English Views: 0 Author: Elsa Long Publish Time: 2026-11-08 Origin: Site
A JUKI SMT machine FAT SAT checklist is worth writing before the purchase order, not the week the machine ships, because the two tests answer different questions and only one of them is cheap to fail. The factory acceptance test asks whether the machine was built and configured as ordered. The site acceptance test asks whether it works with your boards, your feeders, your conveyor and your operators. A defect found at the factory costs a repair; the same defect found on your floor costs a repair plus travel, schedule and credibility, and the difference is not linear.
This checklist runs the two gates in order and treats each as a measurement rather than a demonstration: configuration and documents before power-on, three test vehicles chosen to expose different weaknesses, placement results recorded as numbers, programs and parameters handed over in a form you can keep, line integration tested at the handover boundary, training signed off by task, and every open item closed through a classified punch list. Every calculation below is labeled as an assumption so it can be replaced with your own figures, and the platforms referred to are the ones in the placement machine range this acceptance plan was written for.
The two acceptance gates are usually written as one document and executed as two events, and the confusion between them is where most disputes begin. The factory acceptance test is a build check. It verifies that the machine in the crate matches the configuration that was ordered, that the head, options and software are what the paperwork says, and that the machine places components to its specification on a board chosen to demonstrate that. The site acceptance test is an integration check. It verifies that the machine works inside your line, with your boards, your feeder mix, your conveyor heights and your operators, at the cycle time your production plan assumes.
Both gates can pass and the purchase still disappoint, because a machine that is correctly built and correctly integrated can still be unable to hold capability on your most demanding product. That is a third question, and it is answered by measurement rather than by demonstration. Treating it as part of the site test is what turns a two-day acceptance into a three-week argument, so a JUKI SMT machine FAT SAT checklist has to separate the three questions explicitly: was it built right, does it work here, and does it meet the requirement. Each has its own evidence and its own decision.
The failure modes differ because the available remedies differ. At the factory, a mismatch can be corrected by changing a part, reflashing software, adding an option from stock or re-running the build, and the machine has not yet travelled. At the site, the same mismatch usually requires a return visit, a shipped part and a rescheduled installation, and the line is standing idle in the meantime. That asymmetry, not the size of the defect, is why the factory test deserves more preparation than it usually gets.
The second difference is who can see the problem. Factory tests are witnessed by people who know the machine; site tests are witnessed by people who know the product. A defect that only appears with your board geometry, your panelization or your mark placement is invisible at the factory unless you bring those boards. A defect that only appears with a particular feeder width or a particular nozzle is visible at the factory if those items are in the crate and loaded, and invisible at the site if they were never shipped. The checklist therefore requires the test vehicles and the feeder set to travel with the test, which is a logistics decision made weeks earlier.
Item | Factory acceptance test | Site acceptance test |
|---|---|---|
Question answered | Was it built and configured as ordered? | Does it work in this line, with these boards? |
Typical duration | Half a day to two days | Two to four days including training |
Boards used | Your test vehicles, shipped in advance | Your production boards and panels |
Cheapest remedy | Part swap, reflash, add option from stock | Return visit, shipped part, rescheduled line |
Who witnesses | Process engineer plus maintenance | Process, production and quality |
Payment link | Often none, which is the problem | Balance on signed acceptance |
The reason to spend money on the earlier gate is arithmetic rather than principle. Take a hypothetical defect with a base correction cost of one unit at the factory: a part, a technician's hour, a re-test. The same defect found during the site test typically costs three to five units, because the technician travels, the part ships, the schedule moves and the plant absorbs a day of disruption. Found in the third month of production, the same defect costs ten to fifteen units, because it now includes scrap, expedited freight, overtime and the engineering time spent proving that the fault is the machine's and not the process's. These multipliers are our own planning assumptions, not measured data, and the shape of the curve is what matters rather than the exact values.
The curve has a second implication that buyers rarely price: the cost of finding a defect late is often paid in credibility rather than money. A line that misses a launch because an option was never enabled damages the equipment buyer's standing with production and with the customer, and that cost does not appear in any purchase order. It is the reason the factory test should be treated as the main event and the site test as confirmation rather than the reverse.
One practical consequence: the effort should be front-loaded into the items that are hardest to fix later. Configuration and options are cheap to verify and expensive to correct at the site. Software licences are cheap to verify and impossible to fix without paperwork. Feeder inventory is cheap to verify and expensive to ship. Accuracy on your own board is the one item that genuinely needs the site, because the board and the line are both there, so it belongs in the second gate and should not be allowed to crowd out the first.
