OEE and maintenance software for dairy and cheese plants

Key takeaways

• CIP is part of the recipe, not a chore you can shorten when you are behind, and it consumes a large share of the available hours.
• A pasteuriser flow diversion is two events at once: product pushed back to the balance tank, and an entry in your HACCP record.
• Fermentation is biology. A culture that acidifies slowly cannot be caught up by running the filler faster.
• Dairy yield lives in milk solids, so fat given away in standardisation and fat lost to whey never show up as rejects.
• A cracked heat exchanger plate is a food safety event, so plate integrity testing belongs in the maintenance plan with a date and a result.

Why a dairy needs its own version of this question

In a dairy the product sits inside closed pipework you cannot see into, changing temperature, pH and bacterial count every minute it stands there. That is why a dairy stop is rarely a clean pause: it is a decision about what to do with the milk already in the circuit.

This guide is for plant, production and maintenance managers at liquid milk, cheese, yogurt, butter and milk powder sites.

We have written the same guide for the bakery line and the meat and poultry plant. Dairy differs from both on one point above all: the cleaning cycle is written into the process, not squeezed around it.

From tanker to pallet: the process as a sequence

A dairy is a chain of circuits rather than a conveyor. A stop in one circuit backs up the one before it and starves the one after.

• Intake: tanker sampling and inhibitor testing, unloading, chilling, silo storage.
• Separation and standardisation: a centrifugal separator splits cream from skim, then fat is blended back inline to the declared figure.
• Heat treatment: thermisation, HTST pasteurisation or UHT, usually in a plate exchanger with a regeneration section.
• Homogenisation: a high pressure pump forces milk through a valve, usually two stage and commonly 150 to 250 bar in total, sited inside the pasteuriser loop so the milk is already at 60 to 70 °C and the fat is molten.
• Culture and fermentation: yogurt is inoculated and held warm, cheese milk gets starter and rennet in the vat.
• Cheese making: cutting the coagulum, whey drainage, stirring or cheddaring, pressing, brining.
• Ripening: controlled rooms, turning and coating, for weeks or months.
• Butter: cream ripened or not, continuous churning, buttermilk drained, working to moisture and salt, then wrapping.
• Evaporation and spray drying: falling film evaporator to around 48 to 52% solids, spray dryer, fluid bed, fines return and bagging.
• Filling and packing: cartons, bottles, cups, pouches or 25 kg bags, then coding and secondary packing.
• CIP: every circuit cleaned on a written recipe, then released before the next run.

Regulation (EC) 853/2004 sets the cold chain in stages: on farm, not more than 8 °C for daily collection or 6 °C if collection is not daily, and not more than 10 °C on arrival at the establishment of destination. On acceptance at a processing plant the milk must be cooled quickly to not more than 6 °C and held there until processed.

Two exceptions exist: the milk is processed within four hours of acceptance, or the competent authority authorises a higher temperature for technological reasons in making certain dairy products. The second is the one many cheese plants run under, so know which one is written into your own approval.

A carton filler cycles in seconds, a cheese vat in hours and a ripening room in months, so one plant OEE across them is an average of incompatible units. Measure each as its own line and never roll them into a single number.

Where the litres go, mapped to OEE

The six big losses map onto a dairy cleanly enough. The trap is that two of the largest dairy losses sit outside OEE altogether.

Fat given away during standardisation and fat lost to the whey stream never appear as a reject anywhere. They are solids losses, and we come back to them below.

What makes a dairy plant different

1. CIP is in the recipe, not in the margins

Cleaning in place is a scheduled, parameter driven cycle: pre-rinse, caustic, intermediate rinse, acid, final rinse, and often a disinfection step before the line is released. It is not discretionary, and it cannot be trimmed because the shift is behind.

Each circuit has its own recipe built on four recorded parameters: time, mechanical action, chemical concentration and temperature. In pipework the action comes from turbulent flow, commonly 1.5 m/s and above, while tanks are cleaned by spray devices sized on flow per metre of circumference, so a velocity rule proves nothing about a silo.

The pre-rinse must be cool, typically below 40 °C, because hot water bakes milk protein onto the surface and creates the deposit the caustic then has to fight. Caustic at roughly 1 to 2% and 75 to 80 °C lifts protein and fat, and the acid step that follows is not optional: it removes the calcium phosphate milkstone the caustic leaves behind.

Because CIP takes a large share of the hours in the day, the rule you pick for it moves your OEE more than anything else on this page. Our master sanitation schedule guide covers building the plan itself.

2. Flow diversion is rework with a paper trail

If the temperature after the holding tube falls below the set point, the flow diversion valve sends product back to the balance tank instead of forward. The line keeps turning, the filler starves, and nothing saleable comes out.

