Reduce changeover time by separating preparation from stopped-machine work, putting adjustable settings into controlled recipes and making physical change parts locate repeatably. Measure from the last good part of the old job to the first good part of the next, including setup checks and any restart scrap.
Motionwell applies these methods to assembly machines, packaging equipment and loading cells in Singapore. On a vision-guided SCARA panel line, recipes select robot paths and vision parameters for different panel variants. A tray-fed filling and sealing machine uses machined tray sets for container formats. One example centres on settings; the other needs a repeatable physical swap.
The short answer. Changeover time on an automated machine is cut in four places, in this order. Decide which differences between products will travel in a recipe and which will need a physical change part, and specify the axes and tooling so the recipe carries as much as your variant list allows. Design every change part to locate on a datum that stays put when the part is swapped, so no robot point, vision job or press stroke needs re-teaching. Make each change part fit one way only, and make the machine confirm which one is fitted before it will cycle. Then move everything that does not need the machine stopped, kitting, recipe preparation and program checks, out of the stop. What remains, measured from the last good part of the old job to the first good part of the new one, is the time the mechanism genuinely needs, and that is the number a machine builder can be held to at factory acceptance.
Filling lines have their own version of this problem, with cleaning and format parts in the product path, and it is covered on our filling line changeover page. This page is about the rest of the factory: assembly cells, secondary packaging and end-of-line equipment, and the loaders that feed machine tools and test stations. Much of what is published on changeover was written for a press shop, and it carries over to automated equipment only in part, which is where we start.
Where Does Changeover Time Hide on an Automated Machine?
SMED was worked out on presses, where a setup is one die, its bolts and a shut height. The method’s two moves, separating work that needs the machine stopped from work that does not, then shortening what is left, still hold. What changes on an automated machine is that the setup is no longer one object. It is spread across a dozen small items owned by different people: a nest, a set of gripper fingers, a feeder track, a robot program, a vision job, a torque window, a handshake with the next machine. A press changeover includes setting and first-good-part verification after clamping. An assembly cell changeover is finished when the feeder presents the new part, the gripper holds it, the robot puts it where the nest expects it, the camera accepts it, the fastener lands inside its window and the first assembly passes. Each of those can add time that nobody records as changeover, because each is small and each belongs to somebody else.
| Machine | What physically changes | What is set in software | What has to be proved before the first good part |
|---|---|---|---|
| Assembly cell, whether SCARA, rotary dial or inline pallet | Nests, gripper fingers, feeder tooling, press anvils | Robot paths and pick offsets, vision jobs, torque and force windows, sequence options | A placed part measured in position, a force curve inside its band, a fastener inside its torque window |
| Secondary packaging and end-of-line | Carton guides, flight bars, pusher plates, labelling head position, gripper for the case format | Pattern, pitch on motorised axes, count, print artwork with its verification job, collation | A case closed and squared, a code that reads back, a layer that sits inside the pallet footprint |
| Machine-tending loader | Gripper fingers, blank magazine or tray, chuck jaws or fixture on the machine tool side | Pick grid, place position and approach, handshake with the machine program, part count | A blank seated fully in the fixture and a first part measured against its drawing |
| Filling line | Format parts in the product path, trays | Volume, torque, code, height where the axis is motorised | Covered on the filling line changeover page |
Read the last column first. Geometry is settled at concept, software is whatever a motor or a parameter controls, and the last column is where the time hides on many lines, because proving the first good part is a task nobody designed a place for. Measure the whole thing from the last good part of the old job to the first good part of the new one, and log the mechanical swap and the verification tail as separate entries. Why that split matters to the availability figure is set out in our note on measuring OEE. What the same losses look like across a whole plant is in our guide to improving production efficiency on existing equipment.
What Does SMED Mean When Half the Setup Is Software?
The internal and external split still applies; the list is different. On automated equipment a large share of what is done with the machine stopped is not mechanical. Someone selects a recipe at the panel after the last part. Someone checks it is the right revision. Someone jogs the robot to a point to see whether it still lands. Someone runs the vision job on a sample to see whether the threshold holds. Recipe preparation and offline program checks can often be completed before the stop. Physical setup checks and first-part verification remain in the changeover unless the design provides an equivalent validated method outside it.
| Done with the machine stopped today | Where it can go | What the machine has to support |
|---|---|---|
| Selecting and editing the recipe at the panel after the last part | Recipe queued for the next job while the current one runs, selected by job order or by scan | A recipe store with a next-job slot and a selection that takes effect at the job boundary |
| Confirming the recipe matches the job | A check of recipe identity against the job order before the stop | Recipe identity carried as data with a version, and a comparison the controller can make |
| Re-teaching robot points for the new variant | No teaching at all, because points are referenced to a datum that did not move | Datum-referenced programming, covered below |
| Tuning vision thresholds on the first parts | The vision job for the variant held in the recipe and proven at qualification | One vision job per variant, versioned with the recipe |
| Swapping feeder bowls or track tooling | A second bowl pre-loaded, or a variant that shares the track | Feeder tooling designed per part family |
| Homing, warm-up and a dry cycle | Sequenced after the new setup is confirmed and the operating area is clear | A controlled start with the required guards, tool identity and motion permissions satisfied |
Stable location and qualified programs reduce routine re-teaching, as the next two sections explain. Parallel changeover tasks need both a suitable access layout and a coordinated work sequence, and it is worked through on the filling line changeover page.
