a heater to change the physical world. The loop runs again and again, many times a minute, and the two worlds stay in step with each other.
Three parts make the loop work. Sensing measures what is happening. Computing decides what should happen next. Actuation makes the change: a valve opens, a motor speeds up, a heater switches on, a diverter sends product to a holding tank. A system that only senses is a measuring device. A system that only computes is a model. A cyber-physical system does all three, and it does them while the process is running, not after it has finished.
A thermostat in a home oven is a simple example. A sensor measures the oven temperature, a controller compares it with the set temperature, and a heater switches on or off. A food plant version is larger and more exact: a pasteuriser measures temperature at the holding tube, compares it with the required holding temperature, and adjusts steam flow to hold it steady, while recording every reading with the time. The shape of the loop is the same; the scale, the accuracy, and the record are different.
Quick self-check
Question: A temperature display on a cold store wall shows the temperature but controls nothing. Is it a cyber-physical system?
No. It senses and shows, but it does not decide and it does not act. Add a controller that starts a compressor when the temperature rises and stops it when the temperature falls, and the joined system senses, decides, and acts.
Question: What are the three parts of the sensing, computing, and acting loop?
Sensing measures what is happening, computing compares the measurement with a target and decides what to do, and actuation makes the change in the physical process, such as opening a valve or starting a motor.
The control loop, step by step
Control in a food plant is usually feedback control. The word feedback means that the result of an action is measured and fed back to the controller, so the next action can be corrected. A driver steering a vehicle uses feedback: the eyes see the road, the hands turn the wheel, the eyes see the new position, and the hands correct again. A pasteuriser controller does the same with numbers.
The loop has a target, called a set point. For a pasteuriser, the set point may be a holding temperature. The sensor measures the actual temperature. The controller works out the difference between the set point and the actual temperature, called the error. If the temperature is below the set point, the error is positive in the heating direction, and the controller opens the steam valve more. If the temperature is above the set point, the controller closes the valve. The sensor measures again, and the loop repeats. When the loop runs fast and the adjustments are well chosen, the temperature stays close to the set point even when the milk flow or the steam supply changes.
A good controller does not overreact. If it opens the valve fully at the first small dip, the temperature will overshoot, then fall, then overshoot again, swinging up and down. If it reacts too slowly, the temperature will drift away before it is corrected. Tuning a controller means choosing how strongly it reacts to the present error, how it handles an error that has lasted a while, and how it reacts to a fast change. Food students do not need to do the mathematics of tuning here, but they need to know that tuning exists, that a badly tuned loop swings or drifts, and that the signs of bad tuning are visible on a chart: a steady wave, or a slow wander away from the set point.
Some control is feedforward rather than feedback. Feedforward acts before the error appears, using a measurement of a coming change. If a flow meter shows that milk flow is about to rise, the controller can open the steam valve a little before the temperature falls. Feedforward and feedback are often used together: feedforward handles the expected change, and feedback corrects whatever is left. A bakery that measures flour moisture before mixing and adjusts the added water is using feedforward thinking: the correction is made before the dough is wrong.
There is also simple on-off control, where a device is either fully on or fully off. A cold store compressor that switches on above a set temperature and off below it is on-off control. It is simple and it works where a small swing around the set point is acceptable. It is not enough where a tight, steady temperature is needed, because the temperature will always swing between the on point and the off point. Choosing between on-off control and smooth control is a practical decision about how steady the process must be.
The diagram shows the loop that repeats many times a minute. The steps point to the diagram: the loop never ends while the process runs, and each pass corrects the last one a little.
Quick self-check
Question: Why does a badly tuned controller make the temperature swing up and down?
Because it overreacts. A large correction for a small error pushes the temperature past the set point, so the next correction pushes it back past the set point the other way, and the swing repeats.
Question: Give a food example of feedforward control.
Measuring flour moisture before mixing and adjusting the water added to the dough. The correction uses a measurement of a coming change, before the dough itself shows an error.
The parts inside the system
A cyber-physical system in a food plant is built from parts that a student can learn to name. The sensor measures, as Unit 3 describes in detail. The controller is often a programmable logic controller, a small industrial computer built to survive heat, moisture, vibration, and long hours, and to follow its program reliably. The actuator is the device that makes the physical change: a control valve, a variable-speed motor drive, a heater, a pump, or a pneumatic cylinder that pushes a diverter. The human-machine interface is the screen where an operator sees the process, changes a set point, and answers an alarm. The network is the wiring or wireless link that carries signals between these parts. The record system stores the readings and the events.
Each part can fail, and a good design plans for failure. A sensor can drift, so the system checks it against a second sensor or against a known standard. A controller can lose power, so a backup power supply or a safe state is planned: valves that fail closed on a heating line, or a diverter that sends product to a holding tank when the signal is lost. A network can be interrupted, so a local controller keeps running its own loop even when it cannot reach the central record. These choices are called fail-safe design, and they matter more in food than in many industries, because a failure can send unsafe food forward if the system keeps running blindly.
Timing matters too. A control loop that reads a temperature once a minute cannot hold a fast process steady, while a loop that reads it ten times a second can. The right speed depends on the process: a cold store changes slowly, so a reading every minute is enough, while a pasteuriser holding tube passes milk in seconds, so its control must be fast. The student should ask, for any loop, how fast the process can change, and whether the measurement and the action are fast enough to keep up.
From one system to a smart factory
One pasteuriser that controls itself is a cyber-physical system. A factory in which the pasteuriser, the filler, the cold store, the laboratory, and the packing line all share data and adjust to each other is a smart factory. The difference is not the size of the machines: it is the way they work together.
In a smart factory, the systems are linked in layers. At the bottom are the sensors and actuators on the machines. Above them are the controllers that run the loops. Above the controllers are systems that watch the whole line: they show the state of every machine on one screen, they collect the records, and they pass orders and recipes down to the line. Above those are the planning and business systems that hold orders, stocks of raw material, and delivery plans. Information flows up as measurements and events, and instructions flow down as recipes, set points, and orders. When the flow works, a change in an order can reach the filler as a new pack size, and a fault on the filler can reach the planner as a revised delivery time.
This layered picture has a plain name in engineering: the automation pyramid. At the base is the process itself. The next layer is sensing and actuation. Then comes control, then supervision of the line, then planning, then business. A student does not need to memorise the layers, but the idea is useful: a smart factory is not one big computer, it is layers of systems, each doing its own job and passing clean information to the next layer.
The smart factory also changes how a line handles changeovers, the switch from one product to another. In an old plant, a changeover means stopping the line, changing parts, finding the right recipe card, and running test packs until the quality is right. In a smart factory, the recipe is sent to the machines as data, the set points change together, the right labels are called up, and the first packs are checked against the record of the last good run of that product. Changeovers become faster and more repeatable, which matters for a food factory that makes many products in small batches for different buyers.
Quick self-check
Question: What is the difference between a cyber-physical system and a smart factory?
A cyber-physical system is one joined machine and computer, such as a self-controlling pasteuriser. A smart factory is many such systems linked together, sharing data across the line and the business, so that the whole plant works as one connected system.
Question: Why does a local controller need to keep running when the network fails?
Because the process does not stop when the network does. If the controller stopped too, a pasteuriser could overheat or pass untreated milk. The local loop holds the process safe, and the central record catches up when the network returns.
How the factory sees itself: supervision and alarms
A smart factory needs a way for people to see what is happening. That is the job of supervision systems, shown on screens in the control room and at the line. A good screen shows the whole line at a glance: which machines are running, which are stopped, the key temperatures and flows, and any alarm that needs attention. The best screens are simple: the normal state is quiet, and a problem stands out by colour and position, not by a wall of flashing numbers.
Alarms deserve care. An alarm is a call for a person to act, and it should be used only when a person must act. A factory that raises an alarm for every small swing soon trains its operators to ignore alarms, and the one alarm that matters is then missed. Good practice is to set alarm limits wider than control limits, so that the controller has room to correct a small drift before a person is called, and to give each alarm a clear message: what is wrong, where it is, and what the operator should do first. An alarm that only says something is wrong, without saying where or what to do, wastes the time it was meant to save.
Records from supervision systems are the factory's memory. They show when a line stopped and for how long, which faults repeated, and which shifts ran best. A weekly review of these records often finds more improvement than a new machine would: a filler that stops every hour for the same small jam, a conveyor that slows after lunch each day, a cold store door that is opened far more often than the plan allows. The smart factory is not smart because it has screens: it is smart because someone studies what the screens recorded and acts on it.
Safety, interlocks, and the human in the loop
Food safety depends on some actions happening without fail. A pasteuriser must divert milk that has not reached the required temperature, even if the operator is busy. A mixer must not start while a person is cleaning inside it. A filler must stop if a guard is open. These are interlocks: hard rules built into the system that override normal control. An interlock is not a suggestion from software: it is a wired or programmed rule that acts even when the normal program is confused or a screen is frozen.
The human stays in the loop in three ways. First, people set the rules: the recipes, the limits, the alarm levels, and the interlocks. Second, people handle what the system cannot: a blocked pipe, a strange smell, a label that is wrong in a way no sensor checks. Third, people decide the things that need judgment: whether a batch that was held for a short temperature dip can be reprocessed, whether a supplier's material should be accepted, and whether a line should keep running after a repeated fault. A smart factory that tries to remove people from all three roles is not smart: it is blind to the things only people can see.
Cleaning and maintenance also belong in this picture. A cyber-physical system that runs a cleaning cycle can record that the cycle ran, how long each step lasted, and what temperatures were reached. That record is useful, but it does not prove that the line is clean: only inspection and testing prove that. The system should make cleaning easier to do correctly and harder to skip, and it should leave a record that can be checked, while the people doing the cleaning keep the judgment about whether the job is truly done.
Quick self-check
Question: What is an interlock, and why must it work even when the normal program fails?
An interlock is a hard rule that forces a safe action, such as diverting underheated milk or stopping a mixer while a person is inside it. It must work when the normal program fails because the moment of failure is exactly when the safe action matters most.
Question: Why should alarm limits be wider than control limits?
So the controller has room to correct a small drift by itself before calling a person. If every small drift raised an alarm, operators would be called too often and would learn to ignore alarms.
Working with models: how a controller knows the process
A controller that only reacts to errors can be good enough for a slow, simple process, but a fast or complex process needs more. A model-based controller uses a model of the process, a set of equations or rules that predict how the process will respond, to choose its actions. For example, a model of a dryer can predict how a change in air temperature will change the moisture of the product a few minutes later, so the controller can act now for an effect that will arrive later. A model of a mixing tank can predict how long blending will take to reach uniformity, so the mixer runs only as long as needed, not longer.
Models come from two places. Some are built from first principles: the physics of heat transfer, the chemistry of a reaction, or the mathematics of flow. Irudayaraj's book on food processing operations modelling shows how such models are built for food processes, using the same heat and mass transfer ideas that a food engineer learns in the core subjects. Other models are built from data: the factory records many runs, and a model is fitted to the pattern in those runs. Both kinds need checking against the real process, and both need updating when the process changes, such as a new recipe, a new raw material, or a worn machine.
A model is a simplification, and every simplification leaves something out. A good engineer knows what the model leaves out and checks whether it matters. A dryer model that ignores the moisture of the incoming air will be wrong on a humid day. A mixing model built for one batch size will be wrong for another. The discipline is the same as in the laboratory: a model is trusted only as far as it has been tested, and the test is done against the real process, not against another model.
Where cyber-physical systems go wrong
Systems fail in patterns, and knowing the patterns helps a student to diagnose them. The first pattern is a bad sensor: the loop is only as true as its measurement, and a sensor coated with product, a thermometer in a dead spot, or a flow meter with a blocked line will tell a confident lie. The sign is a reading that disagrees with other evidence, such as a second sensor, a laboratory test, or the look and feel of the product.
The second pattern is a wrong set point: the loop holds the process steady, but at the wrong value, because someone typed a wrong number or loaded the wrong recipe. The sign is a product that is steady but wrong in the same way in every batch. The fix is not in the controller: it is in the recipe handling and in the check that the right recipe is loaded.
The third pattern is a slow actuator: a valve that sticks, a motor that cannot reach the speed asked for, or a heater that has lost power. The sign is a loop that works for small changes but cannot keep up with large ones, or a temperature that drifts slowly away from the set point. The fix is maintenance, not more control.
The fourth pattern is a communication failure: the controller runs, but its readings never reach the record, or the recipe never reaches the controller. The sign is a gap in the record, or a machine running an old recipe after a change was ordered. The fix is in the network and in the check that the machine has received what it was sent.