In the EU, HTST pasteurisation of milk means at least 72 °C for 15 seconds, or 63 °C for 30 minutes, or an equivalent combination, and the treated product must give a negative alkaline phosphatase reaction immediately after treatment. The 15 seconds is the residence time of the fastest particle, not an average.

That has a consequence most OEE reviews miss: the holding tube is a fixed length, so the 15 seconds only holds at or below the flow rate it was sized for. A pasteuriser therefore has no catch up gear, and pushing the feed pump above validated flow is a food safety change, not a performance improvement.

Diverted milk is not thrown away, it goes round again, so the real cost is a second heat load on the same milk plus the time the filler stood empty. Repeated diversion cooks flavour and shelf life out of a batch long before anything shows up as a reject.

Log the count, the duration and the volume returned as their own stop reason, not buried inside a generic "pasteuriser fault". A diversion is a downtime event, a rework event and a HACCP record at once, and usually only the first of the three gets recorded.

3. A cracked plate is a recall, not a repair

The regeneration section of a plate heat exchanger warms incoming raw milk against outgoing pasteurised milk, which saves a great deal of energy. A thin plate and two gaskets are all that separate the two streams.

Protein and mineral deposits build up on those plates, pressure drop climbs, and plates eventually crack or corrode. The safeguard is a booster pump downstream of the holding tube that keeps the pasteurised side above the raw side at every point in the regenerator, so a pinhole leaks pasteurised milk into raw rather than the other way round.

Set the margin from your own validation, commonly at least 0.1 bar, and measure it with a differential pressure transmitter and recorder rather than a pair of gauges. Loss of that differential must divert flow exactly as a low temperature does, and the interlock is worth proving on test, not assuming.

That control only works if the differential is genuinely monitored. Put plate integrity testing and the gasket change history on the maintenance plan, with a date, a method and a result.

4. Fermentation cannot be caught up

Stirred yogurt is inoculated at around 43 °C and fermented in tank until the pH reaches about 4.6, typically in four to six hours, then cooled and smoothed before filling. Set yogurt is filled first and ferments in the cup, so there the incubation room, not the filler, is the constraint.

The usual causes of a slow set are inhibitor residues in the milk, bacteriophage, an inoculum that has lost activity, or an incubation temperature a degree or two low. Under heat treating the base costs you too, because yogurt milk is held at 90 to 95 °C for around five minutes precisely to denature the whey proteins that build the gel.

No amount of filler speed recovers the delay, because the product simply is not ready. Give fermentation overrun its own stop reason, pointed at the culture or the milk base rather than at the filling machine.

The same logic runs through the cheese vat, where coagulation, cheddaring and pressing take the time they take. The cost of a slow set lands on the next batch in the queue.

5. Changeovers are about fat, flavour and allergens

Going from 3.5% to 1.5% fat is a standardisation change with an interface loss, not a full clean. Going from plain yogurt to one carrying hazelnut pieces is an allergen changeover with a validated clean behind it.

A third kind catches plants sharing a pasteuriser between drinking milk and cheese milk. Homogenised milk gives a weak rennet curd and lipolysed flavours, so the homogeniser must be bypassed for the vat run, and that bypass route is worth proving rather than assuming.

Sequence the plan so that allergen bearing and strongly flavoured recipes run last in the block. Our guide to allergen changeover validation covers proving the clean, and SMED covers shortening the stop itself.

6. Yield is counted in milk solids

Cow's milk is roughly 12 to 13% total solids: fat, protein, lactose and minerals. Everything the plant sells is built from those solids, so yield is a solids question and never a litres question.

Cheese yield depends on how much fat and casein ends up in the block instead of the whey, and cutting the coagulum too early or stirring it too hard pushes both into the whey stream. Cut on a multiple of the flocculation time or on a measured curd firmness, never on the clock alone, because rennet activity moves with pH, calcium and milk temperature.

Then follow the escaped solids. A fines saver on the whey line and a whey cream separator turn much of that loss back into product, and a blinded fines screen is a silent yield leak nobody writes a work order for.

A cheese plant can therefore post a healthy OEE while its yield quietly slips. Keep a solids balance next to the OEE board, and review them in the same meeting.

How to count CIP time in OEE

CIP is the biggest single block of non producing time in most dairies, so where you put it swings the headline number by ten points or more. Write the rule into your OEE definition document and apply it to liquid, cheese, yogurt and powder alike.

The table below is our recommendation and not a published rule. Nakajima's original framing already puts planned cleaning outside loading time, but no standard forces it, so a dairy that classifies CIP differently is defensible as long as it says so on the report.