Acceptance is a decision, and decisions need the people who can make them. Three roles belong at the factory test: a process engineer who understands what the machine has to achieve, a maintenance technician who will live with the mechanics, and somebody with authority to sign. Two more join at the site: a production supervisor who can release the line for testing, and a quality representative if the product is audited. Sellers sometimes prefer a single visitor, and a single visitor is exactly the arrangement that produces a signed report nobody can defend later.
The roles also determine what gets tested. A process engineer tests capability and cycle time. A maintenance technician tests access, lubrication points, error recovery and the practicality of the service procedures. A production supervisor tests whether the machine can be run by the people on shift. A quality representative tests the documentation trail. Each of those is a different pass criterion, and a test plan that covers only the first will produce a machine that meets its specification and frustrates everyone else.
Where travel is expensive, the roles can be split: the process and maintenance people attend the factory test, and the production and quality people join the site test with a briefing from the first pair. What does not work is sending nobody to the factory and treating the site test as the whole acceptance. By then the platform choice is closed, the configuration is built, and the only remaining lever is the punch list. The screening that decided the platform is described in the platform screening that decided which machine is being tested, and it should be closed long before this checklist starts.
No board should run until the machine and the paperwork agree. This is the cheapest gate in the whole process and the one most often skipped, because the machine looks right, the seller is friendly and everyone wants to see it place components. The problem with skipping it is that a configuration mismatch discovered after the test has run is ambiguous: nobody can tell whether the missing option was never ordered or was removed during the test, and the conversation becomes an argument about memory rather than a correction.
The gate has three parts: reconciling the machine against the order, assembling the document set that must exist before power-on, and confirming that the utilities and safety conditions allow the test to be run safely at all. All three can be done in two hours with the covers closed, and all three produce findings that are trivial to fix at this stage and expensive later.
Take the order, the build sheet and the machine, and walk them line by line. The order says what was bought; the build sheet says what was built; the machine says what is present. Any divergence between the first two is a supplier issue to resolve before the test starts. Any divergence between the second and third is a shipping or handling issue, and it is worth knowing about before the crate is unpacked rather than after. The reconciliation covers the head type, the option list, the feeder bank, the nozzle set, the conveyor configuration and the software level.
The reconciliation is also the right moment to confirm the identity: serial on the nameplate, serial on the build sheet, serial in the software. Three sources that agree are a solid foundation; two that disagree are a reason to stop and resolve before anything else happens. This step catches the rare but expensive case of a machine delivered with the right badge and the wrong head, which is not obvious from a photograph and is very obvious from a build sheet.
The document list that anchors this step is the same one that belongs to the purchase, and it is worth checking it against the purchase checklist that should have produced these documents. If an item on that list was never obtained during the purchase, it will not appear at acceptance, and the acceptance report will have a hole in it that surfaces during the first audit.
Six documents should be in the room before the machine is energised, and their absence is a finding rather than an inconvenience. The build or configuration sheet establishes what the machine is. The electrical drawing, or at least the terminal diagram, establishes how it is connected. The utility requirement gives voltage, current, air pressure and consumption for this configuration rather than for the platform in general. The calibration record, where one exists, establishes the state of the axes and the vision system. The licence document establishes what the software may do. The spare parts list with lead times establishes what happens when something fails.
Two more documents belong to the test rather than to the machine: the test procedure with pass criteria, and the report format that will be signed. Both should be agreed before the visit, because a procedure written on the day tends to be written around what the machine does. The report format matters more than it sounds: it determines what evidence is captured, and evidence that is not captured during the test cannot be reconstructed afterwards without another visit.
Where a document is genuinely unavailable, record it as an open item with an owner and a date, and decide before the test whether it blocks acceptance. Some items can wait, such as a spare parts catalogue. Others cannot, such as the licence document or the calibration record for a machine whose accuracy is being accepted. The distinction is worth making explicitly rather than leaving to the end of the day when everyone is tired and the machine is running.
Some conditions stop a test regardless of how well the machine performs, and they should be checked first so that the day is not wasted. The supply voltage and phase configuration must match the machine's setting, and the earth arrangement must match what the machine expects; a mismatch produces intermittent faults that look like machine defects. Air pressure at the machine inlet must be measured at the machine, not at the compressor, and during a run rather than at rest. The floor must be level enough for the machine's own levelling to work, and the machine must be anchored or shimmed as the manual requires.
