Commercial
A factory visit tells you more in an hour than a questionnaire tells you in a month — if you look at the right eight things. This is the checklist we use on our own suppliers, written so you can print it and take it into the workshop.
Almost every factory tour ends at a sample wall. Polished edges, a mirror-finish display stand, a beautifully bent shroud, all lit from the right angle. That wall proves exactly one thing: this plant is capable of making one good part. It does not prove the thing that actually matters: that every batch is as good as that one part.
The parts on the wall were selected. What happened to the rest? Reworked, scrapped, or downgraded to a customer who did not measure? The wall cannot answer that. Only records can.
So the audit reduces to three questions: is the equipment what they say it is, are the records actually being filled in, and do problems actually close. If all three hold, the sample wall means something. If any of them fails, the prettier the wall the more you should worry.
What follows is the walk order we use. Eight stations, three observations at each. You do not need to be an acrylic processor to run it — you need to know what evidence looks like.
| Station | What to look at | How to look | What good looks like |
|---|---|---|---|
| 1. Machine nameplates | Router travels and count, laser wattage, oven internal size, polishing equipment | Do not read the equipment list — walk to each machine and photograph the nameplate | Nameplate model matches what sales claimed; machine count supports the capacity promised |
| 2. Sheet storage | How sheet is stacked, condition of the masking film, whether sheets are separated | Look at whether sheet stands upright or lies flat, and whether the saw area has debris piled around it | Upright or compartmented with dividers; faces unmarked; offcuts segregated |
| 3. CNC and laser | Tool management, coolant, extraction, where the first-off part sits | Look for tool numbers and wear marks; look for a first-off part next to the machine | Tools numbered and changed on life; first-off stored with the drawing and measured values written on it |
| 4. Bending and annealing | Oven controller, annealing cabinet, formers | Ask one specific question: how do you ramp a 20 mm cast sheet to an R20 inside corner? | They can quote ramp rate, soak time and cooling rate, and explain why cooling is slow |
| 5. Bonding bench | Adhesive brand and batch, clamping jigs, hygrometer | Look for a hygrometer, and for numbered clamps | Adhesive in date with batch logged; temperature and humidity recorded |
| 6. Inspection and retention | First-article sheet, patrol records, sample cabinet, calibration labels | Pick a batch number at random and ask for every record on it within ten minutes | It arrives in ten minutes, and the numbers are measured values, not ticks |
| 7. Packing area | Interior cushioning, carton spec, fragile marks, stacking method | Watch what is being packed right now, not the display sample | PE film plus foam plus rigid carton; corner protectors fitted |
| 8. Corrective action | Three months of complaints, rework slips, CAPA records | Pick one complaint and ask how many days from detection to closure, and what changed | A named owner and a closure date; the same defect has not recurred in three months |
Walk them in this order rather than following the route you are offered. Station 4 and station 6 are where most visits are steered away from, and they are the two that carry the most information.
The equipment list in the sales deck is usually the group's list, or a list from three years ago. What is running in the workshop may be different machines entirely. Walk up to each one and photograph the plate: model, travel, wattage, year.
Three numbers matter more than the rest. Router travel decides whether they can take your largest part without tiling it. Laser wattage decides how thick a sheet they can cut — 80 W and 150 W are not the same category of machine, and the difference shows up as charred edges and slow feed rates rather than a clean cut. Oven internal dimensions set the ceiling on bent parts: a 1,200 mm oven cannot produce a 1,500 mm shroud no matter what the catalogue says.
If a plant has three routers but claims ten production lines, that is not necessarily a lie — a line is a process chain, not a machine. But ask them to draw the chain for the specific job you are placing, and see whether the numbers survive the drawing.
Sheet storage is the fastest read on a plant. Sheets stacked flat with no interleaving will have pressure marks on the bottom layers, and those marks do not polish out. The correct practice is upright storage, or compartmented flat stacking with rigid dividers every few sheets. If the masking film has been peeled back across large areas, expect handling scratches.
In the CNC area, look at two things: tool control and where the first-off part is. Tools are consumables; whether they are numbered, whether they are changed on hours cut, and whether worn tools have a separate bin — these three decide whether the same drawing produces the same surface finish across batches. A first-off part sitting next to the machine, with the drawing beside it and measured values pencilled on, tells you first-article inspection is real. A first-off part in the office tells you it is paperwork.