Keeping scheduled CIP outside planned production time points OEE at what the shift team can actually change. Report CIP hours on a line of their own, so a cycle creeping from 80 minutes to 110 stays visible to somebody.

Worked example: one milk filling line

Here is a single shift on a line filling 1 litre gable top cartons of pasteurised milk. The numbers are round so you can check every step.

1. Availability

Stops inside planned time were 16 + 9 + 15 = 40 minutes, so run time is 400 minus 40 = 360 minutes.

Availability = 360 ÷ 400 = 90.0%.

2. Performance

At 105 cartons per minute, 360 running minutes should have produced 37,800 cartons, and the line produced 34,776.

Performance = 34,776 ÷ 37,800 = 92.0%, mostly from short stops at the carton magazine and a filler held below rated speed while foam settled after the fat change.

3. Quality

904 cartons were rejected for seal defects, fill level or an unreadable date code.

Quality = 33,872 ÷ 34,776 = 97.4%.

4. OEE, and the one step check

OEE = 90.0% × 92.0% × 97.4% = 80.6%.

Check it in one line: 400 planned minutes at 105 per minute is 42,000 possible cartons, and 33,872 ÷ 42,000 = 80.6%.

What the 97.4% does not include

The milk diverted during those 9 minutes went back to the balance tank and was pasteurised again, so it never passed the filler counter and is nowhere in the 904.

Nor are the water and product interfaces at the start and end of the run, or the milk pushed out of the pipework when CIP began.

Move the 80 minute CIP back inside planned time and the base becomes 480 minutes, or 50,400 possible cartons, which drops the same shift to 67.2%. Both figures describe the shift honestly, so the reporting rule has to travel with the number, and our OEE calculation guide sets out the full method.

Worked example: what a tenth of a fat point costs

Standardisation is where solids leave the plant for free. Take a site processing 300,000 kg of milk a day, declared at 3.5% fat, with an inline average of 3.6%.

• Excess fat: 0.1 percentage points
• Milk processed: 300,000 kg per day
• Butterfat transfer value: 5.00 EUR per kg (use your own)

0.1% of 300,000 kg = 300 kg of fat per day that could have stayed in the cream stream.

300 kg × 5.00 EUR = 1,500 EUR per day, and over 300 operating days that is 450,000 EUR a year.

Halving the overrun to 0.05 points returns 225,000 EUR a year. Narrowing the spread is what buys that, because the target can only come down once the variation does.

The floor is legal, not commercial. Regulation (EU) 1308/2013 fixes drinking milk at not less than 3.50% fat for whole, 1.50 to 1.80% for semi skimmed and not more than 0.50% for skimmed, so a plant with a wide inline spread has to sit well above the limit just to stay inside it.

A cheese vat leaks solids the same way, into the whey instead of the cream tank. Run that arithmetic before anyone argues about half a point of OEE.

The dairy maintenance plan

Dairy breakdowns cluster on a small number of high energy assets and on the hygienic barriers around them. The refrigeration plant is the one that fails without stopping anything: as ice bank capacity or condenser performance drops, the pasteuriser outlet creeps from 4 °C towards 6 °C and you lose shelf life instead of minutes.

Anywhere a lubricant could reach product, specify an NSF H1 registered grade and keep a list of which grade sits at which point. Our H1 vs H2 food-grade lubricants guide explains the difference.

Two traps catch dairies repeatedly, and a third catches almost everyone. A spray ball that has lost two holes still passes a visual check but leaves a shadow in the tank, and a CIP strength reading taken at the tank rather than at the return line says nothing about what reached the far circuit.

The third is compressed air. Air that opens a carton, clears a nozzle or drives a mix proof valve seat sits in the product zone, so it needs a written quality class, commonly ISO 8573-1 class 2.2.1 for direct contact, with dew point checks and particle and oil filter changes on the maintenance plan.

See how to reduce machine downtime for choosing what to fix first, and our preventive maintenance schedule guide for turning the table above into a plan.

Food safety and the evidence an auditor asks for

Regulation (EC) 852/2004 carries the general hygiene duties for any food business. The dairy specific detail, covering raw milk temperatures, heat treatment and the phosphatase test, lives in Regulation (EC) 853/2004.

On top of that, Regulation (EC) 2073/2005 fixes microbiological criteria reaching several dairy products, among them Listeria monocytogenes in ready to eat foods and Salmonella in milk and whey powder. The one that lands on engineering every week is the process hygiene criterion for Enterobacteriaceae in pasteurised milk, tested at the end of the manufacturing process.