Safety checks belong in the same list because they cannot be deferred: guards present, interlocks functioning, emergency stops tested, warning labels legible, and the working area clear. A test run with a bypassed interlock is not a valid acceptance run, and accepting a machine with one is accepting liability. The configuration being verified is the one on the flexible inline platform this plan is written for, and its utility table should be the reference rather than a general assumption about machines of that size.
The last check before power-on is the least technical and the most useful: confirm who is allowed to touch what. During an acceptance test there are two teams with different habits, and a machine adjusted by one while the other is measuring produces numbers nobody can trust. Agree at the start who operates, who observes and who signs, and write it at the top of the report. It takes two minutes and prevents the most common source of disputed results.
Most acceptance tests use one board, and that board is usually chosen because it is available rather than because it is revealing. A single production board tests production, which is valuable, and leaves three other failure modes untested: the component at the edge of the machine's capability, the geometry that stresses transfer and clamping, and the recognition case that the vision system was never configured for. Three vehicles cover those in a day and produce a report that can be defended later.
The vehicles need to be prepared before the machine ships, because they have to travel to the factory test if that gate is to be worth anything. Prepare them as a small kit: ten to twenty boards of each type, taped and labelled, with a written statement of what each one is intended to expose and what result counts as a pass. The statement is the part that turns a demonstration into a test, because it fixes the criterion before anyone sees the machine run.
The limit board carries the extreme components from your BOM, and nothing else that could mask a failure. That means the smallest chip you place, the largest body you place, the tallest part, the finest pitch, and the component with the most difficult recognition characteristics, whether that is a dark moulded body, a reflective lead or a part with almost no contrast against the board. Running this board first tells you whether the machine can do the job at all, before any discussion of rate or capability.
The limit board also validates the tooling. Each extreme component implies a nozzle and a feeder width, and if either is missing the test stops there rather than in the third week of production. Load the widest tape on the limit board even if your volume product never uses it, because the wide feeder path is where jams appear and where a bank that looked healthy reveals its wear. The feeder widths that correspond to your tape mix, and the bank they fit, are described in the feeder widths loaded for the test.
Keep the limit board small and repeatable. Ten boards are enough when the purpose is capability at the extremes, and a small board keeps the cycle short so the test can be repeated after an adjustment without consuming the day. Record which extremes were tested and which were not, because a limit board that covers six of your eight extremes is a partial result and should be reported as one.
The production board is the one you will actually run, at the panelization you will actually use. It tests the machine against the mix rather than against the extremes: the number of part numbers, the feeder positions consumed, the changeover that the family requires, and the cycle time the line plan assumes. This is the board that answers the question the plant cares about, which is not whether the machine is capable but whether it can produce.
Run at least a full panel set, and run it more than once. The first run proves the program; the second proves repeatability; the third, after a reel change, proves that the material path works when the tape is not fresh. The changeover between two products is worth including if your family changes weekly, because changeover is where the feeder bank, the carts and the program structure either work together or reveal that they were never designed as a set.
The production board also produces the data used later in this checklist, so treat the run as a data collection exercise rather than a demonstration. Log the cycle time of every board, the stoppages and their causes, the pick errors by feeder position, and the number of manual interventions. That log becomes the baseline against which the first month of production is compared, and it is far more useful than a single figure for throughput.
The stress board tests the parts of the process that fail intermittently: panel edges, where clamping and support decide whether a component lands accurately; tall parts, where clearance and head height decide whether the machine can pass over without contact; and fiducial marks, where contrast and placement decide whether the machine can find the board at all. None of these failures appears reliably in a short run, which is why the board is designed to provoke them.
Design the stress board deliberately: place components at the extreme corners of the panel, include the tallest body in the assembly near the panel edge, and use the mark pattern your production panels actually carry rather than a convenient one. If your product uses tooling holes or bad marks, include them. Run the board enough times to see whether the failure is systematic or random, and record which position failed and how often, because a failure that repeats at one corner points at support or clamping while a failure that moves points at recognition or nozzle condition.