Laser cutting deserves its own look. A clean laser edge on acrylic should be close to polished. If the edges are brown or hazy on clear sheet, either the power is too high, the feed is too slow, or the assist gas is wrong — and the same settings will bite you on every subsequent batch.
Bending and bonding are the two operations most dependent on operator experience, and therefore the two most likely to fail silently. The simplest way to audit them is to ask one concrete question and listen to whether the answer is a procedure or a story.
Try: “For a 20 mm cast sheet bent to an R20 inside corner, what curve do you run on the oven?” A controlled answer gives you three stages — ramp rate, soak time, cooling rate — and explains why the cooling is slow: to avoid residual stress that cracks the part weeks later. An uncontrolled answer is “our master has done this for fifteen years, he watches the colour.”
Experience is real and valuable. But experience that has not been written down as parameters cannot be replicated on a Tuesday when the master is on leave, and it cannot be transferred to your reorder six months later.
Then look for the annealing cabinet — not the bending oven, a separate programmable oven used for stress relief — and for a recent log sheet. Annealing is the step most often skipped because it produces no visible improvement on the day. It produces its result three months later, by not cracking.
Three items decide bonding quality: adhesive brand and batch, clamping jigs, and a hygrometer.
Solvent cements for acrylic have a shelf life. Cement past its date produces joints that bloom white a few weeks later, and no amount of polishing fixes it because the change is inside the material. Batch logging is the only way to trace that later.
Clamping decides seam flatness, especially on long joints and large panels. Numbered clamps with a defined clamp time are a controlled process; a pile of spring clamps and a feeling is not.
The hygrometer is the tell. Below about 20 °C the solvent evaporates too slowly and the joint hazes; above about 60 % RH the seam can fog. A bench with no hygrometer is telling you that bonding quality depends entirely on how the operator feels that day.
This is the most important part of the visit and the easiest to stage. The technique is simple: pick a batch number at random and ask for the whole file on the spot.
What should come back: the sheet batch and supplier, the first-article sheet with measured values, in-process inspection records, final inspection, and the packing record. Ask for it in ten minutes, without warning. If it arrives, the system is alive. If the answer is “we'll tidy it up and send it tomorrow”, it needs to be written rather than retrieved.
Then read one detail: numbers or ticks. A live record carries measured values — 38.12, 38.08, 38.15. A decorative record carries the word “pass” and a tick. The first can be investigated when something goes wrong. The second can only be argued with.
| Record field | What it should contain | Why it is not optional |
|---|---|---|
| Drawing number and revision | Controlled drawing ID plus revision | Building to a superseded revision is the single most common error |
| Sheet batch | Supplier batch plus internal receiving batch | Colour, transmission and stress all follow the batch |
| Machine and operator | Machine ID plus name | Two machines of the same model can behave differently |
| Tool ID and hours | Tool number, diameter, edge form, hours cut | Tool wear is the leading cause of edge quality drift |
| Spindle, feed, depth | Actual set values, not recommended ones | The same drawing at different parameters gives a different edge |
| Workholding | Vacuum, clamps or jig number | Clamping distortion is invisible once the part is finished |
| Critical dimensions | Three measured points plus gauge ID | One point proves a part; three points reveal drift |
| Sign-off and time | Operator, inspector, timestamp | Without a time the record cannot be sequenced |
Three measured points rather than one is the detail worth arguing for. One point tells you the part passed. Three points in sequence tell you whether the process is drifting: 38.12 → 38.16 → 38.21 is all inside tolerance and still a warning that the tool is wearing or the shop is warming up.
Ask to see a job that is in production right now, ask which sheet batch it is from, then ask for that batch's certificate. A plant with real traceability answers in minutes. A plant without it will explain why that particular job is complicated.
Traceability is what makes a recall possible, a warranty defensible, and a replacement part match the original two years later. It is also the easiest thing to fake with a folder of certificates that belong to nothing in particular — which is why the test is a live job rather than a document.
A related question with a high hit rate: “If I reorder this in eight months, how do you make it match?” The good answer describes a retained sample, a parameter card and a batch record. The weak answer is “we'll use the same material”, which is not a control, because the same grade from the same supplier in a different batch is a different colour.