An unsatisfactory result sends you to the efficiency of the heat treatment and to recontamination after it, which in practice means the regenerator, the pasteurised side pipework and seals, and the last CIP. Treat a run of borderline results as a maintenance signal, not a QA argument.

Read the current consolidated text of each before quoting a clause at anyone. They set out what has to be achieved and recorded, and none of them names a piece of software.

Most retail customers then add a voluntary certification scheme such as BRCGS Food Safety, IFS Food or FSSC 22000, each with its own current issue and clause numbering.

What they all demand is documented evidence, much of it produced by engineering rather than by QA.

A history line that reads "repaired, Tuesday" satisfies nobody. A work order carrying the task, the parts, the technician, the time and the hygiene release does.

An auditor will ask the milk intake and the cheese room the same question, so keep one maintenance record and let the hygiene evidence hang off it. Our HACCP compliance guide sets out the prerequisite programmes that sit under it.

Questions to put to any vendor

Ask every supplier the same questions, and insist the answers are demonstrated on a running line rather than in a slide deck.

Getting the numbers off the plant

• Which PLC protocols does it speak, and has it been connected to your filler and pasteuriser skid brands before?
• If a machine's controller is closed or ancient, what is the fallback: a retrofit sensor, or a camera watching the discharge?
• Will it register a short stop of a few seconds with nobody pressing anything?
• Can the flow diversion signal land as its own named event rather than an anonymous fault?
• Does rated speed vary by product and pack size, so 250 ml and 1 litre are not judged against one figure?

Logic a dairy actually needs

• Can the system tell a planned CIP apart from a changeover clean and a re-clean, and count them differently?
• When a cycle runs long, can the overrun alone be charged to availability?
• Can loss be entered in litres or kilograms, tied to the stage and the stop behind it?
• Is the stop reason list pickable with wet gloves in a couple of taps? See downtime reason code design.
• Where do fat standardisation and solids yield live, alongside OEE or in a clean export?

Maintenance and the evidence trail

• Can a preventive plan be driven by the calendar, by run hours or by a condition, and reused as a template?
• Scanning a QR code on a homogeniser, does a technician get its whole history on a phone?
• Can a job hold a hygiene release step that somebody signs before the line restarts?
• Will the closed record stand up in an audit, showing the person, the date, the work and the parts?
• Do spares carry minimum levels, so seal kits, plate gaskets and valve seats are not ordered the day they fail?

The same questions, asked across the wider sector, sit in our review of the best CMMS for food and beverage.

The first 90 days

Days 1 to 30: one line and a rule everybody signs

Choose the line that hurts most, which in most dairies is filling and packing. Settle the CIP definition on paper before anyone pulls a cable.

Wire up the filler and the pasteuriser skid, enter a rated speed for every product and pack size, then agree a short reason list that names flow diversion and fermentation overrun explicitly.

Register that line's assets, put a QR code on each one, and carry the preventive tasks over unchanged for now.

Days 31 to 60: work the two biggest losses

Sort the losses by minutes lost rather than by how irritating they are. On a dairy filling line the top pair is usually the changeover block and short stops at the carton or cup magazine.

Put SMED on the longest changeover, then reorder the weekly plan so the allergen cleans cluster instead of repeating.

Turn CIP verification, probe calibration, plate integrity testing and hygiene release into recurring tasks, each with a named owner.

Days 61 to 90: solids, proof and the second line

Bring fat standardisation and solids yield into the daily OEE review, held standing at the line rather than in an office.

Shift wear parts such as homogeniser seals, plate gaskets and valve seats off the calendar and onto run hours or a measured condition.

Finally, read a month of your own records as an auditor would, fix what is thin, and only then take on line two.

How Fabrico helps

Fabrico puts OEE measurement and a complete CMMS in one place, which is what the sections above keep asking for.

The measurement side takes machine data from PLC connections, IoT sensors or computer vision cameras. A cup filler whose controller gives you nothing useful can still be counted by a camera.

Availability, performance, quality and OEE are calculated live, short stops included, and each stop carries the reason your team picked for it.

The maintenance side is where your engineers work: they raise work orders, build preventive plans from recurring templates, and reach a machine's file by scanning its QR code on iOS, Android or the web.

What that leaves behind is documented history an auditor can read, job by job. Inventory carries minimum and maximum levels, so homogeniser seal kits, plate gaskets and valve seats are on the shelf before the breakdown.

To be plain about the boundary: Fabrico does not forecast failures, does not open work orders by itself and does not plan production. It records what happened and puts the history, the losses and the plan in one place, and your people do the deciding.

If you would rather measure your CIP window and your filling losses than argue about them, book a 30 minute demo with a Fabrico consultant, no commitment, or contact us.