Test vehicle | What it is built from | What it exposes | Pass criterion to agree in advance |
|---|---|---|---|
Limit board | Smallest, largest, tallest and finest-pitch parts only | Capability at the edges, nozzle and feeder coverage | All extremes placed within the agreed offset band, no manual intervention |
Production board | Your real BOM and panelization | Mix, feeder positions, changeover, cycle time | Three consecutive runs without a stop, cycle time within the agreed band |
Stress board | Corner parts, tall bodies, production marks | Clamping, support, clearance, mark recognition | No placement outside band over the agreed number of runs |
A machine that places components neatly is not the same as a machine that meets a specification, and the difference is only visible in numbers. Observation tells you that the head moves, the feeders index and the boards come out looking right; it does not tell you the mean offset, the spread, or how much margin remains before the product fails at the edges of the panel. Those three numbers are what the acceptance decision should rest on, and they are cheap to obtain once the test vehicles exist.
The measurement plan has to be written before the run, because the sample size, the pattern and the criterion all determine whether the result means anything. A criterion invented after seeing the data is not a criterion. A JUKI SMT machine FAT SAT checklist that survives an audit uses industry-standard characterization logic, which is what makes the numbers comparable between machines and between suppliers, and it keeps the arithmetic simple enough to do on the day.
Thirty boards with twenty measured placements each gives six hundred points, which is enough to estimate a mean offset to within about half a micron when the spread is around ten microns. The arithmetic is the standard error: ten microns divided by the square root of six hundred is roughly 0.4 microns, so a mean estimated from that sample is unlikely to be more than about a micron away from the true value. That precision is sufficient to decide whether the machine is centred, and it costs one shift of production time rather than a week.
Fewer points are acceptable when the product is forgiving and more are needed when it is not. A sixteen-board sample with ten points each is enough to detect a gross offset but not enough to certify a fine-pitch process. The important discipline is to fix the number before the run and to state it in the report, because a sample size chosen after the fact always looks adequate in hindsight. These figures are an illustrative plan with assumed values, not a requirement of any standard.
Sample across the panel rather than from one area. Ten points clustered in the centre of one board describe the centre of one board; twenty points spread across the field describe the machine. If the panel carries multiple boards, sample each one, because the differences between board positions are often larger than the differences between panels and they point at support, clamping or conveyor alignment rather than at the head.
Three numbers come out of the sample and they answer different questions. The mean offset tells you whether the machine is centred; a mean of several microns in one direction is a calibration or teaching question rather than a defect. The standard deviation tells you whether the machine is consistent; a large spread means variability that no re-centring will fix. The capability index combines both against the specification and gives a single number that can be compared with the requirement.
Take a hypothetical sample with a mean offset of 6.5 microns and a standard deviation of 9 microns against a specification of plus or minus 35 microns. The capability index is 35 minus 6.5, divided by three times 9, which is about 1.06. That passes and it passes thinly, which matters: a machine at 1.06 on your easy pattern will likely fall below 1.0 on the edge pattern, and the product will show it as intermittent defects at panel corners. Run the calculation on both patterns and report both numbers.
Where the number falls short, the sequence of remedies is diagnostic rather than arbitrary. Re-centring addresses the mean. Checking support pins and clamping addresses edge-specific spread. Cleaning and re-teaching the vision system addresses spread that appears only on certain components. Nozzle condition addresses spread that appears only on small parts, and the replacement part numbers and orifice sizes are listed by family in the nozzle families and orifice sizes, which is the reference for checking that the set in the crate matches the components being measured.
Placement rate is the number most often accepted on trust and the easiest to measure properly. Count the placements attempted and the placements that required intervention over a defined run, and express the result as a rate rather than an impression. Five thousand attempted placements with twelve interventions is a rate of 0.24 percent, which is a working figure for a healthy machine on production material; the same run with sixty interventions is a machine that will consume an operator's shift in recovery.
Record the failures by feeder position rather than in total, because the distribution is the diagnosis. Failures concentrated on one position point at a feeder, a tape or a nozzle. Failures spread across positions point at vacuum, vision or the head. Failures that appear only after a reel change point at the splice or the tape path. A total without a distribution tells you that something is wrong and nothing about what to fix.
Rate claims deserve the same skepticism at acceptance as they do at purchase. A platform published at a headline figure places that number under favourable conditions on a demonstration board, and the class of machine behind those figures, such as the high-rate rotary class behind the published numbers, will deliver less on a mixed BOM with frequent feeder indexing. What the acceptance test should confirm is the rate on your board, not the rate in the brochure, and the two are separated by the mix factor and the transfer time.
Recognition failures do not announce themselves. The machine places the part, the board looks correct, and the defect appears at inspection as a rotated or offset component that the machine reported as good. The cause is usually a recognition routine that was never taught the specific part, or a lighting and camera setting that works for chips and fails for dark or reflective bodies. The acceptance test is the last cheap moment to find this, because after installation the same discovery costs a production interruption.
