The glossary of words this category uses

Every category has a vocabulary that sounds precise and is not. This one is worse than most, because four of its category names — MES, MRP, ERP and SCADA — genuinely overlap, and because a great many of its measurements are ratios whose denominators are choices somebody made. The result is a specific and expensive failure: two people agree on a word, sign something, and find out in month three that they meant different things by it.

So this is written to be argued with rather than quoted. Where a term has one honest meaning, it gets a short entry. Where it has two, both are here and you are told which one to pin down first. Several of these will be worth reading even if you never buy any software at all.

The categories, and where they blur

MES

Manufacturing execution system. The layer between the commercial system and the machines: which works order is on which machine, in what sequence, with what material, inspected how, labelled how, traceable how. Used two ways. Narrowly it is a defined layer with a defined set of functions. Loosely it is any software on a shop floor that is not ERP — which is how two people agree they need an MES and mean two different purchases. The question that settles it is which functions: scheduling, execution, quality, traceability and labelling are separable, most sites need two or three of them first, and a supplier who cannot tell you which of the five they are strong at is describing a category rather than a product.

MRP

Material requirements planning. Given a demand, a bill of materials and what you hold, work out what to buy and make, and when. MRP II widened that to capacity, labour and money, and in ordinary speech “MRP” now often means the whole planning-and-purchasing half of an ERP — so establish which one is meant before agreeing to anything. The practical point is upstream of the arithmetic: MRP is only as good as the on-hand number it starts from, and on most floors that number is wrong because consumption is recorded somewhere other than where it happened. Decide which system is authoritative for quantity and make the other one visibly refuse the write. Two systems that both accept adjustments will quietly disagree, and you will find out at stocktake.

ERP

Enterprise resource planning: the commercial record. Sales orders, purchasing, stock value, costing, invoicing, and usually the master data everything else borrows. A works order is normally born in the ERP and its cost eventually lands back there. What an ERP is generally poor at is the middle — what actually happened on a machine, minute by minute — and that is structural rather than a criticism. An ERP is built around transactions people enter. A production system is built around events machines emit, at a rate nobody would type.

SCADA and PLC

A PLC is the controller on the machine; it runs the process. SCADA supervises a set of them — screens, setpoints, alarms, control in real time. Two things follow. SCADA answers “what is this machine doing right now, and can I change it”, which is a different question from “of the time we paid for, how much became good parts”. And SCADA and production monitoring software are not competitors, whatever the overlap in the screenshots: software reading a controller over OPC UA, read-only sits alongside SCADA rather than arguing with it, and a site that has SCADA has not already bought monitoring.

Production monitoring

Measuring what machines did — states, cycles, stops, scrap — and turning it into OEE, downtime analysis and a record that survives being questioned. “Machine monitoring” is used for the same thing and also for something much narrower: up or down, and nothing about the job, the part or the scrap. Three questions separate them. Does a stop carry a reason? Is a cycle attached to a works order? And does an empty window say “no production in this window” rather than reporting a convenient 0%? Ours is described on the monitoring page, and what to ask a supplier is a guide of its own.

The OEE family

OEE

Overall equipment effectiveness: availability × performance × quality, as one percentage. Of the time you intended to produce in, how much became good parts at the rate you expected. Three ratios multiplied is what makes it unforgiving — three numbers you would each be reasonably pleased with produce a total you would not. Its weakness is that every input is a definition somebody chose, so an OEE figure quoted without its definitions is a number rather than a measurement. The arithmetic, a worked example and the six usual ways a figure gets flattered are in OEE in plain English; the calculator does it on your own shift.

Availability, performance and quality

The three legs, each with exactly one soft input. Availability is run time over loading time — did it run when it was meant to — and its soft input is what you took out of loading time. Performance is actual output against what the standard cycle time says that run time should have produced, and its soft input is the standard. Quality is good parts over total parts made, and its soft input is which side of the line you put rework. Anybody can move an OEE figure several points by adjusting one of the three without going near a machine, which is why the three inputs are worth more scrutiny than the output.

