Stamped part numbers on metal are hard to read because the characters have the same colour as the surface, have very low contrast and catch glare from every light source. Vision AI on a handheld wearable solves this by looking at the part up close, adjusting brightness and contrast on the spot and reading the number at any angle, so the part can be checked against its production and customer labels before the crate ships.
In this guide we look at why label-to-part mismatches happen in automotive supply, why fixed cameras and visual checks struggle with stamped characters, and how metal part and label verification with TIM INSPECT compares two labels and the part in one short flow.
Why label and part mismatches happen
In automotive supply, a crate typically carries two labels: the production label created in your own plant and the customer label your customer's receiving process relies on. Both must describe the part inside. Problems appear when they do not:
- The two labels disagree: a customer label is printed for the wrong part number or put on the wrong crate.
- The labels agree but the part does not: a very similar part ends up in the crate.
The second case is the hardest. Similar parts often differ only by a number stamped into the metal. A mixed-up crate may only be discovered at the customer's line, where the cost of a wrong part is much higher than at your shipping area.
Why stamped characters are hard to read
Printed labels are designed to be read by machines. Stamped numbers are not. They are pressed into the metal, not printed, so there is no ink to create contrast. The surface reflects light, and the angle at which glare appears changes with every small movement of the part. Oil, scratches and pen marks add noise. On top of that, the part sits differently every time: the number may be upside down, sideways or tilted.
Traditional approaches and their limits
- Visual check by the worker: reading a low-contrast stamped number under factory lighting is tiring and error-prone, especially when similar numbers differ by a single character.
- Barcode scanning of labels only: confirms the labels, not the part. It cannot tell you that the part inside is wrong.
- Fixed camera station: looks at the part from one angle. When glare falls at that angle, the number cannot be read, and parts have to be carried to the station and positioned.
What is needed is a check that reads both labels and the metal, decides whether all three agree and tells you where the error is.
How vision AI verifies metal parts against their labels
In the automotive supplier deployment described on the solution page, the check runs on AXION, held by the worker, with TIM INSPECT AI models doing the reading and comparison on the device. The flow has four steps.
1. Scan both labels
The worker points AXION at the production label first, then at the customer label. The device compares the information on the two. If they do not match, the error surfaces before anyone looks at the part, and the record shows that the problem is in the label.
2. Show the part to the device
The worker holds AXION close to the part. On a glaring surface AXION adjusts brightness and contrast, then reads the number stamped into the metal regardless of its orientation or angle. Because the device is in the hand, it can look from the angle where glare does not hide the characters.
3. Get alerted on a mismatch
If the labels or the part do not match, the device vibrates and the worker is notified. The order of checks makes the source clear: if the labels agree but the part does not, the error is in the part.
4. Write the result to your system
If all three match, the crate is approved. Every result is recorded with an evidence photo and written to SAP, MES or whichever system you use. The check usually takes place before shipping.
The solution page reports 100% accuracy and 100% customer satisfaction for this deployment, with a three-way match of two labels and the part, based on the Thread In Motion project report.
What to plan for before a pilot
Collect real samples of your parts
Surface conditions such as scratches, oil or pen marks are assessed on your own parts during setup. Bring parts from different batches and different stamping tools, including worn ones, so the models see the real variation.
Map part variants and label structure
List the variants that are easy to confuse and the fields on each label that identify the part. Decide which fields must match between the production label, the customer label and the stamped number.
Choose the checkpoint
The natural place is just before the crate is closed or released for shipping. Because the check stays in the worker's hand, it does not need a separate station; you only need to decide at which step in the packing or shipping flow it runs.
Lighting
The device adjusts the image, but stable lighting still helps. Avoid working directly under a single strong light that creates hard reflections on the part.
Integration
In deployments the result is usually written to SAP; MES or another system is also possible. Agree which record the result and photo attach to, so a mismatch is visible to quality and logistics teams. For crate-level checks at the exit gate, shipment and crate verification matches crates against the SAP shipment list with fixed cameras.
FAQ
What is metal part and label verification?
It is a check that a crate's production label, its customer label and the number stamped on the part inside all match. With TIM INSPECT it runs on AXION and the result is written to SAP or MES.
How is a stamped number read on shiny metal?
AXION looks at the part up close from the right angle and adjusts brightness and contrast on the spot, then reads the number regardless of orientation.
Why AXION instead of a fixed camera?
A fixed camera sees the part from one angle; when glare falls there, the number cannot be read. AXION is in the worker's hand, so it finds the right angle and no separate station is needed.
How do you know whether the error is in the label or the part?
The two labels are compared first, then the part is compared with the labels. If the labels disagree, the error is in the label; if they agree but the part does not, the error is in the part.
Next step
To look at your part variants, labels and surface conditions together, request a demo and review the metal part and label verification flow with our team.