Write the pass condition before anyone reaches for a switch
The usual weakness in a supplier demonstration is that nobody has written down what a pass means. Put one line on paper for every measurement you intend to rely on: what is measured, on which material, in which machine setting, and what answer you will accept. A line can look excellent across the first pieces you load, and then drift once the feed, the tooling temperature and the hand-off between operators have all been running for four hours. The membrane electrode hot press bonding machine page states a fitting error, a working temperature, a press pressure and a production cycle. Treat those as the fields to have measured rather than as an answer you can copy, and ask for the same fields recorded on your stack during your trial.
Record what the trial stack was made of
Test data collected on a supplier's demonstration material transfers badly to production material, so log the stack itself: how many layers were bonded, which film or electrode material each layer was, what the incoming tolerance on each layer was, how the parts arrived (tray, reel, magazine) and who placed them. A record that does not say what went in cannot support a decision about what comes out. If you also plan to run a second material later, decide now whether that run happens in the same session, because a comparison done on two different days is really two tests.
This is also where you record what the trial did not cover. If a stack type you will build was not run, write that sentence down rather than leaving the gap for someone to fill in later with an assumption.
Measure alignment where it actually costs money
Stacking accuracy is usually reported as a single fitting figure, and that figure is only useful if you know where on the part it was taken and how often it was taken. Ask for the reading at each position the machine is meant to hold, plus a record of the alignment sequence in use, since the same line can sit differently depending on which recipe is loaded. The membrane electrode vacuum hot pressing machine listing sets its fitting figure against a working vacuum figure, so log both together: alignment measured without the process condition is half a number.
A useful alignment record shows the spread across consecutive pieces, not the single best reading. Log every piece in a run and keep the run order intact, so you can see whether the line holds its figure or wanders and gets corrected.
Log the process settings as they ran, not as they were set
There is a difference between the recipe an operator typed in and what the machine held during the cycle, so capture both for every trial run: the recipe name and its stored values, then the values actually read during the cycle. Temperature, pressure and dwell are the obvious ones. Where a machine holds a tool at a working temperature, note whether that figure is the set point or the reading at the tool face, since the two are not the same statement. For a line that bonds under reduced pressure, note the pressure you were told to expect, then the figure the process actually held. Ask the supplier to hand over the recipe file with the names they will use after delivery, so your records and their records describe the same condition.
Time the line the way your production day will time it
The production cycle on a listing is a rate the machine can support. Your number is a rate that survives changeover, jam clearing, material replenishment and shift handover, so time four things separately: a single cycle with an attentive operator, a run of consecutive cycles, a changeover between two stack types, and the recovery after a stop. Then write down which figure you will use in your plan and why. If the line is meant to sit inside a wider cell, measure the hand-off to the station before and the station after, because the cell rate belongs to the slowest step and not to the bonding press.
Ask what the supplier recorded for availability during their own run, and how they defined it, since two factories can count the same hour differently. Any figure that changes when the layer count changes should be recorded for each layer count you tested, which matters most if your production mix moves between a five-layer and a seven-layer stack.
Keep the defect records together with the settings that produced them
A stack that bonded badly is a paid-for data point, so photograph it, keep it, and attach the settings that were running at the time. The defect list for this kind of work is short but the causes are not, so capture the observable symptom: a mis-aligned layer, a lift at the edge, a trapped bubble, a wrinkle in a film, a crushed or thinned zone, a piece that left the tool in a different shape than it entered. For each one, record which step you believe caused it and what was changed. Keeping the bad pieces with their settings is what lets a later conversation with the supplier be about the process rather than about opinions.
Test the line as an automatic line, which means testing the stop
An automatic bonding line has two behaviours you need evidence for, and most demonstrations test only the first. What it does when everything goes well is the easy half. What it does when something goes wrong is the expensive half. Run these deliberately: feed one part out of position, feed one part missing, stop the feed for a few cycles, ask the machine to change recipe mid-run, open a guard during a cycle, cut the control power mid-cycle and restart. For each one record what the machine detected, what it did with the part already inside, whether an operator had to intervene to clear it, and how long the recovery took. Ask how the interlocks behave while a recipe is loaded, and whether the machine can be started again by the person who runs it or needs an engineer with a password.
Collect the machine's own records and see whether you can read them
Most lines of this type store recipe data, cycle data and fault history, and the value of that data depends entirely on whether your team can get it out and read it. Ask during the test, in front of the machine: which records does the control keep, for how long, in what format, and what happens to them when the power is off or a recipe is edited. Have your own engineer export one shift's worth of records and check them against the run they just watched. If your plan depends on line data feeding a quality or maintenance record, prove the export path during the test rather than after installation. The semiconductor process equipment range on the supplier's own site covers alignment and bonding work where the control layer matters as much as the mechanics, so judge the machine on both.
Record the people, not only the hardware
Have the operator who will actually run the line do the loading during the trial, and write down what they had to be told. Ask them afterwards, without the salesperson present, what felt slow, what they were afraid of touching and what they would change. A supplier engineer running their own machine tells you the machine's ceiling; your operator running it tells you your floor. Record how long it took a new person to load unaided, and whether the setup steps live in one document or in the head of whoever built the machine.
Write the acceptance page while the test is still warm
The point of the whole exercise is a short document that says what was measured, what came back, and what still needs proving. One page per measurement: the trial condition, the figure, the pass or fail judgement, who judged it, and a signature from both sides. Any measurement that failed is worth more on paper than in a verbal promise that it will be fixed, because the paper version travels into the contract. Where the listing's figures were the reason you asked for the test at all, put the listing figure and the measured figure side by side on that page, so nobody argues afterwards about which number was promised.
Two cautions to keep on the same page. Listing figures belong to the build the supplier describes, and a line built to order can differ, so treat every printed number on a product page, including the fitting, temperature, pressure and cycle figures on the bonding listings, as a claim to be measured rather than a fact to be assumed. And a factory test does not prove your site: power quality, air supply, ambient cleanliness, the material you will actually buy and the shift pattern you will run are all outside the test. If the equipment is part of a automated production line solutions scope rather than one stand-alone press, then the interfaces have to be tested too: what the line expects from the station upstream, what it hands downstream, which signals are exchanged, and who is at fault when a stop on one side stalls the other.
The short list to take to the test floor
Take these eleven items as your data sheet and you will come back with something usable: the written pass condition; the trial stack description with incoming tolerances; alignment readings at every position the machine claims to hold, per piece, in run order; the recipe file with its stored values plus the values read during the cycle; four timed figures covering cycle, consecutive run, changeover and recovery; a defect log with photographs and the settings attached; the fault-injection results including what happened to the part already in the tool; a control-data export produced by your own engineer; the operator's own notes; the signature page with listing figures and measured figures side by side; and one list of everything the test did not cover. The last one protects you more than the others. What you have not tested does not disappear; it simply moves into your first month of production, where it is far more expensive to fix.