Loading time

Also called planned production time. The period you intended to produce in: your real shift pattern, minus what you deliberately chose not to count — breaks, planned maintenance, a scheduled shutdown. It is the denominator of availability and therefore the easiest place in the whole calculation to improve a number without improving anything. Two rules keep it honest: whatever you take out must be a category you defined in advance, and you must be able to point at it afterwards. A third is worth asking software for — change the working calendar and history should re-state itself, so that there are not two versions of last March depending on when you ran the report.

TEEP

Total effective equipment performance: the same three ratios measured against all of the calendar — 168 hours in a week — instead of against the time you chose to staff. It answers a different question from OEE. Not “how well did we run the shifts we ran” but “how much of this asset are we using at all”, which is the capital question and the one a finance director actually has. Used two ways in practice, and the argument is always about the denominator: some sites take all 168 hours, others take the hours the business could staff if it wanted to, and “utilisation” gets used for both. Say which hours are underneath and the disagreement disappears.

Time, as the floor talks about it

Downtime and stoppage

A stoppage is the machine not running. Downtime is usually the accounted version — a stoppage that has been given a start, an end and a reason — and the gap between the two is where most reporting arguments live. Three things are worth pinning down: the grace period before a stop becomes an event, whether the event is timestamped from when the machine stopped or from when somebody noticed, and how much of your downtime carries no reason at all. That last figure, reported in its own right, tells you what share of your Pareto is evidence and what share is just the part somebody got round to classifying. Underneath sits the micro-stop: short enough that an event for each would be noise, invisible in every downtime report, and a machine that stalls twenty times a shift loses a shift.

Cycle time and standard cycle time

Cycle time is how long one cycle actually takes. Standard cycle time is how long it is supposed to take. Both get called “the cycle time”, and the confusion is expensive, because performance is measured against the standard. If the standard came off a quotation five years ago, or was set by whoever had the stopwatch, your performance figure is measuring the quotation rather than the machine. A machine beating its standard produces a performance figure over 100% and most systems quietly cap it — which throws away the signal, because a press consistently running at 108% is telling you the standard is wrong. Ask where each of your standards came from before you defend a figure with one.

Takt time

Not a cycle time, and routinely used as though it were. Takt is available time divided by demand: the rate at which you have to finish a part to meet the order book. It is a target set by the customer, not a measurement of the machine. Cycle time faster than takt means you are ahead and building stock; slower means you are late whatever your OEE says. This is the one term in the list that connects the floor to the order book, and using it to mean “how fast the press goes” loses that.

Changeover

The time between the last good part of one job and the first good part of the next. The definition is the whole argument. Last good part to first good part includes the setting, the first-off inspection and the parts you scrapped proving it — that is the honest number and it is usually much larger than the one being reported. Stop the clock when the tool is in the press and you have measured the crane. On a floor of tools a changeover is also not one event: it is a tool change, often a preheat, a person from the toolroom and a first-off approval, any of which can be the thing you were actually waiting for. Planning it as one block of minutes is how it overruns.

SMED

Single-minute exchange of dies. The method for shortening changeovers: separate the work that can be done while the machine is still running from the work that can only be done when it is stopped, then move as much as possible from the second group to the first. “Single-minute” means single digits of minutes, not one minute — worth saying before somebody promises a board a sixty-second tool change. Most of the gain on a first pass is unglamorous and has nothing to do with software: the trolley ready before the machine stops, the bolts all the same size, the paperwork already printed.

Parts, quality and capability

First-pass yield and right first time

The share of parts that were right without rework — two names for broadly the same idea, RFT being the one you will hear in a UK factory. Two traps. Per-operation and whole-route are different numbers: five operations at 95% each is 77% end to end, and quoting the per-operation figure flatters you by nearly twenty points. And rework and concession decide the answer. A part that needed twenty minutes of hand-finishing before it passed is not free, and counting it as good makes quality look like 100% while the cost sits in somebody else’s budget. Whichever side of the line you put rework, write it down and leave it there — a definition that moves between months is worse than one you disagree with.