On the SMED figure itself: the percentage reductions quoted for SMED come from press shops, and no single percentage describes an automated line. A cell whose variants differ only in recipe reaches a figure that a cell with hardware in every changeover never will, so the two should not be averaged into one number. Log recipe-only changeovers and hardware changeovers as separate populations, because a shift in the mix between them moves the average without anything on the machine having changed. The same goes for cobots: a cobot cell changes over as fast as its fixtures, its gripper and its recipe allow, so the arm type is not what any benchmark would be measuring.
How Do You Design a Change Part That Fits One Way Only?
A change part is any piece of the machine cut to one product: a nest, a set of gripper fingers, a feeder track, a carton guide, a pusher plate, a chuck jaw. On our own bills of materials, nest plates, quick-change couplings and format part sets appear as their own line items alongside the frame, because they are drawn, machined and stored as parts in their own right. That is a useful test to apply to any machine: if the change parts have drawing numbers, someone designed the changeover; if they do not, it was left to the fitter.
The design rules are old, and they are worth listing because a machine that misses one of them loses the time in a place the specification never mentions.
| Design choice | Typical practice | What it removes from the changeover |
|---|---|---|
| Location separated from clamping | Two dowels and a face locate; the fastener only holds | Adjustment. The part goes to the same place every time, and the clamp cannot pull it off position |
| Asymmetry | An asymmetric pin pattern, keyed feature or offset locator that physically blocks the wrong orientation; round and diamond pins separately control location without overconstraint | The wrong-way fit and the scrap that follows it |
| Captive hardware | Fasteners retained in the change part; quarter-turn or lever clamps where the load allows | Dropped screws, missing screws, and the search for the right length |
| Hard stops | A shoulder, a stop pin or a keyed slot at the setting position | Setting by eye, and the drift between one operator and the next |
| Identification on the part | Variant marked on the part and in the kit list, with a feature the machine can sense | The question of whether the right part went on, covered further down |
| Wear surfaces | Hardened dowels and bushes, or replaceable inserts on the mating faces | Location drifting after repeated swaps, which shows up as a re-teach nobody can explain |
| Handling | Handles, lifting points, a cart position, and a mass one person can manage | Two-person swaps, and damage to the locating faces in transit |
The first row does the most work. A change part located by the same screws that clamp it sits wherever the screws pulled it, which is a different place every time and the reason a scribed line and a feeler gauge come out at every swap. Two dowels and a face take the position decision away from the operator, and the fastener then only has to hold, so it can be a lever, a quarter-turn or a captive screw reached with one hand.
Whether a gripper should be a change part at all, or built adjustable so one tool covers several formats, is a separate decision worked through on our end of arm tooling page. Feeder tooling is the hardest change part to make quick, because a bowl track is cut to one part’s geometry, and the alternatives are on our part feeding page.
Where Should the Locating Datum Live So Nothing Needs Re-teaching?
A robot program, a vision job and a press stroke are each written relative to something. A stable work frame and repeatable change-part location let qualified variant offsets carry through a swap. Product geometry, fixture wear and tool calibration still need checking; a new part does not automatically require a completely new program. That one sentence is behind much of the difference between a changeover measured in hours and one measured in minutes on a robot cell.
The arrangement looks like this. The machine carries the datum: a dowel pattern on the tooling plate, a locating face and a stop on a pallet, the chuck or fixture on a machine tool. The change part carries the product-specific geometry and locates to the datum on a primary face, a secondary edge and a single stop, so it cannot rock, slide or rotate. The robot is taught once, to the datum. Each variant’s offset from the datum is a number in the recipe taken from the drawing, and it is measured on the machine once, at qualification, never taught at every swap.
Where the offsets cannot be trusted, because the product itself varies or the tray does, a camera measures them. That is the arrangement on the SCARA panel line: the camera finds the fiducial and shifts the robot’s trajectory on each cycle, so the locating job lives in the camera, and a new variant is a new vision job and a new set of paths. The boundary is stated on the case study: a variant that needs a different gripper geometry or a different tray format still needs hardware, and the tray side of it is a nest plate change.
