Paint Surface Defect Inspection: 800 Frames Per Second And What It Changes

Overview

Paint defects are small, low-contrast and distributed over curved surfaces, which makes them difficult to inspect by eye and difficult to photograph well. A human inspector working on a bumper sees a few square centimetres at a time, judges gloss by moving their head, and loses consistency within an hour. Camera-based inspection replaces that judgement with a repeatable measurement, but only if the imaging chain can capture the defect class being specified.

The imaging chain has three parts: how fast the camera samples, how the surface is lit, and how the part is moved through the field of view. This article explains why a frame rate of 800 frames per second changes what a paint inspection station can find, and where the remaining limits are.

Big Bird Industrial (formerly Shimada Big Bird Industrial) has been building cleaning and deburring systems for automotive powertrain components since 2001, and the vision line described here follows the same engineering approach: define the acceptance criterion first, then size the hardware around it.

1. Why Frame Rate Sets The Detection Floor

A paint defect is only detectable if at least one frame captures it with enough pixels to separate it from the surrounding coating. As the part moves, the defect passes through the field of view in a finite time. If the camera samples too slowly, the defect appears in one frame or in none, and no algorithm recovers it afterwards.

Sampling In The Direction Of Travel

For a line-scan arrangement, the sampling rate has to be matched to the surface speed so that consecutive lines overlap by a defined amount. For an area-scan arrangement, the frame rate has to be high enough that a defect of the specified minimum size appears in several consecutive frames. Either way, the requirement is derived from the smallest defect in the specification and the maximum surface speed, not chosen from a catalogue.

What 800 Frames Per Second Buys

A configuration using four cameras at 200 frames per second each produces 800 frames per second in total. At a surface speed of 800 to 1000 mm per second, that throughput supports continuous sampling across the full width of a bumper without gaps between exposures. The practical result is that the station can inspect a complete coated surface in a cycle under 15 seconds while still resolving defects at the scale of a paint particle.

2. Lighting Is The Other Half Of Resolution

High frame rates are worthless on a surface that returns glare. Painted panels are curved, glossy and often dark, which means a diffuse light source produces a bright reflection across part of the surface and hides defects in the rest. The lighting geometry matters as much as the sensor.

Why A U-Shaped Source Fits Curved Panels

A U-shaped source wraps the panel so that the angle of incidence stays within a narrow band across the whole curve. That keeps the specular reflection at a predictable position and lets the defect appear as a local distortion in an otherwise even field. On a flat panel a ring light does the same job; on a bumper the source has to follow the geometry.

Separating Defect Classes By Their Optical Signature

Not every defect responds to the same illumination. Particles and craters distort the specular highlight, pinholes and solvent pops appear as dark points, runs and sags change the local slope, and orange peel changes texture at a larger spatial scale. A single lighting setup rarely covers all of them at equal sensitivity, which is why paint stations usually combine more than one illumination mode and fuse the results.

3. From Detection Rate To Production Decision

Inspection performance is quoted as a detection rate, but what a line actually experiences is a false-call rate and a rework flow. A station that misses 0.2 percent of defects and falsely flags 2 to 5 percent of good parts produces a very different operating cost from one with the same miss rate and a lower false-call rate.

Reading The Two Numbers Together

Detection rate alone rewards over-sensitivity. Every threshold that catches one more defect also flags more good parts, and on a paint line each false call means a part sent to polishing or a human re-inspection. The useful specification pairs a miss rate with a false-call rate and states which one is allowed to move.

Cycle Time As A Constraint, Not A Result

A cycle of 15 seconds or less for a bumper or a set of closures is a line requirement, not an achievement of the camera. It sets a budget: handling time plus scan time plus decision time. If the scan is fast enough but the decision requires a second look from another angle, the cycle is what limits the station, not the frame rate.

4. Linking Inspection To Repair

A defect that is found but not located precisely is only half useful. Paint rework requires knowing where the defect is, in a coordinate frame the polishing robot can use.

Shared Coordinates Between Inspection And Polishing

When the inspection station and the polishing robot share one coordinate system, the defect position becomes a robot target directly. The polishing head can then be selected automatically for the defect class, and the operator is not asked to find the spot again on a moving line. This is the difference between an inspection station and an inspection-and-repair cell.

What Cannot Be Automated

Colour matching remains a human judgement on most lines, because the acceptance criterion depends on a reference panel and on viewing conditions that are hard to reproduce at a station. Texture-related defects such as orange peel are measurable, but the threshold between acceptable and unacceptable is usually set by a customer standard rather than by a physical limit.

5. Specification Checklist

A paint inspection specification that can be quoted and verified contains the defect classes and their minimum sizes, the surface speed and cycle time, the illumination modes, the miss rate and false-call rate, and the coordinate frame shared with repair. Missing any one of these leaves a gap that the supplier fills with an assumption.

Verification follows the same logic. Before acceptance, the station is challenged with a set of reference panels carrying deliberately placed defects at the specification limit, and each panel is presented several times in different orientations. A station that finds a defect on the first pass and misses it on the third has a repeatability problem rather than a sensitivity problem, and repeatability is what production experiences every shift. The same reference set also calibrates the false-call rate, because panels without defects are part of the test.

It is also worth specifying what happens to a flagged part: whether it goes to re-inspection, to polishing, or to scrap. That decision determines whether the false-call rate is an annoyance or a cost.

Frequently Asked Questions

Is A Higher Frame Rate Always Better?

No. A higher frame rate reduces exposure time, which reduces light per frame and raises noise. Beyond the rate needed to catch the smallest specified defect at the maximum surface speed, additional speed costs image quality rather than adding detection capability.

Can One Station Inspect Both A Bumper And A Door?

Usually not with the same lighting geometry, because the panel curvature and the working distance differ. Multi-variant stations handle this by changing the fixture position and the illumination recipe per variant rather than by using one universal setup.

How Is Orange Peel Measured?

It is evaluated as a texture parameter at a defined spatial scale, using the variation in reflected light across the surface rather than a single defect location. Because it is a statistical measure, the sampling area has to be specified together with the threshold.

What False-Call Rate Is Realistic?

On painted exterior panels, a false-call rate of 2 to 5 percent is typical in production, with a miss rate around 0.2 percent. Tightening one without loosening the other is usually not possible on the same imaging chain.

Conclusion

Frame rate determines whether a defect is captured at all; lighting determines whether it is visible once captured. Both have to be sized from the smallest defect in the specification and the surface speed of the line, and both have to be traded against the false-call rate the production team can absorb.

The stations that work in production are the ones where the specification states a miss rate, a false-call rate and a cycle time together, and where the inspection result lands in the same coordinate frame as the repair operation.