The fifth pattern is the operator working around the system: a sensor bypassed because it alarms too often, a limit changed to make a run pass, or a manual mode left on after a repair. The sign is a system that looks fine on the screen while the product is wrong. The fix is management: bypasses and limit changes must be recorded, approved, and reversed, not hidden.
Quick self-check
Question: A temperature loop holds steady at the wrong temperature in every batch. Where would you look first?
At the set point and the recipe, not at the controller. A steady wrong temperature means the loop is working but it is aiming at the wrong target, which points to a wrong number or a wrong recipe loaded.
Question: Why is a sensor coated with product a serious problem for a control loop?
Because the loop trusts the sensor. A coated sensor reads a temperature that is not the true product temperature, so the controller corrects a false error and drives the real process away from its target.
Try it yourself
Pick one machine you know from a food plant or a kitchen: an oven, a fridge, a kettle, or a mixer. Draw its control loop on one page: the set point, the sensor, the controller, the actuator, and the process, with arrows showing the flow of information. Then answer four questions in writing: what happens if the sensor fails high, what happens if it fails low, what happens if the actuator sticks, and who should be warned first in each case? Compare your answers with a classmate who chose a different machine: which failures are dangerous in one machine and harmless in the other, and why?
Key ideas to carry forward
- A cyber-physical system joins a physical process with sensing, computing, and actuation, so the system measures, decides, and acts while the process runs.
- Feedback control compares a measurement with a set point, works out the difference, and corrects it, repeating the loop many times a minute.
- Feedforward control acts before an error appears, using a measurement of a coming change, and it works best together with feedback.
- Tuning decides how strongly a controller reacts: too strong and the process swings, too weak and it drifts.
- Controllers, sensors, actuators, screens, networks, and records form the parts of the system, and each part needs a plan for failure.
- A smart factory links many cyber-physical systems in layers, from the process up to planning, with information flowing up and instructions flowing down.
- Supervision screens should be quiet when all is well, and alarms should call a person only when a person must act.
- Interlocks force safe actions even when the normal program fails, and people keep the roles of setting rules, handling what the system cannot see, and judging what to do with a held batch.
- Models can help a controller predict a process, but a model is trusted only as far as it has been tested against the real process.
- Failures follow patterns: bad sensors, wrong set points, slow actuators, communication gaps, and workarounds, and each pattern has its own sign.
Glossary
Actuator
A device that makes a physical change in a process, such as opening a valve, speeding up a motor, or switching on a heater. Food tech example: a control valve that opens to let more steam into a pasteuriser is an actuator.
Alarm
A signal that calls a person to act because a measurement has passed a set limit. Food tech example: an alarm sounds when a cold store temperature rises above its set limit.
Automation pyramid
A layered picture of factory systems, from the process at the base, through sensing, control, and supervision, up to planning and business systems. Food tech example: a filler controller sits near the base of the pyramid, while the order planning system sits near the top.
Changeover
The switch of a line from one product to another, including changing parts, recipes, and labels. Food tech example: changing a jam line from strawberry jam to apricot jam is a changeover.
Cyber-physical system
A system that joins a physical process with a computer that senses, decides, and acts while the process runs. Food tech example: a pasteuriser that measures its own temperature and adjusts its own steam flow is a cyber-physical system.
Error
In control, the difference between the set point and the measured value. Food tech example: if the set point is 72 degrees and the milk is at 70 degrees, the error is 2 degrees.
Fail-safe
Designed so that a failure leads to a safe state, not a dangerous one. Food tech example: a steam valve that closes by itself when its signal is lost is fail-safe.
Feedback control
Control that measures the result of an action and uses that measurement to correct the next action. Food tech example: a pasteuriser controller that measures milk temperature and adjusts steam flow to hold it steady uses feedback control.
Feedforward control
Control that acts before an error appears, using a measurement of a coming change. Food tech example: adjusting dough water from a flour moisture reading before mixing is feedforward control.
Human-machine interface
The screen and controls through which a person sees a process and gives it instructions. Food tech example: the touch screen at a filling line where the operator sees fill weights and changes the set point.
Interlock
A hard rule that forces a safe action, overriding normal control. Food tech example: a rule that stops a mixer when its lid is opened is an interlock.
Model
A simplified description of a process, using equations or rules, that predicts how the process will respond to a change. Food tech example: a model of a dryer predicts how product moisture will fall when air temperature rises.
Programmable logic controller
A small industrial computer that follows a set of instructions to control a machine reliably in factory conditions. Food tech example: a controller that runs a dough mixer for a set time and stops it at a set temperature.
Set point
The target value that a controller aims to hold, such as a temperature or a flow rate. Food tech example: the set point for a pasteuriser holding temperature is the temperature the controller tries to keep.
Smart factory
A factory in which many cyber-physical systems share data across the line and the business, so the whole plant works as one connected system. Food tech example: a dairy in which the pasteuriser, filler, and cold store share readings and orders is a smart factory.
Supervision system
Software that shows the state of a whole line or plant on screens and collects its records. Food tech example: a control room screen showing every machine on a biscuit line and its key readings.
Tuning
Choosing how strongly a controller reacts to an error, so that the process stays steady without swinging or drifting. Food tech example: tuning a pasteuriser controller so its temperature holds steady when milk flow changes.
References
- Irudayaraj, J., Food Processing Operations Modeling, Design, and Analysis, 2002.
- Singh, R. P., Computer Applications in Food Technology, 1996.
Unit 3: Sensors, Data Collection and the Digital Thread
Note
Unit 3: Sensors, Data Collection and the Digital Thread
A question to start with
A yoghurt plant fills pots, seals them, and stores them in a cold room. One week, a buyer reports that some pots are sour and thin while others from the same day are fine. The laboratory tests a few pots and finds nothing wrong with them. The manager asks the packing team, who remember that the filling line stopped twice that day, and the cold room door was left open during a delivery. Nobody can say which pots were filled before the stops, which were filled after, or which pallets sat nearest the open door. The pots all look the same, and their codes do not tell the story. The plant has a quality problem and an information problem at the same time: the food changed, and the record cannot show when, where, or why.
This unit is about fixing that information problem. Sensors measure what is happening, data collection gathers those measurements together, and the digital thread links them into the story of one batch, from raw material to finished pack. Without sensors, the factory is blind. Without collection, it forgets. Without the thread, it cannot tell the story of one batch when a question is asked.
Questions to think about
- A cold room thermometer shows the air temperature, but the yoghurt in the middle of a pallet is warmer than the air. Which temperature matters for the food, and how would you measure it?
- A factory measures temperature every hour by hand. A sensor measures it every ten seconds. What problems can the sensor see that the hourly reading cannot?
- One batch of jam is made from fruit from three deliveries. How should the record show that, so that a question about one delivery can be answered?
Think about the yoghurt question before you read on
The air temperature is a useful warning, but the product temperature is what decides whether the yoghurt keeps well. Air changes fast when a door opens, while a pallet in the middle changes slowly. To know the product, the factory needs a sensor in or near the product, or a tested rule that links air temperature and time to product temperature. The deeper point is that a measurement is only useful if it measures the thing that matters, in the place where it matters, at the speed at which it changes.
What a sensor does, and what it does not do
A sensor turns a physical quantity into a signal. A thermometer turns temperature into an electrical signal. A pressure sensor turns force on a surface into a signal. A flow meter turns the movement of a liquid into a signal. A load cell turns weight into a signal. The signal is then read by a controller or a recorder, turned into a number with a unit, and stored with a time.
A sensor does not judge. It reports what it sees at its own position, in its own condition. A temperature sensor in a pocket of still air reads the still air, not the moving product beside it. A sensor coated with jam reads the jam stuck to it, not the jam flowing past. A sensor that has drifted reads a steady wrong number with complete confidence. The skill of using sensors is therefore not only in buying them: it is in placing them where they see the real process, keeping them clean and calibrated, and checking their readings against other evidence.
Every sensor has limits that a student should learn to ask about. Range is the span it can measure, from its lowest to its highest value. A sensor chosen for a cold store cannot measure an oven. Accuracy is how close its reading is to the true value. Precision is how repeatable its reading is when the same thing is measured again. Speed is how fast it responds to a change: a thick thermometer in a protective pocket responds slowly, while a thin one responds fast. A sensor can be precise but wrong, if it repeats the same wrong reading, and it can be accurate on average but noisy, jumping around the true value. The choice of a sensor is a choice among these qualities, and the right choice depends on the process: a pasteuriser needs a fast, accurate temperature sensor, while a silo level sensor can be slower and less exact.
Quick self-check
Question: A sensor reads the same value every time, but the value is 3 degrees above the true temperature. Is it accurate, precise, or both?
It is precise but not accurate. Precision means repeatability, and it repeats well. Accuracy means closeness to the true value, and it is 3 degrees away from it.
Question: Why does the position of a temperature sensor matter as much as its accuracy?
Because the sensor only reads its own position. A well-placed sensor reads the product, while a badly placed one reads still air, a wall, or a dead spot, and even a perfect sensor in the wrong place gives a wrong picture of the process.
The sensors a food plant uses most
Temperature sensors are the most common in food, because temperature decides safety and quality in cooking, pasteurisation, cooling, freezing, and storage. Resistance thermometers change their electrical resistance with temperature and are steady and accurate. Thermocouples join two different metals and produce a small voltage that changes with temperature; they are tough and fast, and they are used in hot places such as ovens and fryers. Infrared sensors read temperature without touching the food, by measuring the heat it gives off, which is useful for a moving product on a belt, though they need a clear view and a known surface.
Pressure sensors measure the pressure in a pipe, a tank, or a retort. In a retort, pressure and temperature together decide whether a can is safely processed, so the pressure reading is part of the safety record. In a pipe, pressure can show a blockage: a rising pressure upstream of a filter means the filter is clogging.
Flow meters measure how much liquid or gas passes a point, as a rate or as a total. They are used to measure milk into a tank, water into a dough mixer, and steam into a pasteuriser. Some measure by the spin of a rotor, some by the pressure drop across a restriction, and some by the effect of the moving fluid on a magnetic field or on sound waves. The choice depends on the food: a meter that works for clean water may clog on a thick sauce, and a meter for milk must be cleanable to food standards.
Level sensors measure how full a tank or silo is. They may use a float, a pressure reading at the bottom of the tank, or sound waves bounced off the surface. Level matters for planning, for preventing a tank from overflowing, and for knowing how much raw material is on hand.
Weight is measured by load cells under a tank, a hopper, or a packing machine. A checkweigher weighs every pack on a fast line and rejects packs that are too light or too heavy. A tank on load cells shows how much product it holds without opening it, and it can show the rate at which product is leaving the tank.
Other sensors measure the food itself. Moisture sensors measure water in flour, grain, or powder, often by the way the material responds to an electrical field. Colour sensors and cameras check the colour of biscuits, the browning of baked goods, and the fill level in a jar. Near-infrared sensors shine light at a product and read what comes back, giving a fast estimate of moisture, fat, or protein without a laboratory test. Gas sensors watch the air in a store or a pack for oxygen or carbon dioxide, which matters for modified atmosphere packing. Vibration sensors on a pump or a motor show wear before failure, as Unit 1 described.
Each of these sensors has a cleaning and hygiene side. A sensor that touches food must be made of food-safe material, must survive cleaning chemicals and hot water, and must not create a dead spot where food can sit and grow microbes. The best measurement in the world is useless if the sensor itself becomes a source of contamination.
Quick self-check
Question: Why would a plant use an infrared temperature sensor on a biscuit line instead of a probe?
Because the biscuits are moving on a belt and cannot be pierced one by one. An infrared sensor reads the surface temperature without touching the product, so it can check every biscuit as it passes.
Question: A checkweigher rejects packs that are too light. What two things does it need to do its job?
It needs to weigh each pack accurately while the line is moving, and it needs a way to remove a bad pack from the line without stopping the whole line.
From a reading to a record: data collection
A reading becomes useful when it is collected with three companions: when it was taken, where it was taken, and what it belongs to. A temperature of 72 degrees means little on its own. A temperature of 72 degrees, in pasteuriser 2, at 10:04, during batch 118, is a fact that can be used, checked, and traced. Data collection is the work of gathering readings with those companions attached, and storing them where they can be found again.
