Write the target around the measured condition, not just around the value
A bare tolerance line — alignment accuracy within a stated deviation — describes an ideal case. To make it verifiable, the document has to state four things alongside the number: the measurement interface, the axis pair the deviation is read against, the handling condition of the part, and the method used to read it.
For a glass panel bonded to PI film, the interface changes what counts as acceptable. A panel corner can sit inside tolerance while a bonded edge drifts, if the reference is the panel outline rather than the alignment mark. Naming the mark pair, and naming the surface the deviation is measured on, removes that ambiguity. It also fixes a second problem: two measurements of the same cell taken at different points in the cycle can legitimately differ, because the part is held differently before and after the chamber is pumped down.
Separate the three numbers that get merged into one claim
In a glass-to-film alignment cell, three distinct figures are commonly compressed into a single specification sentence. They are verified differently, and they fail differently.
- Positioning capability — how the placement stage performs when the machine positions the parts, without an operator or a load on the feeding side.
- Process repeatability — how much the result varies when the same cell type is processed again, under the same recipe.
- Line yield — the share of cells that pass downstream inspection, which absorbs handling loss, film variation and contamination as well as placement error.
Write each one separately, with its own test condition. When they are merged, a supplier defending a positioning figure and a buyer tracking scrapped cells are arguing about different measurements. Once the three lines are separate, disagreements narrow to the condition attached to each line, which is a question that can be settled.
Record the equipment's own capability before signing off on yield
The TGV alignment packaging machine page from Shenzhen PowerDe Automation Precision Equipment is a useful reference for how much detail a specification should carry. The listed equipment combines manual feeding with camera photography for positioning on both the glass panel and the PI film, movement on linear motor modules, holding on electromagnetic and electrostatic ceramic platforms, and an X, Y and θ adjustment axis. The page states alignment accuracy of ±0.03 mm, with a stated throughput figure for glass vacuum bonding where one product requires four bonding operations.
Read that specification as a buyer drafting acceptance criteria and several commitments follow. Manual feeding means the document has to state who loads the part and how repeat placement is judged, because operator technique moves the measured result. Camera positioning means the record must say when the camera is triggered — the page notes alignment occurs once the negative pressure reaches the set value, so acceptance photos taken at a different point in the cycle are not comparable. A stated throughput figure based on four bonding operations per product only translates into line throughput once the number of operators and the feeding interval are fixed; quoting the machine figure as a line figure is where capacity promises usually break.
Keep process windows inside the target, not outside it
Two features on that page carry process windows rather than single values. The bonding sequence — adsorption platform pressing, pre-bonding, vacuum pumping, then bonding — is a sequence whose timing affects the bonded result. The selectable program formulas exist so the machine can be matched to different product processes; the same formula will not carry two different film thicknesses at the same accuracy.
Both belong inside the target document, with the accepted range written down: which formula is the reference recipe, which parameters in it are fixed, and which may be adjusted when the incoming film changes. An automated production line solution inherits these windows from each station it contains. A line document that omits them cannot be verified station by station, and a change made at one station goes unnoticed until downstream inspection rejects the cells.
The same applies to the data the machine produces. The page describes a vision system with a measurement function whose data can generate local marks or networked data, and MES data connection with on-board storage. Those are the instruments the target document should name as the acceptance record: which dataset, at which sampling rate, retained for how long. If the record is not agreed in advance, the acceptance meeting has no shared evidence to work from.
A checklist to hand the supplier with the enquiry
Before a specification is signed, each of the following should have a written answer attached to it. If a line is blank, the target is not yet a target.
- Panel size and film size the accuracy figure applies to, plus the number of bonding operations per cell.
- The alignment mark or edge used as reference, and the surface the deviation is read on.
- Feeding method, and who performs it during acceptance testing.
- Trigger condition for positioning — pressure, sensor or sequence step.
- Holding method, and whether it changes between pre-bonding and final bonding.
- Reference recipe, with the parameters that are fixed and the parameters that may be adjusted.
- Test sample count, sample source, and the reject definition used for yield.
- Which measurement system produces the acceptance record, and how that record is stored.
Send that checklist with the enquiry rather than after the quotation. Suppliers answer specification questions more precisely when the questions are already on paper, and the answers come back in a form that can be compared across quotations. Where the required panel or film dimensions fall outside what a listed machine covers, the difference should be settled in writing before the build. Product development work of that kind is scoped against the documented target, which is another reason to write the target properly the first time.
The practical test of a performance target document is simple. Hand it to an engineer who was not in the negotiation and ask them to run the acceptance test from it alone. If they need to phone someone to interpret a line, that line is not finished yet.
