1. Machine identity and the exact recipe under test
This is the least glamorous section and the one most often missing. Before any process number means anything, the record has to say which configuration produced it:
- Model, serial number, controller and firmware versions, and the software build used for the run
- Recipe or program name, revision, and every parameter setpoint with its tolerance band
- Configuration options fitted to that specific unit — alignment cameras, roller type, chamber size, vacuum pump package
- Software build and test-firmware versions for any instrument used for measurement, not only for the machine
Recipes get edited. On a line where several engineers share one machine, a result recorded without the recipe revision cannot be reproduced, and the same set of numbers may correspond to two different processes.
2. Process parameters and their tolerances, logged as curves
Averaged values hide the events that matter. For bonding, lamination and hot-pressing equipment specifically, collect:
- Chamber or plate temperature over time, including ramp rate and soak duration, not just the setpoint
- Pressure profile — press force, dwell time, and the actual time to reach vacuum, plus leak-up rate at the end of the cycle
- Cycle time broken into stages, so a slow cycle can be traced to the stage that caused it
- Alarm and interlock events with their timestamps
Where the machine publishes limits, record the actual values against them. Our high-pressure debubbling machine listing, for example, states a tank of 1000 × 1200 mm in SUS304 stainless steel; the test record should show the pressure and temperature actually reached in that tank, along with how the pressure was measured and by which gauge.
3. Material and consumable traceability
Process equipment does not run in isolation, and a defect that looks like a machine problem often starts upstream. For each run, log:
- Lot and batch numbers of adhesives, optical films, protective films, glass and PI film
- Storage time before use and, for temperature- or shelf-life-sensitive materials, the time out of refrigeration
- Substrate dimensions, thickness and incoming inspection result
- Consumable life counters: roller, seal, vacuum pump oil and filter hours since replacement
A defect rate that jumps when the film lot changes is a materials conversation. Without the lot number in the record, it reads as an unexplained machine fault and the machine gets stripped down for nothing.
4. Environment, utilities and infrastructure
Results move with the room. A machine qualified in one facility can under-perform in another, and only the environmental record explains why:
- Cleanroom class, plus particle counts at representative points during the run
- Ambient temperature and relative humidity over the run period
- Compressed-air pressure and dew point; cooling-water temperature and flow where applicable
- Supply voltage and frequency, recorded at the machine terminal rather than at the panel
- Nitrogen or other process-gas purity and flow for equipment that uses them
Our 3D curved glass hot bending forming machine listing includes oxygen monitoring in four nitrogen-filled chambers. On equipment like this, oxygen concentration is not an optional extra reading — it is part of the process record, and the data log should show it against each cycle.
5. Quality output and defect data at panel level
This is where most test programmes stop too early. A yield percentage on its own cannot be improved. Capture:
- Defect type, count and physical location on the panel or sheet, tied back to the cycle that produced it
- Alignment accuracy and fitting error against the machine's stated specification
- Bubble, particle and delamination counts, with size classification
- Images of representative defects, kept with the cycle record
- Operator name, station, shift, and the inspection method or instrument used for each check
The value of panel-level location data shows up quickly: a defect pattern that repeats in the same corner of every panel points at tooling or a roller, while scattered defects point at materials or environment. A total defect count cannot tell you that.
6. Reliability, endurance and failure records
Endurance data is what turns a delivered machine into a predictable one. Track:
- Uptime and availability, with the definition used for each term stated in the report
- Mean time between failures and mean time to repair, separated by fault category
- Drift in key parameters over the endurance run: vacuum level, temperature uniformity, alignment accuracy
- Every stoppage, including short ones, with cause and action taken
- Wear-part condition at defined intervals, with photographs
Published figures should always be tied to the test that produced them. The listing for our 3D surface vacuum bonding machine gives uptime of 98% or better and a 20-second cycle from a 5.5-inch soft-to-hard test. Those numbers belong to that specific test configuration — same product family, different substrate and material stack, and the cycle time will differ.
7. Maintenance, calibration and change history
A test result measured with an out-of-calibration gauge is not a measurement. Keep alongside the run data:
- Calibration certificates and due dates for all measuring instruments, including the machine's own sensors
- PM schedule adherence — planned versus actual dates
- Spare-part and consumable changes, with part numbers and lot codes
- Any recipe edit or hardware change, with who made it, when, and why
8. Documentation package for acceptance
For equipment bought for production rather than for a lab, the paperwork is part of the machine:
- Risk assessment, electrical safety records and the declarations of conformity required at the destination
- Raw data exports, not only summary charts — the CSV behind the chart is what you re-analyse a year later
- Spare-parts list, wear-part lifetimes and recommended stock
- Operator and maintenance training records
Our engineering side covers hardware testing for equipment being developed, which is where these data structures are normally defined — before the machine is built, rather than after a problem appears on the production floor.
Common data-collection gaps, and how to close them
- Sampling too slowly. A temperature spike that lasts three seconds between two 30-second samples never appears in the log. Set the sampling period against the fastest event you need to catch.
- Storing setpoints instead of actuals. The recipe says 160 °C; the sensor reading is what happened. Keep both.
- No cycle identifier. If process data and inspection data cannot be joined on a common cycle or serial number, they are two unrelated datasets.
- Undocumented manual overrides. An engineer who manually adjusted pressure mid-cycle has changed the experiment. Without that entry, the next person inherits an unrepeatable result.
- No unit and range definition. Record the pressure unit explicitly. A log that mixes kPa and mbar across a shift costs more time than the sensor that produced it.
Before the test starts
A practical order of work: define the acceptance criteria first, then list the data that proves each criterion, then fix the sampling period, tags, units and naming convention before the first run. Adding channels afterwards leaves gaps that cannot be filled retroactively.
If you are specifying test data requirements for a new machine, PowerDe semiconductor process equipment covers vacuum lamination, bonding, debubbling and related production equipment, and contact our engineering team with your substrate, material stack and target cycle time — we can agree the data schema and logging interfaces during the quotation stage rather than at commissioning.