In a modern plant, readings travel from sensors to controllers, then to a data collection system, often called a historian, which stores them in time order. The historian is built for large volumes of time-stamped readings, and it can show a chart of any measurement over any period: the last hour, the last shift, or the last year. A laboratory system adds test results to the same story, linked by batch number and sample time. A planning system adds orders and deliveries. When these systems share batch numbers and times, the factory has one story instead of many separate ones.
Sampling rate is a practical choice in data collection. A reading every second gives a fine picture but a large store of data. A reading every minute gives a smaller store but can miss a short event, such as a three-minute dip in pasteuriser temperature. The right rate follows the speed of the process: fast processes need fast readings, slow processes can be read more slowly. A useful rule is to read at least several times during the shortest event that matters. If a three-minute dip matters, a reading every minute is the slowest rate that can still show it, and a reading every ten seconds is safer.
Missing data is also data. A gap in the record should be visible and explained: a sensor was being calibrated, a network was down, a line was stopped. A system that fills gaps with invented values, or hides them, destroys trust. A good record shows what was measured, when, and where the gaps are, so that anyone who uses it knows what it can and cannot prove.
The diagram shows how a single reading becomes part of a batch story. The steps point to the diagram: the value matters, but the time, place, and batch number are what make it usable.
Quick self-check
Question: A pasteuriser temperature dips for three minutes. How often must the plant record temperature to see the dip?
At least several times during those three minutes. A reading every minute would show it roughly, and a reading every ten seconds would show it clearly. A reading every hour would miss it entirely.
Question: Why should a gap in the record be shown, not hidden?
Because a hidden gap pretends the plant knows something it does not know. A visible gap tells the user where the record is silent, so nobody draws a conclusion from data that is not there.
Calibration: keeping the numbers true
Calibration is the comparison of a sensor's reading with a known standard, and the adjustment or correction that follows. A thermometer is checked in ice water and in boiling water, or against a reference thermometer whose own accuracy is known. A weighing machine is checked with known weights. A flow meter is checked by collecting the liquid it measured and weighing it. Without calibration, a sensor's readings drift slowly away from the truth, and nobody notices, because the drift is slow and the readings still look smooth.
A calibration plan names each sensor, how often it is checked, what standard is used, and what happens if it fails the check. Safety-critical sensors, such as the temperature sensor in a pasteuriser holding tube, are checked most often. The result of each check is recorded, so that the factory can show when a sensor was last known to be true. If a sensor is found to be wrong, the factory must decide which batches are affected: every batch since the last good check is in doubt, until the record shows otherwise. That is why short calibration intervals matter for safety sensors: a short interval limits the number of batches that fall into doubt.
Calibration is also a habit of mind. A surprising reading should be checked, not obeyed at once. If a temperature jumps suddenly, the first question is whether the sensor, its wire, or its position has changed, before the process is changed to chase the new number. A factory that checks surprising numbers keeps its process steady, while a factory that chases every jump wears out its valves and confuses its operators.
The digital thread: one story for one batch
The digital thread is the chain of information that follows one product from its start to its end. For a batch of jam, the thread begins with the fruit deliveries, including which fields or suppliers they came from and when they arrived. It continues with washing, cooking, and filling, each step adding its times, temperatures, and quantities. It ends with packing, labelling, storage, and despatch, with the pack codes that let one jar be traced back to its batch. At each step, the thread records what went in, what was done, what was measured, and what came out.
The thread works because each step uses the same identity for the batch. A batch number is given when the batch starts, and it travels with the product, on paper, on a screen, and on the pack. When a batch is split, such as one cooking batch filling two packing lines, the thread splits with it, and each part keeps the link to its parent. When two batches are mixed, such as fruit from three deliveries going into one cooking batch, the thread records all three parents, so that a question about one delivery can find every batch it entered.
The value of the thread appears when a question is asked. A buyer asks which deliveries went into one jar: the thread answers in minutes. A laboratory finds a high count in one batch: the thread shows which raw materials, which line, and which operators were involved, so the search for the cause starts in the right place. A raw material is found to be a problem: the thread lists every finished batch that used it, so only those batches are held, and good batches are not wasted. Without the thread, each of these questions takes a day of searching through separate records, and the answer is never certain.
Building the thread is less about buying a single system than about agreeing on identities and links. Every system in the factory, the receiving log, the process record, the laboratory, the packing line, and the despatch log, must use the same batch number, the same clock, and the same names for materials and steps. Where systems cannot talk to each other directly, a person or a small piece of software must carry the link across. The thread is therefore a management agreement as much as a technical one: it works only when every department records the same way.
Seeing patterns in the thread
Once the thread exists, it can be studied. A factory can compare batches that turned out well with batches that did not, and look for the difference: a longer cooking time, a warmer cold room, a different flour delivery, a slower line speed. It can look at a whole year of batches and find a slow drift that no single day showed, such as a dryer losing efficiency over months. It can look at stoppages and find the fault that repeats most often, and fix that fault first.
This study is sometimes done with simple charts, and sometimes with software that searches for patterns across many measurements at once. The method matters less than the discipline: a pattern is a suggestion, not a proof. A factory that finds a link between a measurement and a quality problem should test the link by changing the measurement in a controlled way and seeing whether the quality follows. If it does, the factory has found a cause. If it does not, the link was a coincidence, and chasing it would waste time. The thread makes such tests possible, because it holds the history that the test needs.
The thread also supports the laboratory. In the food analysis subject, sampling was shown to decide the value of the whole analysis. The thread tells the laboratory which batch a sample came from, when it was taken, and what the process was doing at that moment. A laboratory result that disagrees with the process record is a clue, not a nuisance: either the sample, the test, or the process reading is wrong, and the thread lets the factory find which one.
Networks: how the readings travel
Readings travel on networks, and a food student should know the plain facts about them. Inside a plant, sensors and controllers are often wired together, because a wire is reliable, fast, and safe from interference. Wireless links are used where a wire cannot go, such as a moving pallet, a rotating machine, or a sensor in a cold store, but wireless signals can be blocked by metal, water, and thick walls, and they need their own care. The network that carries control signals should be kept apart from the office network, so that office work cannot slow or disturb control, and so that a problem in one does not spread to the other.
Every device on the network needs a name and an address, so that its readings can be traced to the right place. Every reading needs a time from a clock that all devices share. If the pasteuriser clock and the laboratory clock disagree by ten minutes, the thread will link a laboratory result to the wrong part of the process record, and the factory will chase a false cause. Keeping one shared clock across the plant is a small job with a large return.
Security is part of the network too. A plant network that is open to the outside world can be reached by people who should not reach it, and a recipe or a control limit can be changed from far away. The plant should limit who can reach the control network, limit what each person can change, and keep a record of every change. These are the same habits as in the laboratory: only the right person, with the right reason, changes a record, and the change is written down.
Quick self-check
Question: Why must all the clocks in a plant agree?
Because the thread links records by time. If the pasteuriser clock and the laboratory clock disagree, a test result will be linked to the wrong part of the process record, and the factory will look for the cause in the wrong place.
Question: A pattern in the data links a measurement with a quality problem. What should the factory do next?
Test the link by changing the measurement in a controlled way and seeing whether the quality follows. A pattern is a suggestion, and only a test can show whether it is a cause.
Try it yourself
Choose a food with at least four steps from raw material to pack, such as bread, cheese, or canned beans. On one page, draw its digital thread: each step as a box, with the batch number travelling through them. Under each box, write the measurements that step should add to the thread, and the laboratory test, if any, that belongs to that step. Then write a question a buyer might ask about one pack, and trace the answer back through your thread, naming each record you would open. If your thread cannot answer the question, add the missing link and note what you added.
Key ideas to carry forward
- A sensor turns a physical quantity into a signal, and it only reports what it sees at its own position, in its own condition.
- Range, accuracy, precision, and speed describe what a sensor can do, and the right choice depends on the process.
- Temperature, pressure, flow, level, weight, moisture, colour, gas, and vibration sensors each serve a different measurement need in a food plant.
- A reading becomes useful when it carries a time, a place, and a batch number with it.
- Sampling rate should follow the speed of the process, so that the shortest event that matters is still seen.
- Calibration compares a sensor with a known standard, and the interval between checks limits how many batches fall into doubt when a sensor is found wrong.
- The digital thread is the unbroken chain of information for one batch, from raw materials through every step to the finished pack.
- The thread works only when every system uses the same batch number, the same clock, and the same names for materials and steps.
- Studying the thread finds patterns, but a pattern must be tested before it is treated as a cause.
- Plant networks should keep control traffic apart from office traffic, share one clock, and limit who can change a recipe or a limit.
Glossary
Accuracy
How close a reading is to the true value. Food tech example: a thermometer that reads 72 degrees when the true temperature is 72 degrees is accurate.
Batch number
An identity given to one batch, which travels with the product and links all its records. Food tech example: batch 118 links a pasteuriser record, a laboratory result, and a pack code for the same milk.
Calibration
Comparing a sensor's reading with a known standard and correcting it if needed. Food tech example: checking a thermometer in ice water and adjusting it if it does not read 0 degrees.
Checkweigher
A machine that weighs every pack on a moving line and removes packs that are too light or too heavy. Food tech example: a checkweigher on a biscuit line rejects packs below the target weight.
Digital thread
The unbroken chain of information that follows one product from raw material through processing and packing. Food tech example: the thread for one batch of jam links the fruit deliveries, the cooking record, and the pack code.
Flow meter
A device that measures how much liquid or gas passes a point, as a rate or as a total. Food tech example: a flow meter measures milk into a holding tank.
Historian
A system that stores large volumes of time-stamped readings in time order and can show charts for any period. Food tech example: a historian shows the pasteuriser temperature for the whole of last week.
Infrared sensor
A sensor that reads temperature without touching the product, by measuring the heat it gives off. Food tech example: an infrared sensor reads the surface temperature of biscuits moving on a belt.
Load cell
A device that turns weight into an electrical signal, used under tanks, hoppers, and packing machines. Food tech example: load cells under a mixing tank show how much dough is inside it.
Near-infrared sensor
A sensor that shines light at a product and reads what comes back, giving a fast estimate of moisture, fat, or protein. Food tech example: a near-infrared sensor estimates the moisture of flour as it enters a mixer.
Precision
How repeatable a reading is when the same thing is measured again. Food tech example: a balance that gives the same weight for the same pack ten times in a row is precise.
Range
The span a sensor can measure, from its lowest to its highest value. Food tech example: a cold store sensor with a range from minus 30 degrees to plus 30 degrees cannot measure an oven.
Sampling rate
How often a measurement is recorded, such as every second or every minute. Food tech example: a pasteuriser temperature recorded every ten seconds has a faster sampling rate than one recorded every hour.
Sensor
A device that measures a physical quantity and turns it into a signal a computer can read. Food tech example: a temperature sensor inside an oven sends its reading to a control screen.
Thermocouple
A temperature sensor made from two joined metals that produce a small voltage changing with temperature. Food tech example: a thermocouple in an oven reads high temperatures quickly.
Traceability
The ability to follow a food from its raw materials through every step to the finished pack, and back again. Food tech example: a pack code lets a jam maker find the fruit delivery behind one jar.
References
- Singh, R. P., Computer Applications in Food Technology, 1996.
- Irudayaraj, J., Food Processing Operations Modeling, Design, and Analysis, 2002.
Unit 4: Cloud Computing and Big Data in the Food Plant
Note
Unit 4: Cloud Computing and Big Data in the Food Plant
A question to start with
A food company runs four plants. Each plant keeps its own records: one on paper, one on a computer in the office, one on a computer beside the line, and one in a spreadsheet that only one person understands. The company wants to answer a simple question: which plant wastes the least energy per tonne of product, and what does it do differently? The answer should take an afternoon. It takes three weeks, because the records use different names for the same things, different units, and different batch numbers, and two of the files are on computers that nobody can open because the person who set them up has left. The company does not lack data. It lacks a shared place to put its data, a shared way to name it, and a shared way to study it.
This unit is about that shared place and that shared study. Cloud computing gives a factory computing power and storage that live elsewhere and are reached through a network. Big data is the name for the large volumes of information a connected factory produces, and for the methods used to store, clean, and study them. Together they let a factory, or a group of factories, see its own work as a whole, instead of as separate piles of records.
Questions to think about
- A factory stores its records on a single computer in the office. The computer fails and the records are lost. What should have been done differently, and when?
- Two plants both record oven temperature, but one records in degrees Celsius and the other in degrees Fahrenheit, and one names its ovens by number while the other names them by product. Can their data be compared? What must change first?
- A manager asks for a chart of energy use per tonne for the last year. The data exists, but it is spread across five systems. What is the hard part of answering: finding the data, or trusting it?