Three tells that the datum is in the wrong place. A re-teach after every swap can indicate inconsistent fixture seating or a reference-frame problem. Offsets that grow over months call for checks of datum wear, clamping and tool calibration. And a machine that runs well only after one particular technician has set it, which means the datum is in that technician’s hands. On a rotary dial the datum is the tooling plate and the nest locates to it, which is why the plate is drawn before the dial, as set out in how we design multi-station rotary assembly machines. On a loader the datum is on the machine tool side, and the loader’s job is to seat the blank against it, which is why a chip on the locating face matters more than the gantry’s repeatability, a point made on our machine tending page.
What Belongs in the Recipe on an Assembly Cell, a Packaging Line and a Loader?
Reducing machine changeover time with the PLC comes down to one rule, set out on the filling changeover page: a setting enters the recipe only if a drive or a parameter controls it. What differs by machine type is what that list contains and where its edge sits.
| Machine | The recipe holds | Still hardware |
|---|---|---|
| Assembly cell | Robot paths and offsets, grip profile, vision job, torque and force windows, sequence options, parts per tray | Nest geometry, gripper finger geometry, feeder tooling, press anvil |
| Rotary dial | Station dwell where the indexer is servo driven, press stroke and force window, vision job | Nest geometry and the stations on the tooling plate |
| Secondary packaging and end-of-line | Layer pattern, pitch on motorised axes, count, print artwork with its verification job, collation | Guide geometry outside the motorised range, a gripper for a different case family, flight pitch on a fixed chain |
| Machine-tending loader | Pick grid, place position and approach, handshake, the program number called on the machine, count | Gripper fingers, magazine or tray, chuck jaws or fixture |
Motorising an axis is what moves an item from the right-hand column to the left, and it has a price in drives, guarding and commissioning that should be paid only where the variant list demands it. A packaging line whose formats differ in carton length but share a height is a candidate for one motorised axis. A loader whose blanks share a diameter and differ in length needs a recipe entry for the pick height and one set of fingers, which is a cheap changeover, whereas blanks that differ in diameter need fingers per family, and the family count is what the cell should be priced on.
Two things about the recipe itself. It has a version and an owner, and the machine should record which version ran each batch, because a changeover to the wrong revision of the right recipe is invisible until the parts are measured. And who may select a recipe is a different permission from who may edit one; on a regulated line that split is a controlled record, and the record-keeping side is on the filling changeover page. Where the recipe physically lives, in the PLC, on the panel or on a server, decides who can reach it during a network fault, and that question is taken up in our comparison of SCADA and HMI roles on a machine.
How Do You Stop the Machine Accepting the Wrong Change Part?
A recipe and its fitted tooling can disagree on a press, assembly cell or packaging machine. If the recipe calls for variant B while variant A tooling is installed, the result can be interference, equipment damage or parts made to the wrong settings. A changeover design is not complete until the machine can tell what is fitted.
| What can be wrong after a swap | How the machine finds out | Typical implementation |
|---|---|---|
| Wrong change part for the recipe | Identity check before cycle start | A coded pin pattern read by proximity sensors, a tag on the part, or a scan at fitting, compared with the recipe’s kit list |
| Right part, unseated | Seated check | A sensor on the datum face, or clamp position confirmation |
| Right part, worn | Trend on the first-part result | First-part measurement compared with previous changeovers on the same recipe |
| Right hardware, wrong recipe revision | Recipe identity and version compared with the job order | Version carried with the recipe and logged at selection |
| Gripper not changed | Tool identity at the coupling | Identification contacts on the coupling read at cycle start |
The identity check is an item often missing from specifications, and it costs little to design in. A pin pattern on the change part and a bank of proximity sensors on the machine give a code the controller can compare with the recipe’s kit list before it allows a cycle to start. A tag or a scan does the same with less machining and more electronics. Either way the rule is the same: the recipe carries a list of the hardware it expects, and the machine refuses to run until the list is satisfied. That refusal has to be logged, because the record of a refused start is what tells you afterwards why a changeover took longer than the mechanism did.
How Do You Shorten the Verification Tail After the Swap?
The changeover ends at the first good part, and on many lines proving that part takes longer than the swap, because verification was designed as a quality task done at a bench, when it could have been a machine function done in the first cycle.
Three design moves shorten the tail. First, an approved setup-verification sequence: confirm the fitted tooling and motion permissions, then initiate the required homing or dry run with the operating area protected. Clearance checks use the qualified paths and the observations or sensors specified for that test. Second, measured values on the first cycle, where most machines report only pass or fail. A press station that reports its force curve, a vision station that reports the measured position, a screwdriver that reports the torque and angle it reached, all tell the operator where the first part sits inside its window, and a part that passes at the edge of the window is the earliest warning that a change part is seated wrong or worn. On the SCARA line the fastener torque is set per fastener type from the recipe and the result is logged against the panel serial, so the first panel’s evidence exists the moment it is made. Third, verification at the station where the fault can occur, and never only at the end of the line, so a wrong placement is caught before further operations are performed on it; the station-ordering rule behind that is on the rotary machine page.