Test recognition explicitly on the components that are hardest to see: the smallest chip, the darkest moulded body, the part with reflective leads, and any component whose orientation matters. Record what the machine reports for each, and compare it with what the vision system was configured to recognize. Where a part is recognized only under a particular lighting setting, write that setting down, because the next operator will not rediscover it and will instead conclude that the machine is unreliable.
Include the tray or stick-fed parts if your product has them. These are outside the tape path and are often configured late, which means they are configured during acceptance under time pressure and never documented. A recognition result for a tray part that depends on an undocumented teaching step is a latent failure, and the checklist item is not only that it works but that the procedure for making it work is written down.
The machine will run on the day of acceptance whatever happens to the knowledge behind it, because the engineer who commissioned it is standing there. The test of a good handover is what happens on the first day that engineer is not there. Programs, parameters, mark settings and the recognition routines for specific parts all live in the machine and in somebody's head, and the acceptance checklist should require them in a form the plant can use without the person who created them.
This is also the gate where new and used purchases diverge most. A new machine's programs are built during commissioning and the builder is contractually present; a used machine arrives with whatever archive the seller transferred, and the parameter set may reflect a different product entirely. The verification logic for a machine that has already been in service, including what to ask for and what to check before shipping, is set out in the verification framework written for machines already in service, and the acceptance report should record which of its items were satisfied.
Ask for the programs in three forms: the machine's own backup format, an export in a documented interchange format if one exists, and a written list of which program belongs to which product. The backup is the operational copy; the export is the escape route if the software generation changes; the list is what makes the archive usable by somebody who was not present when it was built. A folder of files without the list is an archive that will be explored rather than used.
Then test the restore. Having the seller or the commissioning engineer delete a program from the machine and restore it from the backup is a two-minute demonstration that proves the backup is a backup rather than a copy. The test also reveals whether the restore requires a parameter set that lives elsewhere, which is a common gap: the program comes back but the vision settings do not, and the machine places components badly until somebody notices.
Record the software version alongside the programs. An archive is only restorable on a compatible version, and a plant that later updates the machine may find that the archive no longer loads. Keeping the version number in the report costs nothing and turns a future failure into a documented condition rather than a mystery. The same record should state who holds the licence and where the media are stored.
Parameters are the settings that make a machine suit a product, and they are the least documented part of any installation. Machine-level parameters cover axis calibration, conveyor width limits, vacuum thresholds and safety timings. Program-level parameters cover feeder offsets, placement heights, recognition thresholds and mark tolerances. Both sets are editable, neither is usually printed, and a single well-meaning adjustment during a production problem can move the machine away from the state in which it was accepted.
The checklist item is a captured baseline: a full parameter export, taken after acceptance and stored with the report, plus a written note of which parameters were changed during commissioning and why. Where the software supports access levels, agree who may change machine-level parameters and record it. The purpose is not to prevent adjustment but to make restoration possible, because without a baseline there is no definition of correct to return to.
Two specific settings deserve their own line in the report. The first is the conveyor width and the clamping arrangement, because it determines what the machine will accept and is easily changed for one job and forgotten. The second is the vacuum or air-pressure threshold, because a marginal setting produces intermittent pick failures that are diagnosed as feeder faults for weeks. Both are worth photographing on screen as well as exporting, since a photograph survives a software change.
Three kinds of data are usually forgotten because none of them is a program. The mark and panel definitions live in a separate structure in most systems, and they are what allows the machine to find the board; losing them means re-teaching every product. The component library, with its recognition settings and dimensions, is the accumulated knowledge of what each part looks like. The alarm and event log is the record of what the machine has complained about, and it is the most useful document for the first month of production because it shows the pattern of faults that the acceptance run may not have provoked.
Ask for all three at handover and store them with the acceptance report. The alarm log is worth exporting even when the machine is new, because it establishes a baseline: a fault that appears in week three can then be compared with a log that shows the machine was clean at acceptance. Without that baseline, every early fault becomes a debate about whether it existed on delivery.
The audit trail belongs in the same package if the product is regulated. Records of who changed which program, when the machine was calibrated and which version of the software was running are exactly what an auditor asks for, and reconstructing them after the fact is impossible. Capturing them at acceptance costs an hour and is the difference between a documented process and a claim about one.