Think about the four plants question before you read on
The hard part is not the lack of data: it is that the data cannot be joined. Different names, different units, and different batch numbers mean that a temperature in one plant cannot be matched with a batch in another. The fix is agreement before storage: shared names, shared units, shared batch identities, and one place where the cleaned records live. Cloud storage can provide the shared place, but it cannot provide the agreement: that is a management decision, made by people, before the technology is chosen.
What cloud computing is, in plain words
Cloud computing means using computers, storage, and software that are held elsewhere and reached through a network, instead of owning and running all of the computing inside the factory. The factory's sensors and controllers still run on site, because control must keep working even if the network fails, but the long-term storage of records, the study of large volumes of data, and the sharing of information between plants and offices can run in the cloud.
Three features define the idea. First, the factory uses what it needs and can use more or less as its needs change, instead of buying a large computer that sits half empty. Second, the factory reaches its systems from wherever it is working: the office, the laboratory, another plant, or a manager's desk elsewhere. Third, the provider of the cloud looks after the hardware, the power, the cooling, and much of the upkeep, while the factory looks after its own data, its own users, and its own decisions about who may see what.
It helps to name the three common ways cloud services are offered. Infrastructure as a service provides raw computing and storage, like renting a bare computer room, and the factory installs its own software on it. Platform as a service provides a ready platform on which the factory's own programs can run, without looking after the underlying machines. Software as a service provides a finished program, reached through a web browser, such as a record system or a planning tool, where the factory uses the software but does not install it. A food factory may use all three at once, and the student does not need to choose between them: the point is to know what is being rented and what stays the factory's own job.
Quick self-check
Question: A factory moves its batch records to a cloud system. Does the pasteuriser controller move to the cloud too?
No. Control must keep running even if the network fails, so the controller stays on site, beside the process. The cloud holds the long-term records and the study of them, not the second-by-second control of the machine.
Question: What is the difference between software as a service and owning software on a factory computer?
With software as a service, the program runs elsewhere and is reached through a browser, and the provider looks after the machines and updates. With owned software on a factory computer, the factory looks after the machine, the installation, and the updates itself.
Why a food plant would use the cloud
A food plant has four plain reasons to use cloud computing, and each one should be weighed before the decision is made.
The first is storage that grows. A connected plant records temperatures, flows, weights, and laboratory results every few seconds, day after day. That volume soon outgrows a single office computer. Cloud storage can grow as the record grows, and old records stay available for study, instead of being deleted to make room.
The second is sharing between places. A company with several plants, a central laboratory, and a head office needs the same record in several places at once. A cloud system lets each place see the same record, with the same batch numbers and the same names, instead of passing copies around by email and wondering which copy is newest.
The third is study at scale. Finding a pattern across a year of batches, or across four plants, needs computing power that a small office computer does not have. Cloud computing can bring that power for the hours it is needed, and release it when the study is done, instead of the factory buying a large machine that is used once a month.
The fourth is care of the record. A cloud provider keeps copies of the data in more than one place, so that a fire, a flood, or a failed computer in one place does not destroy the only copy. That does not remove the factory's duty to keep its own safe copies and to know how to get its data back, but it makes the care of the record easier than keeping a single copy in one room.
These reasons are real, and so are the limits. A cloud system needs a network, and a plant with a poor connection will find its records slow to reach. A cloud system costs money every month, and the cost grows with the volume stored and the amount used. A cloud system holds the factory's data on someone else's computers, so the factory must be clear about who can see it, where it is held, and how it can be taken back. None of these limits is a reason to refuse the cloud: each one is a question to answer before choosing it.
What big data means in a food plant
Big data is often described with three words: volume, velocity, and variety. Volume is the amount: a connected plant produces far more readings in a day than a paper logbook ever held in a year. Velocity is the speed: readings arrive every few seconds, and some decisions need them at once, such as stopping a filler that is drifting. Variety is the range of forms: numbers from sensors, text from laboratory reports, pictures from cameras, and records from orders and deliveries, all describing the same batches in different ways.
A fourth word belongs beside them: veracity, which means truthfulness. Large volumes of data are useful only if they are true. A sensor that has drifted, a clock that is wrong, a batch number typed by hand with one digit wrong, or a laboratory result entered against the wrong sample all poison the study that follows. Cleaning data, finding and fixing such errors, is often the largest part of a big data job, and it is the least glamorous. A student who learns to clean data carefully, to check a surprising value before trusting it, and to keep a record of what was changed and why, has learned the most useful big data skill there is.
Big data work in a food plant follows a plain order. First, the data is collected from its sources: sensors, controllers, laboratory systems, and planning systems. Second, it is cleaned: errors are found, units are made the same, names are made the same, and gaps are marked. Third, it is stored in a form where it can be searched, often in a single store that brings the sources together. Fourth, it is studied: charts, comparisons, and pattern searches that answer a question the factory has asked. Fifth, the answer is put to work: a set point is changed, a maintenance job is planned, a supplier is questioned, or a recipe is adjusted, and the result is measured.
The diagram shows the order of big data work. The steps point to the diagram: cleaning comes before study, and study comes before action, because a conclusion drawn from dirty data is a wrong conclusion with a neat chart behind it.
Quick self-check
Question: Why is cleaning often the largest part of a big data job?
Because data arrives from many sources with different names, units, errors, and gaps. Each of these must be found and fixed before the data can be joined and studied, and the fixing takes more time than the chart that follows.
Question: What do volume, velocity, and variety mean for a food plant?
Volume is the large amount of readings, velocity is the speed at which they arrive, and variety is the range of forms, from sensor numbers to laboratory text to camera pictures, all describing the same batches.
Studying the data: from charts to prediction
The study of data starts with simple tools, and a food student should learn those first. A chart of one measurement over time shows a drift, a swing, or a sudden step. A chart of two measurements together shows whether they move together: does biscuit colour darken as oven temperature rises, does fill weight fall as line speed rises? A comparison of batches, grouped by shift, by line, or by supplier, shows which group runs best and which runs worst. These simple studies answer most factory questions, and they need no special software beyond a spreadsheet and a careful eye.
Averages and spreads are the next tools. An average says where a measurement sits, and a spread says how much it varies. A filling line whose average weight is correct but whose spread is wide will still produce many underweight and overweight packs. A line whose average is wrong but whose spread is narrow has a different problem: a set point that needs moving. Knowing which of the two problems is present decides the fix, and the fix for a wide spread is different from the fix for a wrong average.
More advanced study looks for patterns across many measurements at once. Software can search thousands of batches for the combination of conditions that came before a fault, or before a good run. It can build a model that predicts a quality result from the process conditions, so that the factory can see a problem coming before the product is made. Such predictions are useful only when they are tested: a model that predicts well on old batches may fail on new ones, if the process has changed. The factory should test a prediction on fresh batches, in real running, before trusting it to change a process.
One warning belongs here, and it is worth repeating: a link in the data is not proof of a cause. Ice cream sales and sunburn rise together in summer, but one does not cause the other: both follow the warm weather. In a food plant, a link between a measurement and a fault may mean that both follow a third thing, such as a change in raw material, a hot day, or a new operator. The study finds the link, and a controlled test decides whether it is a cause.
Dashboards: the factory's daily picture
A dashboard is a screen that shows the key numbers for a plant, a line, or a shift, in one place, updated as the work runs. A good dashboard answers the questions a manager asks every day: how much was made, how much was wasted, how long the line stopped, whether quality is in limits, and whether energy use is where it should be. A bad dashboard shows everything, and therefore answers nothing, because the eye cannot find the number that matters among fifty others.
Building a good dashboard starts with the questions, not with the data. The factory lists the five to ten numbers that decide whether the day went well, agrees how each one is calculated, and then shows those numbers clearly, with their targets beside them. A number without a target is only a fact: a number beside its target is a judgment. A chart that shows the number over the last days or weeks adds the trend: is the plant improving, steady, or slipping?
Dashboards also need honesty about freshness. A number labelled live should be live. A number that is updated once a day should say so, and should show the date it covers. A dashboard that shows an old number as if it were fresh misleads the people who use it to make decisions, and a factory that learns its dashboard cannot be trusted will stop looking at it. The rule from Unit 1 applies here too: the record must be true, timed, and complete, or it is worse than no record.
Keeping data safe and private
Food plant data includes recipes, supplier names, costs, and customer orders, and some of it is commercially sensitive. It also includes records that prove food was made safely, and those records must be kept whole and unaltered. Keeping data safe therefore has two sides: keeping it from being lost, and keeping it from being changed or seen by the wrong people.
Backup is the first side. A copy of the data is kept apart from the original, and the copy is tested by restoring it, not only by making it. A backup that has never been restored is a hope, not a protection. The factory should know how long a restore takes, and whether the plant can keep running while it happens.
Access is the second side. Each person should see the data they need and be able to change only what their job allows them to change. An operator can change a set point within limits, but should not be able to change a recipe or delete a batch record. A laboratory worker can enter a test result, but should not be able to change a result after it has been reported without a recorded reason and an approval. Every change to a recipe, a limit, or a reported result should carry the name of the person who made it and the time it was made.
The third side is the network itself. The control network should be kept apart from the office network and from the outside world, as Unit 3 described. Connections to the cloud should pass through a controlled path, with only the data that needs to travel being sent, and with the traffic protected. A factory that sends every reading to the cloud without asking which readings need to leave the plant has given away control of its own record.
Quick self-check
Question: Why is a backup that has never been tested by restoring it only a hope?
Because a backup can fail silently: the copy can be incomplete, the storage can be damaged, or the restore steps can be wrong. Only a real restore proves that the data can be brought back.
Question: Why should an operator be able to change a set point but not delete a batch record?
Because changing a set point within limits is part of running the process, while deleting a batch record destroys the proof of what was made. Different jobs need different powers, and the record must survive the people who use it.
Working across plants: comparing fairly
A company with more than one plant gains the most from shared data, and it also faces the hardest trap: comparing plants that are not alike and blaming the wrong one. Two plants may make the same product with different equipment, different raw materials, and different pack sizes. A direct comparison of their energy use per tonne, without allowing for those differences, will punish the plant with the harder job and reward the plant with the easier one.
Fair comparison needs shared definitions. Energy per tonne must be calculated the same way in each plant, over the same period, with the same products included. Waste must be defined the same way: does it include rework, does it include product given away as samples, does it include cleaning losses? A company that agrees these definitions before it compares will learn something. A company that compares first and defines later will argue about the numbers instead of learning from them.
Shared data also lets plants learn from each other. If one plant holds its pasteuriser temperature steadier than the others, the others can ask how: a different sensor position, a different tuning, a different steam supply, or a different routine. The data shows where to look, and the people show how it is done. A cloud system that only produces a ranking of plants, without the detail that explains the ranking, wastes the chance to learn.
What big data cannot do
Big data has limits that a student should name plainly. It cannot fix a process that is not understood. A factory that does not know why its product varies will not learn the reason from more data alone: it needs experiments, observation, and food science beside the data. It cannot replace a laboratory test where only a test can answer the question, such as whether a microbe is present. It cannot see what was never measured: if no sensor watched a step, the record is silent about that step, and no amount of study can fill the silence. And it cannot make a decision about values, such as whether to hold a doubtful batch or to release it: that judgment belongs to a person who bears the responsibility.
The cost side matters too. Storing every reading from every sensor forever is expensive, and much of that data will never be used. A sensible factory decides what to keep in full detail, what to keep as summaries, and what to discard, and it writes that decision down. A temperature record for a safety step is kept in full, because it proves the food was treated correctly. A vibration reading from a healthy pump may be kept as a daily summary, because the detail is needed only if a fault appears. Keeping everything is not careful: it is wasteful, and it makes the useful data harder to find.
Try it yourself
Take a small set of numbers you can collect yourself, such as the time it takes to boil a litre of water on three different days, with the starting temperature and the amount of water noted each time. Make a simple table, then a chart of the times. Write three sentences: what the average is, how much the times vary, and what could explain the variation. Then write one question that your table cannot answer, because the information was never recorded. That missing column is the lesson of this unit: a study can only answer the questions its data was collected to answer.
Key ideas to carry forward
- Cloud computing uses computers, storage, and software held elsewhere and reached through a network, while control stays on site.
- Infrastructure, platform, and software as a service are three ways cloud services are offered, and a factory may use all three.
- The cloud gives storage that grows, sharing between places, study at scale, and safer copies, and it needs a network, a monthly cost, and clear rules about who sees the data.