SPC

Statistical process control: charting a characteristic over time to separate ordinary variation from a real change, so you act on the second and leave the first alone. Easy to draw and easy to get wrong, and the failure modes are always the same three — limits recomputed from the very data being judged, capability quoted without saying how sigma was estimated, and a chart drawn from six points. It is also a clear case of a term used two ways: to a quality engineer SPC is a method with rules about when you are allowed to act, and to a lot of software it is a chart shape. The mechanics of the first version are on the quality and SPC page.

Control limits and specification limits

The most consequential confusion in this list. Specification limits come from the drawing — what the customer will accept. Control limits come from the process — what it does when nothing unusual is happening, calculated from its own variation. They have nothing to do with one another, and all four combinations exist: inside spec but out of control means something changed and you are getting away with it; in control but outside spec means nothing is wrong and the process was never capable of the tolerance. Drawing spec limits onto a control chart is common and destroys the only thing the chart was for. In software, the thing to check is that control limits are frozen from an audited baseline study rather than recomputed on page load — otherwise a process drifting out of control can quietly redraw its own limits and look in control all the way out.

Cp and Cpk

Capability indices, and they ask different questions. Cp asks whether the tolerance is wide enough for the spread of the process — could it fit. Cpk asks whether it does fit, where the process is actually sitting, so the gap between the two is pure centring. Both depend on how sigma was estimated, within-subgroup or overall, and the same data gives different answers under each — which is why the estimator belongs on screen next to the number rather than in a methods note. Pp and Ppk are the overall-sigma pair of the same idea. The long version, including how much data a study honestly needs before the number means anything, is in Cp vs Cpk.

Following the part

Works order

Also job, production order, shop order: the same object under four names. The instruction to make a quantity of a part, and the thing against which time, material, scrap and cost are booked. It is the hinge of this entire category, because it is the one record both the commercial system and the floor system want to own. It is also the reason no shop floor is fully automatic — a works order is a decision, not a signal. No machine can tell you which job it is running, so somewhere a person scans or selects it, and from that moment everything measured automatically attaches itself to the right order.

Lot and serial

A lot is a quantity made or received together and treated as one unit for traceability. “Batch” is the same thing; “lot” is the word a recall notice uses. A serial identifies one physical item. Most manufacturers need both, arranged as a hierarchy — serials under lots, lots under items — so that a unit still has a parent when the supplier sent no lot code of their own. The real decision is what gets serialised: serialising everything is expensive, and serialising nothing makes the scope of a recall a guess. It is also worth keeping two codes where they differ, the one your floor scans and the one a customer’s recall notice will quote.

Traceability

Used two ways, and the difference between them is a weekend. Sometimes it means the records exist somewhere — certificates in a folder, labels in a file, a box in the archive. Sometimes it means you can ask the question and get an answer: backwards from a finished unit to every lot inside it, forwards from a suspect lot to every unit carrying it. Those are the same walk run in either direction, so a system that can do one should be able to do both. Ask for the forwards one in a demo, because it is the direction a real recall runs and it is the harder of the two. Ours is on the stock and traceability page.

Andon

Originally a lamp: a signal from the floor that something needs attention, visible from a distance without anybody having to ask. Now usually a screen on a wall. What makes one work is not the screen — it is that the number on it is the same number the office is looking at, and that it is readable at a glance by everyone. Colour alone is not enough, because roughly one in twelve men on a shop floor cannot reliably separate the red and the green that andon boards are built from. Colour plus a glyph plus the word costs nothing and is the difference between a board people use and a board people walk past.