The machine on its own is a specification; the machine in a line is a process. Integration is where the acceptance test stops being about the machine and starts being about the interface: conveyor heights, handshake signals, board transfer times and the cycle balance with the stations on either side. Training is where the test stops being about the equipment and starts being about the people who will run it. The punch list is where both are converted into obligations with dates.
These three items are usually compressed into the last afternoon of a site visit, which is why they generate most of the first-month problems. They deserve their own half day, scheduled before the visit rather than after the measurements, because a line that cannot hand boards across is not going to demonstrate a cycle time no matter how well the head performs. On a JUKI SMT machine FAT SAT checklist, these three items are the ones that most often move from "passed" to "still open" a month later.
Three interface conditions decide whether the machine works in the line. The mechanical one is height and position: the conveyor must sit at the line height within the adjustment range, and the machine must be positioned so the transfer distance matches the neighbouring stations. The electrical one is the handshake, which in SMT lines conventionally follows the SMEMA interface standard for board-available, machine-ready and transfer signals. The timing one is the transfer time, which has to fit within the cycle of the slowest station or the line accumulates boards in the wrong places.
Test the handshake with real boards rather than by inspecting wires. Run boards from the upstream station through the machine to the downstream station, and watch for the failure modes that appear only in motion: a board that is released before the machine is ready, a transfer that double-feeds, a sensor that misses a dark board. Repeat with the thinnest and the thickest board in the product range, because sensor sensitivity and conveyor gap are set for one and not necessarily for the other.
Record the settings that made it work: conveyor width for each product, sensor positions, transfer speed and the handshake polarity. These settings belong in the parameter baseline described earlier, and they should be repeated in the training notes, because the first product change after acceptance is when they get disturbed. A line that has to be re-tuned after every changeover was never properly documented at acceptance.
Throughput should be accepted as a cycle time measured over a run, not as a rate quoted from a specification. Time thirty consecutive boards from the moment the first enters to the moment the last leaves, compute the median and the ninety-fifth percentile, and convert both to boards per hour. The median tells you the capability; the ninety-fifth percentile tells you the practical output, because a line is limited by its slow boards as much as by its fast ones.
Compare the measured figure with the promised figure using the same arithmetic on both sides. A machine published at a headline rate, such as the classes described in how a rated rate turns into boards per hour on your line, delivers less on a real BOM because of feeder indexing, transfer time, vision overhead and changeover. If the promised effective rate assumed a mix factor of 0.82 and the measured result corresponds to 0.71, the difference is a finding for the punch list rather than a rounding error.
Do the conversion in front of both parties and write the factors down. The calculation is not contentious when the inputs are visible: placements per board, effective rate, transfer time, availability allowance. What is contentious is a single number presented without its inputs, which is why the report should carry the arithmetic rather than the conclusion.
Training is the item most often satisfied by a signature and least often by capability. A session measured in hours can be delivered entirely as demonstration, with the trainee watching and nodding. A session measured in tasks produces a record of what each person can actually do. The tasks worth listing are the ones that occur weekly: building a program, loading and threading a feeder, changing a nozzle, recovering from a pick error, reading the alarm log, and performing the daily checks the manual specifies.
A practical format is a short checklist per role, signed by the trainee and the trainer, with the date. Operators, technicians and process engineers have different lists. The operator list covers running, loading and recovery. The technician list covers nozzle changes, feeder repair, conveyor adjustment and the daily and weekly maintenance items. The process engineer list covers program structure, recognition teaching, parameter export and the interpretation of placement data.
The signature matters less than the demonstration, so build the demonstration into the acceptance schedule: each trainee performs the task once, unaided, while somebody records the result. Tasks that were not performed become training items on the punch list with a date, which is a far more useful outcome than a certificate that says four people attended for six hours.
Every acceptance produces open items, and the only question is whether they are classified before the team leaves. A three-level classification works because it maps directly to money. An A item prevents production or breaches a safety or contractual requirement, and it must be closed before acceptance is signed. A B item degrades performance or reliability but does not stop production, and it is closed within an agreed window with the cost borne by the supplier. A C item is cosmetic or documentary and can be closed at leisure.
Each item needs four fields: what is wrong, what evidence shows it, who owns it, and by when. An item without an owner is a note. An item without a date is a wish. The classification also determines the payment conversation, because retention is normally released against A items only, with B items tracked separately and C items recorded for the file.