- Big data in a food plant means volume, velocity, variety, and veracity: much data, arriving fast, in many forms, and useful only if true.
- Big data work runs in order: collect, clean, store, study, and act, and cleaning is often the largest part.
- Simple charts, averages, and spreads answer most factory questions, and advanced prediction must be tested on fresh batches before it is trusted.
- A link in the data is not proof of a cause, and only a controlled test can show whether it is.
- A dashboard should show a few key numbers with their targets and trends, and it must be honest about how fresh each number is.
- Data safety means tested backups, access limited to each job, and a record of every change to a recipe, a limit, or a reported result.
- Comparing plants fairly needs shared definitions, and shared data is most useful when it shows how the best plant achieves its result.
- Big data cannot fix an ununderstood process, replace a laboratory test, see what was never measured, or make a judgment that belongs to a person.
Glossary
Average
The sum of a set of numbers divided by how many there are, showing where the set sits. Food tech example: the average fill weight of a hundred packs shows whether the filler is centred on its target.
Backup
A copy of data kept apart from the original, so that the data can be restored if the original is lost. Food tech example: a copy of a year's batch records kept in a separate place from the office computer.
Big data
Very large volumes of data, collected fast and from many sources, studied with methods that can find patterns across them. Food tech example: temperature, weight, and laboratory readings from a year of biscuit production form big data for that factory.
Cleaning data
Finding and fixing errors in data, making units and names the same, and marking gaps, before the data is studied. Food tech example: changing all oven temperatures to degrees Celsius and fixing batch numbers typed with one digit wrong.
Cloud computing
Using computers held elsewhere, reached through a network, to store and study data, instead of keeping all the computing inside the factory. Food tech example: a bakery stores its batch records on a cloud system and reads them from the office and the laboratory.
Dashboard
A screen showing the key numbers for a plant, line, or shift in one place, updated as the work runs. Food tech example: a dashboard shows packs made, waste, line stops, and quality results for the day.
Data store
A place where cleaned data from many sources is kept together in a searchable form. Food tech example: one store holding sensor readings, laboratory results, and orders for the same batches.
Infrastructure as a service
Cloud service that provides raw computing and storage, on which the factory installs its own software. Food tech example: renting cloud computers to run a factory record system.
Platform as a service
Cloud service that provides a ready platform on which the factory's own programs can run, without looking after the underlying machines. Food tech example: a platform where a factory's own batch tracking program runs.
Prediction
A forecast of a future result made from a model built on past data. Food tech example: a model that predicts biscuit colour from oven temperature and belt speed before the biscuits are baked.
Software as a service
Cloud service that provides a finished program, reached through a web browser, where the factory uses the software but does not install it. Food tech example: a planning tool used through a browser to schedule production.
Spread
How much a set of numbers varies around its average. Food tech example: the spread of pack weights shows whether a filler is steady or erratic.
Trend
The direction in which a measurement is moving over time, such as rising, falling, or staying steady. Food tech example: a chart showing energy use per tonne falling across a year shows a trend of improvement.
Variety
In big data, the range of forms data takes, from numbers to text to pictures. Food tech example: sensor numbers, laboratory reports, and camera pictures of the same batch show variety.
Velocity
In big data, the speed at which data arrives. Food tech example: temperature readings arriving every ten seconds have a high velocity.
Veracity
In big data, the truthfulness of data, whether it can be trusted. Food tech example: a drifted sensor gives data with poor veracity, even though the volume is large.
Volume
In big data, the amount of data produced. Food tech example: a year of readings taken every few seconds from a whole plant is a large volume.
References
- Singh, R. P., Computer Applications in Food Technology, 1996.
- Irudayaraj, J., Food Processing Operations Modeling, Design, and Analysis, 2002.
Unit 5: Digital Twins: a Virtual Copy of the Factory
Note
Unit 5: Digital Twins: a Virtual Copy of the Factory
A question to start with
A biscuit factory wants to run its ovens a little cooler and its belts a little faster, to save energy and make more biscuits in a day. The change sounds simple, but the manager is afraid to try it on the real line: if the biscuits come out pale or underbaked, a whole shift of product is wasted, and if the change damages the oven, the line stops for a repair. The engineers have a different idea. They build a computer model of the oven and the belt, fed with the same temperatures, speeds, and flour data as the real line, and they try the change in the model first. In the model, the biscuits come out pale at the new settings, so the engineers try a smaller change, and then another, until the model shows biscuits that are golden at a lower temperature. Only then do they try the best setting on the real line, and it works. The model did not bake a single biscuit, but it saved a shift of wasted product.
That working model has a name: a digital twin. This unit explains what a digital twin is, how it is built, what it can and cannot do, and how a food factory uses it to test changes safely before making them in the real plant.
Questions to think about
- A factory has a drawing of its dryer on a computer, but the drawing never receives live data. Is it a digital twin? What is missing?
- A model predicts that a new oven setting will bake biscuits well, but the real biscuits come out pale. What should the factory check first: the model, the sensor, or the flour?
- If a digital twin is wrong, is it better to have no twin at all, or to have a twin that is known to be wrong in a certain way?
Think about the biscuit question before you read on
A drawing that never receives live data is only a drawing, not a twin. A twin stays in step with the real plant through live data from sensors and machines, so that its state matches the real state at the same time. When the model and the real plant disagree, the first check is usually the data: a sensor that has drifted, a flour delivery that differs from the one the model was built with, or a setting that was typed wrongly. A twin that is known to be wrong in a certain way can still be useful, because its limits are known: a twin that is wrong without anyone knowing is the dangerous one.
What a digital twin is
A digital twin is a computer model of a real thing, a machine, a line, or a whole factory, that is kept up to date with live data from that thing, so that the model and the real thing stay in step. The twin has three parts. The real thing is the physical plant: the oven, the dryer, the pasteuriser, the line. The virtual thing is the model in the computer: equations, rules, and pictures that describe how the real thing behaves. The link is the flow of data between them: sensors send readings to the model, the model updates its state, and the model sends back predictions, warnings, and suggested settings.
The link is what separates a twin from an ordinary model or a simulation. An ordinary simulation is run once, with chosen inputs, and its result is studied. A twin runs continuously, fed by live data, and its result is the present state of the plant as the model understands it. When the real oven temperature rises, the twin's oven temperature rises. When a pump slows, the twin's pump slows. A person looking at the twin sees the plant as it is now, not as it was when the model was built.
Twins exist at different sizes. A twin of one machine, such as a dryer, models that machine alone. A twin of a line models the machines on the line and how they affect each other, such as an oven and the conveyor that feeds it. A twin of a whole plant models the flow of material from reception to despatch, including stores, lines, and utilities such as steam and cold water. A small factory will usually start with a twin of one machine or one problem, not of the whole plant, because a small twin can be built, tested, and used, while a whole-plant twin is a large job that needs much data and much care.
Quick self-check
Question: What three parts does a digital twin have?
The real thing, the physical plant; the virtual thing, the model in the computer; and the link, the flow of live data between them that keeps the model in step with the real plant.
Question: How is a digital twin different from a simulation run once?
A simulation is run once with chosen inputs. A twin runs continuously, fed by live data, so that it shows the present state of the plant and stays in step with it over time.
How a twin is built
Building a twin starts with a clear purpose, not with software. The factory names the question the twin must answer: can this dryer run cooler, what happens if the line speed rises, where does the bottleneck sit on this line, or how much steam will this change save? A twin built without a question tries to model everything, and it models nothing well. A twin built for one question can be kept simple enough to test and to trust.
The next step is to describe the real thing. Engineers list the machines, the pipes, the flows, and the steps of the process, and they write down the science that governs it: heat moves from hot to cold, water evaporates when heat is supplied, a liquid flows faster through a wider pipe, and a mixture becomes uniform only after enough mixing time. These are the same heat and mass transfer ideas that a food engineer learns in the core subjects, and Irudayaraj's book on food processing operations modelling shows how they are turned into models of real food processes. The description does not need to include every detail: it needs to include the details that decide the answer to the question.
The next step is to connect the data. Sensors and controllers send their readings to the twin, with the time attached, as Unit 3 described. The twin checks the data as it arrives: a reading that is outside the possible range, a sensor that has stopped sending, or a value that jumps without a cause is flagged, not silently accepted. A twin fed with bad data will give bad answers with great confidence, so the data check is part of the twin, not an extra.
The last step is to test the twin against the real plant. The twin is run beside the real process for a period, and its predictions are compared with what really happens. If the twin predicts the oven temperature well but predicts biscuit colour badly, the factory knows where the twin is weak, and it can improve the colour part of the model or accept the limit and use the twin only for temperature. Testing never ends: each time the process changes, a new recipe, a new raw material, or a worn machine, the twin must be checked again, because a twin built for yesterday's plant may not describe today's.
The diagram shows the building order. The steps point to the diagram: the question comes first, the science comes second, and testing comes before use, because an untested twin is only a guess in a neat form.
Quick self-check
Question: Why must a twin be built for one clear question, not for everything at once?
Because a twin that tries to model everything becomes too large to test and too slow to trust. A twin built for one question can be kept simple, tested against the real plant, and used with known limits.
Question: A twin predicts oven temperature well but biscuit colour badly. How should the factory use it?
It should use the twin for temperature questions and not for colour questions, until the colour part of the model is improved. Knowing where a twin is weak is part of using it well.
What a twin is used for
A twin has four plain uses in a food plant, and each one saves either product, time, or energy.
The first use is testing changes safely. A new temperature, a faster belt, a different recipe, or a changed cleaning time can be tried in the twin first, and the result can be studied without wasting product or stopping the line. Only the best few settings are then tried on the real plant. This is the biscuit example that opened the unit, and it is the most common use.
The second use is finding bottlenecks. A twin of a line can show which machine limits the output of the whole line: the filler that cannot keep up, the oven that needs a longer time, or the packing machine that stops most often. The factory can then spend its money on the true bottleneck, instead of speeding up a machine that was never the limit. Speeding up a machine that is not the bottleneck does not increase the output of the line: it only builds a pile of work in front of the true bottleneck.
The third use is training. An operator can learn on the twin: what happens if a valve is opened too far, if a line stops, or if a temperature drifts. Mistakes on the twin waste nothing and harm nobody, and the operator can repeat the lesson until it is learned. Training on the real line, by contrast, wastes product and risks safety, so it is kept short. A twin lets a new operator practise the unusual events, the faults and the alarms, that rarely happen on a real shift but matter most when they do.
The fourth use is watching for faults. A twin that knows how the plant should behave can compare its prediction with the real readings. When the real plant drifts away from the prediction, the difference is a warning: a sensor may be failing, a machine may be wearing, or a raw material may have changed. The warning arrives before the product is wrong, because the twin sees the drift while it is still small. This use needs a twin that is kept accurate: a twin that has drifted away from the real plant will raise false warnings, and operators will learn to ignore them.
Quick self-check
Question: Why does speeding up a machine that is not the bottleneck fail to raise line output?
Because the output of a line is set by its slowest step. Speeding up another step only makes product pile up in front of the slowest step, while the slowest step still sets the pace.
Question: How can a twin warn of a fault before the product is wrong?
By comparing its prediction of how the plant should behave with the real readings. A growing difference between the two shows that something in the real plant has changed, while the change is still small.
The science inside the twin
A twin of a food process is built from the same science as the process itself. Heat transfer decides how fast a product heats or cools: a thick piece heats more slowly than a thin one, and a product in moving air heats faster than one in still air. Mass transfer decides how water moves: moisture leaves a drying product at its surface, and it moves from the wet centre to the surface inside the product. Fluid flow decides how a liquid moves through a pipe: a thick sauce needs more pressure than water, and a narrow pipe limits the flow. Mixing decides how long it takes for a blend to become uniform: a powder added to a liquid needs time and movement to spread evenly.
A twin does not need to model every molecule. It uses simplified equations that capture the main effects, with numbers fitted from real runs of the process. For example, a dryer twin may use a simple rule that links air temperature, air flow, and inlet moisture to outlet moisture, with the rule fitted from many real drying runs. The rule is then tested on fresh runs, and its error is measured: how far, on average, is the prediction from the real outlet moisture? That error is the honest measure of the twin's quality, and it should be written down beside the twin, so that nobody trusts it beyond its error.
Some parts of a food process are harder to model than others. The chemistry of browning, the growth of microbes, and the texture of a dough depend on many factors at once, and a simple model may miss them. A twin should name these hard parts plainly and either model them with a tested rule or leave them out and say so. A twin that pretends to predict flavour or texture with no tested basis is not a twin: it is a guess with a picture.