Poka-yoke

Mistake-proofing. Design it so the wrong thing cannot be done, rather than training people so that it usually is not — a fixture that only accepts the part one way round is the classic. The software equivalent is a refusal at the point of action rather than a warning in next month’s report: a quarantined lot that cannot be issued to a job however you reach it, a mis-scan or an already-consumed batch rejected at the scan. The test of any poka-yoke, physical or otherwise, is what it does when somebody is in a hurry at the end of a shift.

Getting the data off the machine

OPC UA

An open, vendor-neutral standard for getting data out of industrial controllers, and the one worth asking for by name. It carries structure as well as values, so a reading arrives with a type and a meaning attached rather than as a bare number you have to interpret, and it can be read strictly read-only — which is what lets monitoring software sit alongside an existing SCADA without anyone having to worry about writes. Two honest notes. OPC UA on a datasheet does not guarantee that the value you want is actually exposed. And plenty of perfectly good machines have none at all, which is not a dead end: counting a spare relay contact or reading the machine’s electrical signature produces the same production record by a different route.

Modbus and MQTT

Two much older and much simpler ideas that still do most of the work. Modbus is a register protocol: you read numbered registers, and it is entirely on you to know that a particular register holds kilowatt-hours and is scaled by ten. It carries no meaning at all, which makes it robust and makes a wrong register map look exactly like a right one — so a new instrument is worth proving against a known load before anybody reports off it. Modbus TCP is how most three-phase meters report. MQTT is not a machine protocol at all; it is a lightweight publish-and-subscribe transport. A device publishes to a broker and whatever cares subscribes. It tells you that a value arrived. What the value means is a separate agreement, and the projects that go wrong are the ones that treated a topic name as a specification.

Cost, energy and the moulding floor

Cost per part

What one part cost to make — and the word to listen for is what is inside it. A figure built only from energy is an energy cost, and it is a small fraction of the answer. A usable one is the composite: energy plus material, labour and a fair share of overhead, tied to the works order that produced it, so it can be set against the standard cost you quoted from. Two details decide whether it survives contact with a moulding floor. Cavities multiply parts but not energy, so a four-cavity tool makes four parts a shot and the arithmetic has to know that. And the standard you are comparing against is, on most sites, a number worked out years ago and adjusted by feel ever since — which is worth establishing before anybody celebrates a variance.

Specific energy consumption (SEC)

Energy against output rather than energy in total. For plastics processors under a sector agreement the unit is kilowatt-hours per kilogram of product made. The point of using a ratio is that growing your output is not failing — a site whose energy per unit creeps up between assessments is. Both halves can be measured where the work happens: kilowatt-hours metered at the machine, and pieces from the cycle count with cavities included. The step from pieces to kilograms is your own part weight, which is one multiplication you control rather than something that falls out of a demo. If your sector holds a Climate Change Agreement, SEC is the number the target is assessed on — the scheme itself, what missing it costs and how often it is reassessed are in our CCA guide.

Shot, cycle and pressing

Three words for one event, and which you use says what you make. A shot is injection vocabulary: one injection of material, producing as many parts as the tool has cavities. A pressing is the compression equivalent. Cycle is the neutral word, and the one to use when the room contains both. It matters more than vocabulary, because counting in cycles is what lets a tool’s service interval follow actual usage instead of the calendar — “this mould is 4,000 shots from its 250,000 service” — and lets the count travel with the tool when it moves between presses. On a moulding floor the tool is the asset and the press is only where it happened to be that week. Why that changes maintenance is on the moulders’ page.

The three worth settling before any meeting

If you take nothing else from this page, take three definitions and get them agreed in writing before anybody demonstrates anything. What is in loading time, because it decides availability and it is the easiest number in the building to move. Where each standard cycle time came from, because performance is measured against it and a stale standard flatters you all year. And which side of the line rework sits on, because that decides quality and it is the one people change without noticing they have changed it.

Settle those three and every OEE figure you are shown becomes comparable with every other one, including your own. That is worth doing whether you buy anything or not — and if you are at the stage of buying something, the questions worth sending a supplier first follow on from here.