An ISO 9001 certificate proves that a quality management system has been audited against a standard. It does not prove that a particular part was made correctly, and it does not audit your product. Treat it as evidence that the paperwork discipline exists, not as evidence about your order.
Two details are worth checking on the certificate itself: the scope statement and the validity dates. The scope tells you which processes are covered — many certificates cover “manufacture of acrylic products” while laser cutting, annealing or printing are subcontracted and therefore outside the audited boundary.
Material reports are a separate category. A RoHS report on cast sheet is a statement about the material the supplier bought, not about the part they made. If your compliance file needs a part-level statement, ask what would be tested, by whom, at what cost, and how long it takes — a supplier who answers that precisely is more useful than one who says “everything is certified”.
The failure mode to watch for is a supplier who cannot distinguish between these categories. It usually means the documents are being used as decoration rather than as evidence.
Almost every fabricator subcontracts something — plating, large-format printing, sometimes bending beyond their oven size. That is normal. What matters is whether they tell you, because the subcontractor is where your delivery date and your quality go to die quietly.
Three practical checks. First, count the process: if they print in-house, there should be a print area with racks, inks and a curing setup. Second, ask what they send out and write the answer into the quotation — a supplier who names their subcontractors is a supplier who has thought about the risk. Third, ask what happens to your delivery date when the subcontractor is late, and see whether the answer includes a buffer or an excuse.
For anything with a compliance file — medical, automotive, food contact — ask specifically whether the subcontracted process is inside or outside their ISO scope. That single question sorts out most of the confusion later.
If you cannot travel, a video walk-through is worth doing, but be clear about its limits. Video establishes presence; it does not establish constancy. It can show you a nameplate, a retention cabinet, and what is being packed today. It cannot show you what the records looked like three months ago, or how the line behaves during a rush.
Three rules make a video audit worth the hour. Send the checklist first, so they walk your route rather than their showroom route. Ask for the camera on the nameplate and the record book, not a wide shot — wide shots can be staged. Pick the batch number during the call, so the part has to be fetched rather than arranged.
A useful addition: ask them to walk to a machine you did not ask about, chosen at random from the list they sent you. Unrehearsed corners are where the real state of the plant shows.
Score each station 0–2 and add up. The total is less interesting than where the zeros fall.
Two zeros at stations 4–6 (bending, bonding, records) matter far more than a couple of zeros at stations 1–2. Equipment can be rented; process control cannot. A plant scoring 10/16 with both zeros in records is a riskier partner than one scoring 8/16 with its zeros in machine age.
And the score is not a verdict on the relationship. It is a map of where you need to build contractual protection: incoming inspection, retained samples, staged payments, or a pilot batch before the full order.
It does not tell you whether they can hold your tolerance on your geometry. That is a DFM question, answered by sending a drawing and asking for a written response on which features are feasible and which are not.
It does not tell you how they behave when something goes wrong. Complaint handling is a separate audit: read the log, pick one, and ask what changed. A plant with an honest log and a boring closure is worth more than one with no log at all.
And it does not replace a pilot. Whatever the visit shows, the cheapest way to find out whether a fabricator suits you is one prototype and one small batch, with the acceptance criteria written down before either is ordered.
The most useful thing you can do at the end of the visit is spend half an hour on a single question: “Walk me through how you would make this part.” Hand them one of your own components. Ask what sheet, how many operations, which step is most likely to go wrong, and what they would do if it did. That conversation carries more information than the previous eight stations combined, because it cannot be prepared in advance.
Half a day. Two and a half hours on the floor, one hour on records, half an hour on the walk-through of your own part. A one-hour visit shows you the shop floor and nothing about control.
Announced, but without time to stage it. Tell them the day before, do not fix a route, and do not allow the workshop to be tidied. Fully unannounced visits are rarely practical and rarely necessary.
For a first order, no. Video establishes presence, a site visit establishes constancy. After two or three orders, video is adequate for re-qualification.
Depends which records. A missing packing record is a process gap that can be closed. Missing first-article measurements or batch traceability is a system gap that takes months to fix.
Redacted batch records and parameter cards, yes. Customer names and drawing details are masked — that is the industry norm, and we apply the same rule to our own customers' files.
“Show me the records for this batch, now.” Everything else can be answered from memory. That one cannot.
We will have the relevant machines running, the certificates on the desk and a job in progress to trace.