Class | Examples | Close-out window | Effect on acceptance |
|---|---|---|---|
A | Safety interlock not working; option missing from the build; capability below the agreed index; handshake failing | Before signature | Blocks acceptance and final payment |
B | One feeder position intermittent; recognition marginal on one part; cycle time above the agreed band | Agreed days to weeks | Acceptance signed with retention held |
C | Document missing; label incorrect; cosmetic damage; spare parts list incomplete | Agreed weeks | Recorded, no payment effect |
A failed test is not the end of a purchase; it is a negotiation that the contract should have anticipated. The difference between a failure that is resolved in a week and one that costs a quarter is whether the remedies were agreed before the test, in the order they should be applied, with the payment consequences stated. Buyers who negotiate remedies after a failure negotiate from weakness, because the machine is already on their floor and the schedule is already committed.
The sequence below is deliberately ordered from cheapest to most drastic, and the order matters: a supplier asked first for a price reduction will offer one, and a price reduction does not fix a machine that cannot hold capability. Working up the ladder keeps the objective, which is a machine that meets the requirement, in front of the alternative, which is being compensated for one that does not. It is also the part of a JUKI SMT machine FAT SAT checklist that is best written while everyone is still friendly.
The first rung is rework in place: an adjustment, a part, a reflash, a recalibration. Most failures end here, and it is the correct first request because it preserves the schedule and keeps the liability with the supplier. The second rung is replacement of the specific assembly that failed, which is appropriate when a component is defective rather than misadjusted, and the third is replacement of the machine, which is rare and usually reserved for identity or configuration failures that cannot be corrected on site.
The fourth rung is a price reduction, and it belongs low on the ladder for a reason. A reduction is appropriate for a B item that genuinely does not affect production, such as a slower changeover than promised. It is inappropriate for a capability shortfall, because accepting money for a machine that cannot hold the index transfers the problem to the plant permanently and converts a supplier obligation into a production constraint. The fifth rung, rejection, is the last resort and depends entirely on the payment structure: a buyer who has paid in full has no practical rejection right.
Where the shortfall concerns throughput rather than accuracy, the comparison should be made on the same basis as the purchase decision, because the class of machine behind the quoted figures, such as the high-rate rotary class quoted for this project, is sold on an optimum figure that no production board reproduces. The remedy then becomes a renegotiation of the promised effective rate rather than a rejection, and the arithmetic in the acceptance report is what makes that conversation possible.
Retention is the mechanism that makes the punch list meaningful. A typical structure for an equipment purchase links payment to three events: a deposit on order, a payment on shipment or delivery, and the balance on signed acceptance, with a small percentage held for a defined period after acceptance to cover B items. The exact percentages are commercial, but the principle is not: some money must remain unpaid while obligations remain open.
The acceptance report should state which items release which payment. A items block the balance. B items release it with retention held until closure. C items do not affect payment but are recorded. Without that mapping, the report becomes a list of complaints and the payment conversation happens separately, usually with less information and more emotion. With it, the conversation is about a document both sides have already signed.
A retention period of thirty to ninety days is common and it should have an end date written down, because an open-ended retention becomes a permanent discount. The period exists to expose the faults that only appear in production, and the evidence for closing it is the same kind of data collected at acceptance: pick rates, stoppages, capability checks on the parts that matter.
Close-out is a document set rather than a meeting. It contains the signed acceptance report, the punch list with its closed and open items, the parameter baseline export, the program archive, the licence documents, the calibration record, the training records by role, and the spare parts list with lead times. Each of those items appeared earlier in this checklist, and close-out is simply the moment when the collection is complete and stored somewhere the plant can find it in two years.
Two signatures belong on the report: the buyer's authorised representative and the supplier's. A report signed only by the buyer is a note; one signed by both is an agreement. Where an A item remains open, the report should say so explicitly and state what happens if it is not closed by the agreed date — normally that the retention is released to the buyer or that the supplier bears the cost of an additional visit.
The storage decision matters more than it appears. Files kept on the engineer's laptop leave the plant when the engineer does. Files stored in the maintenance system with the machine's serial as the key are still there when the machine is audited, resold or transferred to another line. Ten minutes at close-out decides which of those two futures applies.
A machine that passes acceptance can still disappoint in production, and the thirty-day review is where that becomes visible while the supplier still has an incentive to act. The review compares the acceptance baseline with the first month of real running: pick rate, stoppage hours by cause, capability checks on the critical parts, changeover time, and the number of interventions that required engineering rather than operator attention. The comparison is only possible if the baseline was recorded, which is the practical argument for doing it at acceptance.