Keeping the twin true
A twin goes stale. Machines wear, pipes scale, sensors drift, recipes change, and raw materials vary with the season. Each of these changes moves the real plant away from the model that was built for it. A twin that is not updated becomes a picture of a plant that no longer exists, and its predictions become steadily worse. The factory therefore needs a routine for keeping the twin true: regular checks against the real plant, updates when the process changes, and a record of when the twin was last tested and how well it predicted.
The routine is simple to state and hard to keep. After any change to the process, a new recipe, a replaced machine part, or a changed supplier, the twin is run beside the real plant and its predictions are compared with reality. If the error has grown, the model is adjusted or rebuilt. If the error is small, the twin is confirmed for another period. The result of each check is recorded, with the date and the error, so that the factory can see whether the twin is staying true or slowly drifting.
People also matter here. A twin needs an owner: a named person who looks after it, who knows its limits, and who decides when it needs updating. Without an owner, a twin is built with enthusiasm, used for a few months, and then quietly forgotten, while the plant changes around it. With an owner, the twin stays a living tool, and the factory keeps the benefit it was built for.
Limits, costs, and honest use
A twin costs money and time to build, and the cost must be weighed against the benefit. Building a twin of a whole plant is a large job, and it is rarely the right first step. Building a twin of one machine for one question is a smaller job, and its benefit can be measured: product saved, energy saved, or time saved in changeovers. A factory should start with the small twin, measure the benefit, and then decide whether a larger twin is worth building.
A twin can also mislead. A model that is trusted beyond its tested error will lead the factory to make a change that fails on the real line. A twin that is fed with bad data will give confident wrong answers. A twin that is used to justify a decision already made, instead of to test it, is not a tool but a prop. The honest use of a twin is to ask it a question, listen to its answer, and then test the answer on the real plant in a small, careful way, before changing the whole production.
The twin does not replace the operator, the engineer, or the laboratory. It gives them a safe place to test ideas, a clear picture of the plant as it is now, and an early warning when the plant drifts from its expected behaviour. The judgment about whether to change a process, whether to release a batch, and whether a product is good enough still belongs to people, using the twin as one piece of evidence among others.
Quick self-check
Question: Why does a twin need a named owner?
Because a twin goes stale as the plant changes, and it needs someone who checks it, updates it, and knows its limits. Without an owner, the twin is forgotten while the plant changes around it.
Question: A twin is used to justify a decision that was already made, instead of to test it. What is wrong with that?
The twin is being used as a prop, not a tool. Its purpose is to test ideas honestly, including the possibility that the idea is wrong, and using it only to confirm a fixed decision wastes that purpose.
Try it yourself
Choose a simple process you know well, such as drying clothes, boiling rice, or baking a cake. Write the question a twin of that process could answer, such as how long the rice needs at a lower heat. List the three measurements the twin would need from the real process, the science it would use, and one way you would test whether the twin is right. Then write one question the twin could not answer, and explain why. Bring your page to your next session and compare it with a classmate: which twins are easy to test, and which are hard, and why?
Key ideas to carry forward
- A digital twin is a computer model of a real machine, line, or factory, kept in step with the real thing by live data.
- The twin has three parts: the real thing, the virtual model, and the link of live data between them.
- A twin differs from a one-off simulation because it runs continuously and shows the present state of the plant.
- Building a twin starts with one clear question, then a description of the process using its science, then live data, then testing.
- Twins test changes safely, find bottlenecks, train operators, and warn of faults by comparing prediction with reality.
- The science inside a twin is heat transfer, mass transfer, fluid flow, and mixing, simplified into rules fitted from real runs.
- A twin is judged by its error: how far its predictions sit from real results, and that error should be written down.
- Twins go stale as machines wear and processes change, so they need regular checks, updates, and a named owner.
- A twin costs time and money to build, so a factory should start with one machine and one question and measure the benefit.
- A twin does not replace operators, engineers, or the laboratory: it gives them a safe place to test ideas and an early warning when the plant drifts.
Glossary
Bottleneck
The slowest step in a line, which sets the output of the whole line. Food tech example: a packing machine that can only pack half as fast as the filler is the bottleneck of that line.
Digital twin
A computer model of a factory, line, or machine, kept up to date with live data, used to test changes before making them in the real plant. Food tech example: a model of a dryer lets engineers test a new temperature setting before changing the real dryer.
Error
In modelling, how far a prediction sits from the real result. Food tech example: a dryer twin that predicts outlet moisture 1 percent away from the real moisture has an error of 1 percent.
Heat transfer
The movement of heat from a hot place to a cold place, by conduction, convection, or radiation. Food tech example: heat moves from hot air into a biscuit in an oven by convection.
Mass transfer
The movement of a substance, such as water, from one place to another, for example from the centre of a product to its surface during drying. Food tech example: water moves out of a drying slice of fruit to the air around it.
Model
A simplified description of a process, using equations or rules, that predicts how the process will respond to a change. Food tech example: a model of a dryer predicts how product moisture will fall when air temperature rises.
Prediction
A forecast of a future result made from a model built on past data. Food tech example: a model that predicts biscuit colour from oven temperature and belt speed before the biscuits are baked.
Simulation
Running a model with chosen inputs to see what result it gives, without a live link to the real process. Food tech example: running a dryer model with a new air temperature to see the predicted outlet moisture.
Virtual copy
A computer representation of a real thing, kept in step with it by data. Food tech example: a virtual copy of an oven shows its temperature and belt speed as they are now.
Fluid flow
The movement of a liquid or gas through a pipe or a machine. Food tech example: sauce flowing through a pipe to a filler is fluid flow.
Mixing
The movement that spreads ingredients evenly through a product. Food tech example: a mixer blends flour, water, and yeast until the dough is uniform.
Live data
Readings sent from sensors and machines as they happen, with the time attached. Food tech example: oven temperatures sent every ten seconds are live data.
Validation
Testing a model against the real process to see whether its predictions match reality. Food tech example: comparing a dryer twin's predicted moisture with laboratory results for the same batches is validation.
Training
Learning how to run a process, including its faults and alarms, by practising on a model before working on the real line. Food tech example: an operator practises a pasteuriser fault on a twin before facing it on the real plant.
References
- Irudayaraj, J., Food Processing Operations Modeling, Design, and Analysis, 2002.
- Singh, R. P., Computer Applications in Food Technology, 1996.
Unit 6: People and Machines: the Connected Worker
Note
Unit 6: People and Machines: the Connected Worker
A question to start with
A new operator starts on a jam line. On her first day, she is given a thick paper manual, a clipboard, and a tour of the line. By lunchtime she has forgotten half of what she was told, she cannot find the page that explains the filler fault light, and she is afraid to ask the same question twice. On the next line, another new operator starts the same day, but her training is different. She carries a small tablet that shows each job as a short list of steps, with a picture of the machine at each step, and a video of the last person who did the job well. When the filler fault light comes on, the tablet shows what the light means and what to check first. When she finishes a job, she taps to record it, and the record goes straight into the batch file. The two operators have the same job, the same line, and the same ability. One works with paper that cannot answer back, and the other works with information that arrives where the work is done.
This unit is about the second operator: the connected worker. It explains what changes when people in a food factory carry information with them, how machines and people share work, what skills the new way of working needs, and how a factory keeps its people safe, skilled, and in charge.
Questions to think about
- A tablet shows an operator the steps for cleaning a filler. The operator has cleaned that filler a hundred times and knows a faster way. Should she follow the tablet, follow her own way, or tell someone about the difference?
- A machine stops and shows an error code. The operator can look up the code on a screen, or she can call the maintenance team. When is each choice the right one?
- A factory gives every worker a tablet. One older worker says he prefers paper and his hands. What should the factory do: insist, offer training, or leave him on paper?
Think about the new operator question before you read on
The connected worker is not a person who obeys a screen without thinking. She is a person whose screen gives her the current, approved steps, the record of the last good run, and the fastest way to get help, while she brings judgment, senses, and experience that no screen has. If her faster way is genuinely better and safe, the right act is to tell someone, so that the approved steps can be checked and changed for everyone. If she quietly follows her own way, the factory loses the improvement and the record becomes false, because the record says the approved steps were followed when they were not.
What a connected worker is
A connected worker is a person in the factory who uses digital tools as part of daily work: a screen at the machine, a tablet or a mobile device carried on the line, and a clear flow of information to and from the systems described in the earlier units. The connection works in two directions. Information flows to the worker: the job to do, the recipe, the steps, the set points, the warnings, and the help. Information flows from the worker: the job done, the check made, the fault found, the sample taken, and the observation that a machine sounds wrong.
The point is not the device. A tablet in a pocket does not make a worker connected. The point is that the right information arrives at the right place, at the right time, in a form the worker can use while working. A recipe that lives in an office binder does not help the person at the mixer. A recipe that appears on the mixer screen, with the amounts for this batch size already worked out, does help. A warning that arrives after the shift ends does not help. A warning that appears on the line, while the batch can still be saved, does help.
The connected worker also stays connected to other people. An operator can send a photograph of a fault to the maintenance team, instead of describing it in words. A laboratory worker can send a result straight to the line, instead of walking a paper slip across the plant. A supervisor can see which lines need help and send a person there, instead of discovering the problem at the end of the day. The technology supports these human links: it does not replace them.
Quick self-check
Question: What flows in each direction for a connected worker?
To the worker flow the job, the recipe, the steps, the set points, the warnings, and the help. From the worker flow the record of the job done, the checks made, the faults found, the samples taken, and the observations that only a person on the line can make.
Question: Why is a tablet alone not enough to make a worker connected?
Because connection is about the information, not the device. A tablet that shows old, wrong, or unclear information does not help, while the right information at the right time, in a usable form, is what makes the worker connected.
Digital work instructions: the job, step by step
The most common tool of the connected worker is the digital work instruction: a job shown as a short list of steps on a screen, with a picture or a short video for each step that needs one, and a check at the points that matter. A cleaning instruction, for example, lists the steps in order, shows which valves to close, shows where to check that the cleaning solution has reached every part, and asks the worker to confirm each check with a tap. A changeover instruction lists the parts to change, the recipe to load, and the first packs to check.
Good instructions have a plain style. Each step is one action, written in short words. A picture shows the part, the valve, or the dial, so the worker does not have to guess which one the words mean. A warning appears only where the step is dangerous or where a mistake is costly, so that warnings keep their force. The instruction shows the current version, with its date, so that nobody follows an old copy from a drawer. When the instruction changes, the old version disappears from the screens, instead of sitting beside the new one and causing doubt.
Digital instructions also make the record. When a worker confirms a check, the confirmation is stored with the time, the worker's name, and the batch, and it joins the digital thread from Unit 3. When a step is skipped, the system shows the skip, instead of hiding it. That honesty is useful: a skipped check that is visible can be discussed and fixed, while a skipped check that is hidden on paper looks like a completed job.
Instructions should be written with the people who do the job, not only by the people who design it. A worker who does a cleaning job every day knows where the instruction is unclear, where a step takes longer than the plan allows, and where a photograph would help more than a paragraph. Writing the instruction together makes it truer, and it makes the worker more willing to follow it, because it is their instruction as well as the factory's.
Quick self-check
Question: Why should a warning appear only where a step is dangerous or a mistake is costly?
Because warnings lose their force if they appear everywhere. A worker who sees a warning on every step will stop reading them, and the one warning that matters will be missed.
Question: Why is a visible skipped check better than a hidden one?
Because a visible skip can be discussed and fixed, while a hidden skip looks like a completed job. The record must show what really happened, not what should have happened.
Sharing the work between people and machines
People and machines are good at different things, and a smart factory shares the work to suit each. Machines are good at repeating the same action exactly, at working fast without tiring, at measuring precisely, and at handling heavy or hot work. People are good at noticing what is unusual, at judging a smell, a colour, or a texture, at fixing a fault that has never happened before, and at deciding what to do when the rules do not cover the case.
The division of work should follow that difference. A machine should fill, weigh, seal, and check every pack, because it can do those jobs steadily all day. A person should watch the pattern, check the first packs of a run, taste and judge the product, and step in when the pattern changes. A machine should lift heavy pallets, because it does not strain its back. A person should decide whether a batch that was held for a short fault can be reprocessed, because that decision needs judgment about the food, the risk, and the customer.
Automation changes jobs rather than removing the need for people. When a filler becomes automatic, the operator's job moves from turning knobs to watching trends, checking quality, and fixing faults. That new job needs more knowledge, not less: the operator must understand the process, read a chart, and know when to call for help. A factory that automates without training its people for the new job will have machines that nobody understands and faults that nobody can fix.