The loader case is different and worth stating. On a machine-tending cell the first part after a changeover is usually proved by measuring the machined part against its drawing, which is a machine tool question. What the loader contributes is seating confirmation before the machine program is allowed to start, so a blank that is not fully home is caught before it is cut.
When Is Scheduling the Cheaper Fix Than Hardware?
Changeover loss is the number of changeovers multiplied by the duration of each. Hardware and recipe design shorten the duration. Sequencing reduces the number, and it costs nothing on the machine.
The mechanism is the changeover matrix. For every pair of variants the machine runs, write down what has to change between them: recipe only, recipe plus a gripper, recipe plus a nest, recipe plus feeder tooling. Variants that differ only in recipe can follow each other with a changeover measured in minutes, which is the SCARA line’s case for its panel variants. Variants that share change parts should be grouped so the hardware swap happens once per group. A scheduler that knows the matrix can order the week so the expensive changeovers cluster and the cheap ones fill the gaps, which is the practical meaning behind changeover time reduction with a scheduler: the scheduler is only as useful as the matrix it is given, and the matrix is a design output. That matrix is what we ask for at concept review, because it is the same information that decides how many change part sets to make.
Two cautions apply. Grouping jobs has a cost in inventory and lead time that belongs to planning, and it reduces how often a swap happens without shortening any of them. And a scheduler cannot fix a changeover that needs re-teaching, because a re-teach takes the same time whichever order the jobs run in.
What Does a Changeover Change in the Safety Case?
Changeover can involve access inside guarding or powered setup tasks. A task-based assessment distinguishes work requiring isolation from controlled powered setting, and ISO 12100 lists setting, teaching or programming and process changeover among the phases of use that hazard identification has to cover, alongside normal operation, cleaning and maintenance. A changeover designed for speed and assessed only for production mode has a gap in it.
Three consequences follow for the design. Access built for a quick swap has to be the same access the safety case allows: a hinged guard on an interlock, so opening it is fast and stops the hazard. Any task that needs motion with a guard open, jogging a robot to check a point or indexing a dial to reach a nest, is a setting mode with its own reduced speed and hold-to-run control, defined and validated as a mode in its own right, and never a defeated interlock. And the change part itself is part of the hazard picture: a nest half-fitted when a cycle starts is a projectile, which is a second reason for the seated check described above beyond scrap. How a task-based risk assessment is built for a machine is covered in our guide to machine safety risk assessment, and how access points and interlocks are laid out for a cell is on our machine guarding design page.
How Do You Prove a Changeover Time Before You Accept the Machine?
You prove it by running it, with your people, before the machine leaves the builder’s floor. A changeover requirement that reads as a demonstration can be tested; one that reads as an adjective cannot. Write it as a pair of variants, a named operator, the parts kitted at the cell, the clock running from last good part to first good part, and the verification included in the clock.
What the builder needs from you to design to that requirement is short. The variant list, with the differences between variants tabulated as dimensions, because the difference matrix is what decides which differences go in the recipe and which in steel. The changeover frequency, per shift or per week, because that decides whether a motorised axis or a second change part set is worth its cost. Who performs the changeover, and how many people are available for it. And which variants are coming in the next few years, because a change part set can be added later but a motorised axis or a datum scheme generally cannot.
At factory acceptance the demonstration is run on the machine at our Woodlands Link facility by your operators, with the time recorded against the requirement. Where the machine has an identity check and inline first-part verification, both get exercised on the day. The items that belong on that day’s list are set out in our factory acceptance test checklist. Where the machine is one station in a longer line, the changeover has to be proved across the line, and the line-level version of this argument is on our packaging line integration page.
The figure for your own line is established at acceptance, on your parts and with your people, which is why the requirement above is written as a demonstration. The two delivered builds illustrate different changeover mechanisms: recipe-selected assembly parameters on the SCARA line, and machined tray-format sets on the filling platform.
Which Four Parts Make Up a Changeover Redesign?
Reducing changeover time on an automated machine is a design exercise with four parts: a recipe wherever a drive or a parameter controls the setting, change parts that locate on a fixed datum and fit one way, a machine that confirms what is fitted and proves the first part itself, and everything else moved out of the stop. Scheduling reduces how often you pay; design reduces how much. Which end of that range your line lands at, a recipe-only change or a swap of machined steel, is decided by the variant list before the frame is drawn.