The review has three possible outcomes, and all three are useful. The machine is performing as accepted, in which case the retention is released and the punch list is closed. The machine is underperforming because of an unresolved B item, in which case the item is escalated with data attached. The machine is underperforming because the process around it changed, which is not the machine's fault but is worth knowing, because otherwise the equipment gets blamed for a printer, a profile or a program that moved.
Hold the review as a short meeting with the log in front of you rather than as a conversation about impressions. Thirty days of data settles arguments that a year of opinion cannot, and it gives the supplier a fair basis to respond, which tends to produce faster action than escalation. It also produces the internal record that justifies the next machine purchase, which is a benefit the equipment buyer collects two years later.
Remedy | When it applies | Who bears the cost | Schedule effect |
|---|---|---|---|
Rework in place | Adjustment, part, reflash, recalibration | Supplier | Hours to days |
Assembly replacement | Defective head, feeder bank, conveyor module | Supplier | Days to weeks |
Machine replacement | Identity or configuration that cannot be corrected | Supplier | Weeks |
Price reduction | B items that do not affect production | Supplier, as credit | None |
Rejection | Requirement not met and no remedy agreed | Per contract | Restart |
Write the two test plans before the machine is built
If you send us the BOM and panel data for the products the machine will run, the accuracy your process actually needs, the feeder and nozzle set you intend to load, and the line the machine will join, we will turn them into a factory test plan and a site test plan with the criteria already written: which boards travel to the factory, what is measured on each, how many points constitute a sample, which parameters are captured as a baseline, and what closes an A item versus a B item. We will mark which figures come from the product pages on this site, which come from the manufacturer's published specification and which are our own planning assumptions, so the criteria can be defended to production and to quality rather than argued at the end of a long day. A test plan written before the build costs a few hours; the same discussion held after a failed acceptance costs a quarter, and the machine that fails is usually the one nobody defined a pass criterion for.
The factory acceptance test verifies that the machine was built and configured as ordered; the site acceptance test verifies that it works in your line with your boards, feeders and operators. FAT is a build check and SAT is an integration check, and they fail differently: a FAT failure is usually corrected with a part or a reflash, while a SAT failure costs a return visit and a stopped line. Both need pass criteria written in advance, and SAT should not be the first time anyone measures anything.
Enough to estimate the mean and the spread, not enough to consume the schedule. Thirty boards with twenty measured placements each gives six hundred points and estimates the mean offset to roughly half a micron when the spread is around ten microns. For a forgiving product, sixteen boards with ten points each is acceptable; for fine pitch, go higher. Fix the number before the run, state it in the report, and sample across the panel rather than from one area.
The configuration reconciliation, the measured placement data with the calculation, the pick rate with failures by feeder position, the programme and parameter baseline, the licence and calibration documents, the training records by role, and the punch list with owners and dates. It should also state the software version and what was not tested. A report that records only that the machine ran cannot support a warranty claim, an audit or a resale, and it cannot be reconstructed later without repeating the acceptance.
At minimum a process engineer who understands the requirement, a maintenance technician who will live with the mechanics, and somebody with authority to sign. Production and quality join at the site test. Sending nobody to the factory and treating the site visit as the whole acceptance removes the cheapest opportunity to correct a build error, because by then the machine has travelled and the schedule is committed. Where travel is expensive, split the roles between the two gates and brief the second pair.
Work up the remedies ladder in order: rework in place, replace the failed assembly, replace the machine, then price reduction, then rejection. Most failures end at the first rung. Price reduction belongs low on the ladder because it does not fix a capability shortfall and converts a supplier obligation into a permanent production constraint. The outcome depends on the payment structure, which is why retention held against acceptance is worth more than any clause about good faith.
Two to four days for a single machine: one day for installation, utilities and levelling, one or two days for programs, runs and measurements, and a half day for integration, training and the punch list. Compressing it into one day is possible and removes the part of the test that finds intermittent faults, because intermittent faults need repetition and repetition needs time. Schedule the integration and training half day explicitly rather than leaving it to the last afternoon.
The structure is the same and the evidence standard changes, because some parameters have already drifted and some documents do not exist. Add a serial and build-sheet reconciliation, a witnessed capability run on your own board in place of a factory certificate, a serviceable feeder count by width, and a licence transfer condition. Price the recalibration into the offer rather than discovering it afterwards, and link the balance to a signed report instead of to delivery.