Robots and collaborative machines belong in this picture, and the word collaborative matters. A traditional industrial robot works behind a fence, moving fast and carrying heavy loads, and a person must not enter its space while it runs. A collaborative machine, often called a cobot, is designed to work beside a person, moving more slowly, stopping when it touches something, and sharing a task such as lifting a box or holding a part. In a food plant, such machines may load trays, stack boxes, or move pallets, while the person does the judging and the fine work beside them. Safety rules for these machines are strict, and the factory must follow them: the machine must stop safely on contact, and the person must know how to work with it.
The diagram shows the division of work. The steps point to the diagram: the machine and the person each do what they are best at, and the record holds the result of both.
Quick self-check
Question: A filler becomes automatic. What does the operator's job become?
It moves from turning knobs to watching trends, checking quality, and fixing faults. The new job needs more knowledge of the process, not less, because the operator must know when the pattern is wrong and what to do about it.
Question: What makes a collaborative machine different from a traditional industrial robot?
It is designed to work beside a person, moving more slowly and stopping safely on contact, while a traditional robot works behind a fence, moving fast, and a person must not enter its space while it runs.
Skills for the connected worker
The connected worker needs skills that a food student can start learning now. The first is reading data: looking at a chart and seeing a trend, a swing, or a step, and knowing whether it matters. The second is recording honestly: entering a result, a check, or a fault in the system at the time it happens, with the right batch and the right time, so that the record is true. The third is following a digital instruction and knowing when to stop and ask: a step that does not match the real machine, a reading that does not make sense, or a warning that is unclear are all reasons to stop and check, not to guess.
The fourth skill is basic care of the tools: keeping a tablet charged, keeping a screen clean, reporting a sensor that looks damaged, and never sharing a password. The fifth skill is communication: describing a fault clearly, with a photograph and the time it happened, so that the maintenance team can fix it without a long search. The sixth skill is food knowledge itself: the microbiology, chemistry, and engineering that let a worker judge whether a product is safe and good, and whether a number on a screen can be true. Without that knowledge, a worker can follow a screen into a mistake, because the screen cannot judge the food.
Training for these skills should be short, practical, and repeated. A worker learns to read a chart by reading real charts from their own line, not by studying charts from a textbook. A worker learns to record honestly by recording real jobs, with feedback on the first ones. A worker learns to describe a fault by describing real faults, with a photograph, until the description is clear. Training that is done once, in a classroom, and never repeated will fade, while training that is part of daily work will stay.
Safety, health, and the human side
Connection brings new questions about safety and health, and a factory must answer them plainly. Screens and tablets must not distract a worker from the machine in front of them. A worker carrying a tablet across a wet floor, or looking at a screen while a forklift passes, is in danger. The factory should set clear rules: where devices may be used, where they must be put away, and when the machine comes first. A device that is used while walking is a risk, not a help.
Ergonomics, the fit between the work and the body, matters too. A screen placed too high or too low, a tablet held for a whole shift, or a job that now asks a worker to stand and watch a screen for hours will cause strain. The factory should place screens at a comfortable height, give workers a place to sit for jobs that need watching, and change tasks through the shift so that no one posture lasts too long.
There is also a question of trust and privacy. A connected factory records when a worker did a job, how long it took, and what they reported. That record is needed for traceability and for food safety, and the factory should be open about what it records and why. It should not use the record to watch workers in a way that has nothing to do with the food, and it should not punish a worker for reporting a fault or a mistake honestly. A factory where people are afraid to report a fault will learn about its faults late, from customers, instead of early, from its own people. The safest factory is the one where reporting a problem is praised, not punished.
Wearable devices, such as a wrist device that shows a warning or a headset that gives instructions, may appear in some plants. Their use should follow the same rules: they must help the work, they must not distract from safety, and the worker must know what is recorded and why. A device that monitors a worker's body without a clear safety reason does not belong in a food plant, and a factory should not introduce it without a plain explanation and the workers' understanding.
Quick self-check
Question: Why should a factory praise a worker who reports a fault honestly?
Because the factory learns about its faults early, from its own people, instead of late, from customers. A worker who fears punishment will hide the next fault, and the factory will lose its earliest warning.
Question: A screen is placed so high that a worker must look up all shift. What is wrong, and what should change?
The screen does not fit the body, and it will cause strain. It should be moved to a comfortable height, or the job should be changed so that watching the screen is not a whole-shift posture.
Keeping people in charge
The last idea in this unit is the most important: people stay in charge. Machines and software give information, make suggestions, and carry out routine actions, but the decisions that matter, whether food is safe to sell, whether a batch should be held, whether a recipe should change, belong to people who understand the food and bear the responsibility. A screen that says a batch is good does not release it: a person releases it, using the screen as evidence.
This matters most when the system and the person disagree. A sensor may say a temperature is fine, while the product looks, smells, or feels wrong. In that case, the person is right to stop and check, and the factory should support that decision, even if the check shows the sensor was right after all. A factory that punishes a worker for stopping the line on a genuine doubt will soon have workers who let doubtful food pass, because stopping is punished and passing is not. The rule should be written down and repeated: when in doubt about safety, stop, check, and ask, and the person who stopped will be thanked, not blamed.
The connected worker is therefore not a smaller role than the old operator. It is a larger one: more information, more judgment, more responsibility, and more need for food knowledge. A factory that invests in machines without investing in its people will have clever machines and confused workers. A factory that invests in both will have machines that work well and people who know why they work, what to do when they do not, and when to trust their own senses over a screen.
Quick self-check
Question: A sensor says the temperature is fine, but the product smells wrong. What should the operator do?
Stop and check, and ask for help. The senses are evidence too, and a sensor can be wrong. When in doubt about safety, the safe act is to stop and check, not to let the food pass because a screen said so.
Try it yourself
Think of a job you do often, such as cooking a meal or cleaning a room. Write it as a digital work instruction in six steps, with one picture described in words for the step that is hardest to explain, and one check that must be confirmed before moving on. Then give your instruction to a friend or classmate and ask them to follow it exactly, without asking you any questions. Watch where they hesitate or go wrong, and rewrite those steps. That hesitation is the lesson: an instruction is good only when another person can follow it without you beside them.
Key ideas to carry forward
- A connected worker receives the right information where the work is done, and sends back the record of the job, the checks, and the faults.
- Connection is about the information, not the device: the right information at the right time, in a usable form, is what matters.
- Digital work instructions show one action per step, with a picture where words are unclear, and they make the record as the work is done.
- Instructions should be written with the people who do the job, and a skipped check that is visible can be fixed, while a hidden skip looks like a completed job.
- Machines are best at repeating, measuring, and lifting, while people are best at noticing, judging, fixing new faults, and deciding doubtful cases.
- Automation changes jobs toward watching trends, checking quality, and fixing faults, and it needs more knowledge, not less.
- Collaborative machines work beside people and stop safely on contact, while traditional robots work behind a fence.
- Connected workers need to read charts, record honestly, know when to stop and ask, care for their tools, describe faults clearly, and know their food science.
- Screens and devices must not distract from safety, and the work must fit the body, with screens at a comfortable height and tasks changed through the shift.
- Records of work are for traceability and safety, and reporting a fault honestly should be praised, not punished.
- People stay in charge: a screen gives evidence, a person releases the batch, and when in doubt about safety, the safe act is to stop, check, and ask.
Glossary
Collaborative machine
A machine designed to work beside a person, moving slowly and stopping safely on contact. Food tech example: a machine that lifts boxes beside a worker on a packing line is a collaborative machine.
Connected worker
A person in the factory who uses digital tools as part of daily work, receiving information where the work is done and sending back the record of the job. Food tech example: an operator with a tablet showing the cleaning steps for a filler is a connected worker.
Digital work instruction
A job shown as a short list of steps on a screen, with pictures where needed and checks at the points that matter. Food tech example: the steps for changing over a jam line, shown on the line screen.
Ergonomics
The fit between work and the body, so that tasks, screens, and tools do not cause strain. Food tech example: a control screen placed at eye height, so the operator does not look up all shift, is good ergonomics.
Fault
Something wrong with a machine or a process, such as a jam, a leak, or a wrong reading. Food tech example: a filler that stops every hour with the same jam has a repeating fault.
Judgment
A decision made from knowledge, experience, and the evidence at hand, where the rules do not cover the case exactly. Food tech example: deciding whether a held batch can be reprocessed needs judgment.
Maintenance team
The people who repair, look after, and improve machines in the factory. Food tech example: the maintenance team fixes a pump that a vibration sensor showed was wearing out.
Operator
A person who runs a machine or a line in the factory. Food tech example: the operator at the mixer starts the mixer, watches the dough, and records the batch.
Record
A written or stored account of what was done, measured, or checked, with the time and the person attached. Food tech example: a record showing who cleaned a filler, when, and which checks were confirmed.
Robot
A machine that carries out a set of actions automatically, often moving and handling material. Food tech example: a robot that stacks boxes on a pallet at the end of a packing line.
Supervisor
A person who oversees a line or a shift, assigns work, and helps where it is needed. Food tech example: a supervisor sees that one line needs help and sends a worker there.
Tablet
A small portable screen that a worker can carry on the line to read instructions and record work. Food tech example: a tablet showing the steps for cleaning a filler at the filler itself.
Training
Learning how to do a job, by practising it with guidance and feedback, until it can be done well. Food tech example: a new operator learns to read a line chart by reading real charts from their own line.
Wearable device
A device worn on the body, such as on the wrist, that shows information or warnings while the wearer works. Food tech example: a wrist device that vibrates when a cold store temperature rises.
References
- Singh, R. P., Computer Applications in Food Technology, 1996.
- Irudayaraj, J., Food Processing Operations Modeling, Design, and Analysis, 2002.
Unit 7: Getting Started: a Practical Road Map for a Food Plant
Note
Unit 7: Getting Started: a Practical Road Map for a Food Plant
A question to start with
A small sauce factory wants to become a connected factory. The owner has read about sensors, cloud systems, digital twins, and smart factories, and a seller has offered a large system that promises to do everything at once. The owner is tempted, and also afraid: the price is high, the factory would have to stop for weeks while it is installed, and nobody in the plant knows how to use it. The production manager has a different idea. She suggests starting with one problem that costs the factory money every week: the filling line overfills some jars and underfills others, and the rework takes an hour a day. Fix that one problem first, she says, with a checkweigher and a shared record, and learn from it. Then fix the next problem, and the next. In a year, the factory will have changed a great deal, and it will have paid for each step with the saving from the step before.
This unit is about the production manager's way of starting. It gives a practical road map for a food plant: how to choose the first step, how to carry it out, how to know whether it worked, and how to keep going, step by step, until the plant is connected in the way that suits it.
Questions to think about
- A factory can afford one improvement this year. Should it buy the most impressive technology, or fix the problem that wastes the most time and product each week?
- A new system is installed, but nobody uses it after a month. What went wrong: the technology, the training, or the choice of problem?
- A plant manager wants to show that the first project saved money. What numbers should have been recorded before the project started?
Think about the sauce factory question before you read on
The manager is right. A large system that promises everything at once asks the factory to change its machines, its records, and its working habits at the same time, while stopping production and spending a large sum before any benefit appears. A small first project, chosen for a real problem that wastes time and product every week, can be installed without stopping the plant, learned by the people who use it, and judged by a clear saving. That saving then pays for the next step, and the factory learns what works in its own plant, instead of learning what a seller promised.
First, look at the plant as it is
Every road map starts with an honest look at the present plant, before any new equipment is bought. The factory walks its lines and writes down what it has: the machines, their age and condition, the sensors already fitted, the controllers, the records kept on paper and on computers, and the places where information is missing. It asks the people who do the work where time is wasted, where faults repeat, where quality varies, and where they guess instead of knowing. It looks at the records it already keeps: which batches were held, why lines stopped, how much was wasted, and how much energy was used.
This review, often called an assessment, should be short and practical. A week of walking, watching, and asking is enough for a small plant. The result is a list of problems in plain words, with a rough size for each: how much product is wasted, how many hours are lost, how often a fault repeats. The list is then ordered, not by which problem is most interesting, but by which one costs the most and is easiest to fix. The road map starts at the top of that list.
The review should also name what is already good. A plant that already keeps careful paper records has a habit of recording that a digital system can build on. A plant whose operators already check their work and report faults has a culture that a connected system needs. A plant whose machines are worn out and whose cleaning is poor should fix those basics first: no sensor will save a line that is not maintained, and no dashboard will fix a process that is not understood. The road map must be honest about this, or the new technology will be blamed for old problems.
Quick self-check
Question: Why should the review ask the operators where time is wasted, instead of only looking at the machines?
Because the operators live with the problems every day. They know where the faults repeat, where they guess instead of knowing, and where a small fix would save an hour a day, and a walk that ignores them misses the best information in the plant.
Question: A plant has worn machines and poor cleaning. Should it buy sensors first, or fix the machines and cleaning first?
It should fix the machines and cleaning first. A sensor on a worn, dirty line will only record the poor result more precisely, while fixing the basics improves the product and gives the later sensors a process worth measuring.
Choosing the first project
A good first project has four marks. First, it solves a real problem that people in the plant recognise, such as overfilling, a cold store that warms too often, or a pasteuriser temperature that drifts. Second, it can be measured: the factory can record the waste, the stops, or the variation before the project, and again after it, and the difference is plain to see. Third, it can be done without stopping the plant for long, and without changing everything at once. Fourth, the people who will use it are willing to learn it, because they helped choose it and they can see the benefit.
Examples of good first projects are easy to list. A checkweigher on a filling line, with a shared record of fill weights, attacks overfilling and underfilling at once. Temperature sensors and a simple alarm in a cold store attack spoilage and waste. A shared digital record of batch numbers, replacing separate paper logs, attacks the slow search for a batch history. A sensor on a pasteuriser holding tube, with a record of every reading, attacks the doubt about whether each batch was treated correctly. Each of these is small, measurable, and useful on its own, and each one teaches the plant something it needs for the next step.
A poor first project has the opposite marks. It is chosen because the technology is new, not because a problem needs it. It cannot be measured, so nobody can say whether it worked. It needs the whole plant to change at once. Or it is imposed on people who were not asked, so they do not use it. The most common poor first project is a large dashboard that shows everything and answers nothing, bought before the plant has agreed what its key numbers are. The road map avoids that trap by starting with a problem, not with a purchase.
Quick self-check
Question: What are the four marks of a good first project?
It solves a real problem people recognise, it can be measured before and after, it can be done without stopping the plant for long or changing everything at once, and the people who will use it are willing to learn it.
Question: Why is a large dashboard a poor first project?
Because it shows everything and answers nothing. Before a dashboard can help, the plant must agree what its key numbers are and how each one is calculated, and a dashboard bought before that agreement has no clear purpose.
The road map in stages
A road map is a sequence of stages, each one building on the last, and each one judged before the next begins. The stages below suit a small or medium food plant, and a large plant can move faster through them, but the order stays the same.
Stage one is to see. The plant adds measurement where it is missing, on the few things that decide quality and safety, and it starts a shared record with time and batch number, as Unit 3 described. The plant learns to look at charts, to spot a drift, and to keep its sensors calibrated. Nothing is automated yet, and nothing is changed automatically: the gain is visibility, and the measure of success is that the plant can answer, for any batch, what happened and when.
Stage two is to connect. The measurements, the laboratory results, and the planning records are joined by the same batch numbers and the same clock, so that the digital thread from Unit 3 runs through the plant. The plant can now trace a batch in minutes, compare batches, and find the cause of a problem from the record instead of from memory. The measure of success is the time it takes to answer a question about one batch, and whether the answer is trusted.
Stage three is to control. The plant adds the feedback loops from Unit 2 where they pay: a pasteuriser that holds its own temperature, a filler that corrects its own weight, a cold store that holds its own temperature. Each loop is tuned, tested, and watched, and the people who run it understand it. The measure of success is steadier quality: less variation between batches, fewer packs rejected, and fewer stops.
Stage four is to learn. The plant studies its stored data, as Unit 4 described, and finds the patterns that explain its waste, its stops, and its best runs. It builds a dashboard of a few key numbers, it compares shifts and lines fairly, and it tests a change in a controlled way before adopting it. The measure of success is a change that was found in the data, tested, and kept because it worked.
Stage five is to predict and to test. The plant builds a digital twin of one machine or one line, as Unit 5 described, and uses it to test changes safely and to warn of faults early. The plant also gives its people the connected tools from Unit 6: digital instructions, tablets where they help, and training for the new skills. The measure of success is a change that was tested in the twin first, and a fault that was found early because the twin saw the drift.
No stage should start until the one before it is working and its benefit has been measured. A plant that jumps to a twin before it can measure its own process will build a twin of a process it does not understand. A plant that jumps to control before it can see its process will automate a process it cannot judge. Patience here is not slowness: it is the way to keep each step paying for the next.
The diagram shows the stages in order. The steps point to the diagram: each stage builds on the one before it, and no stage should start until the one before it is working and its benefit has been measured.
Quick self-check
Question: Why should a plant not build a digital twin before it can measure its own process?
Because a twin is a model of the process, and a model of a process that is not measured and not understood will be wrong. The twin needs the measurement and the understanding that the earlier stages build.
Question: What is the measure of success for stage two, to connect?
The time it takes to answer a question about one batch, and whether the answer is trusted. If the thread is working, a batch can be traced in minutes from a single record.
Measuring whether it worked
Every project on the road map needs a baseline, a measurement of how things stood before the project, taken in the same way as the measurement after it. Without a baseline, the factory cannot say whether the project helped, and it cannot decide whether to spend more. The baseline should be recorded for a few weeks before the change, so that a good week or a bad week does not mislead the comparison. It should cover the thing the project aims to change: fill weight variation for a checkweigher, waste for a cold store alarm, stoppage time for a fault-finding project, or energy use per tonne for an oven project.
The measures should be few and plain. Waste as a share of production, packs rejected, line stops and their length, energy use per tonne, and the time to trace a batch are enough for most projects. Each measure needs a clear definition, agreed before the project, so that the before and after numbers can be compared fairly. A waste figure that changes its definition halfway through the project proves nothing.
The factory should also record the cost: the price of the equipment, the time spent installing it, the training time, and the running cost. The benefit, measured in product saved, time saved, or energy saved, can then be weighed against the cost, and the factory can say, in plain words, whether the project paid for itself and how quickly. A project that pays for itself in a year is a strong case for the next step. A project that does not pay should be studied, not hidden: the lesson may be that the wrong problem was chosen, or that the equipment was not used as planned, and that lesson is worth more than a neat report.
People, training, and change
A road map is a plan for people as much as for machines. Each stage needs a named owner, a person responsible for making it work and for reporting whether it did. Each stage needs training, short and practical, using the plant's own lines and records, as Unit 6 described. Each stage needs time for people to learn: a new screen or a new instruction will feel slow at first, and the factory should expect that and allow for it, instead of judging the project in its first week.
Change should be introduced with the people who will live with it. Operators, laboratory workers, maintenance staff, and supervisors should be asked what would help them, should see the new tool before it is installed, and should be able to say what is wrong with it after it is installed. A tool that is imposed without that conversation will be used unwillingly, and its record will be incomplete. A tool that is built with its users will be used, and its record will be trusted.
The factory should also plan for the people who find change hard. An older worker who has run a line well for years on paper should be offered training, time, and a colleague to learn beside, not a demand to change at once. A worker who prefers paper should be shown, with a real example from their own line, how the new record answers a question faster than the paper one. Respect for experience is not a soft extra: it is the way to keep the knowledge of the plant inside the plant while the tools change around it.
Risks, safety, and keeping the record true
A road map must name its risks and plan for them. The first risk is a network failure: the plant must be able to keep running safely when the network is down, with local controllers holding their loops and paper backup for the record until the network returns. The second risk is a sensor or a system that reads wrong: the plant must keep its calibration routine, must check surprising numbers, and must have a rule that a doubtful batch is held, not passed. The third risk is loss of the record: the plant must keep tested backups, as Unit 4 described, and must know how to restore them. The fourth risk is a change made by the wrong person: the plant must limit who can change a recipe, a limit, or a reported result, and must record every change with a name and a time.
Food safety rules do not change because the tools change. A pasteuriser must still hold its temperature for its time, a cold store must still hold its temperature, and a batch must still be traceable. The new systems must prove that they do these things, not only claim it. A factory that can show, from its own record, that each batch was treated correctly and can be traced, has used Industry 4.0 well. A factory that has new screens but cannot show that proof has changed its appearance, not its safety.
The road map should also be reviewed. Once a year, or after each stage, the factory should ask: what did we set out to do, what did we measure, what worked, what did not, and what should the next stage be? The answers should be written down, with the numbers beside them, so that the next stage starts from evidence, not from enthusiasm. A road map that is never reviewed becomes a list of purchases. A road map that is reviewed becomes a record of learning, and the learning is the real product of the journey.
Quick self-check
Question: Why does every project need a baseline recorded before it starts?
Because without a before measurement, taken the same way as the after measurement, the factory cannot say whether the project helped, and it cannot decide whether to spend more on the next step.
Question: What should a plant be able to do when its network fails?
It should keep running safely: local controllers should hold their loops, and the record should continue on paper backup until the network returns, so that no batch is made blind and no record is lost.
Try it yourself
Choose a small food business you know, such as a bakery, a dairy, or a sauce maker. Write a one-page road map for it in five stages, using the stages in this unit. For stage one, name the three measurements you would add first and why. For each stage, write one sentence saying how you would know it had worked. Then write one risk for the whole road map and one sentence saying how you would guard against it. Bring your page to your next session and compare it with a classmate: whose first stage costs less, whose measures are clearer, and whose risk is best guarded?
Key ideas to carry forward
- A road map starts with an honest look at the present plant: its machines, its records, and the problems its people name.
- Basics come first: worn machines and poor cleaning should be fixed before sensors are bought.
- A good first project solves a recognised problem, can be measured before and after, can be done without a long stop, and has willing users.
- The road map runs in five stages: see, connect, control, learn, and predict and test, each building on the last.
- No stage should start until the one before it is working and its benefit has been measured.
- Every project needs a baseline, a few plain measures with clear definitions, and a record of its cost, so that benefit can be weighed against cost.
- Each stage needs a named owner, short practical training, time to learn, and the involvement of the people who will use the new tools.
- Change should respect experience: training, time, and a colleague to learn beside, not a demand to change at once.
- Risks must be planned for: network failure, wrong readings, loss of the record, and changes by the wrong person.
- Food safety rules do not change with the tools: each batch must still be treated correctly and be traceable, and the record must prove it.
- The road map should be reviewed after each stage, with the numbers beside the judgments, so that the next stage starts from evidence.
Glossary
Assessment
A short, practical review of a plant as it is now, listing its machines, records, and problems. Food tech example: a week of walking the lines and asking operators where time is wasted is an assessment.
Baseline
A measurement of how things stood before a project, taken the same way as the measurement after it, so that the change can be judged. Food tech example: fill weight variation recorded for three weeks before a checkweigher is installed is a baseline.
Benefit
The gain from a project, such as product saved, time saved, or energy saved. Food tech example: fewer underweight packs rejected is a benefit of a checkweigher.
Cost
What a project takes: the price of equipment, installation time, training time, and running cost. Food tech example: the cost of a cold store alarm includes the sensor, the fitting, and the calibration.
Measure
A number that shows how well a process is doing, such as waste as a share of production or the time to trace a batch. Food tech example: energy use per tonne is a measure for an oven project.
Owner
A named person responsible for making a project or a stage work and for reporting whether it did. Food tech example: the production manager is the owner of the filling line project.
Pilot project
A small first project, done on one line or one problem, to learn whether an idea works before it is used more widely. Food tech example: a checkweigher on one filling line is a pilot project.
Review
A planned look back at what was done, what was measured, what worked, and what should happen next. Food tech example: a yearly review of the road map, with the numbers beside the judgments.
Risk
Something that could go wrong, such as a network failure or a wrong reading, which the plan must guard against. Food tech example: loss of the batch record is a risk that tested backups guard against.
Road map
A sequence of stages, each building on the last, that takes a plant from where it is to where it wants to be. Food tech example: the five stages from seeing to predicting and testing form a road map.
Stage
One step in a road map, with its own work and its own measure of success. Food tech example: the stage of connecting records into one thread for each batch.
Success
Meeting the measure set for a stage or a project, such as steadier quality or a faster batch trace. Food tech example: a checkweigher project succeeds when fill variation falls and the saving is measured.
References
- Singh, R. P., Computer Applications in Food Technology, 1996.
- Irudayaraj, J., Food Processing Operations Modeling, Design, and Analysis, 2002.