Top Machine Vision Solutions for Automotive Manufacturing

Matching the right technology to weld, surface, measurement and assembly inspection

A modern vehicle is checked by machine vision hundreds of times before it leaves the plant. Most of those checks happen in seconds, around the clock, on surfaces that reflect like mirrors and parts that can arrive hot. Automotive is one of the hardest environments vision has to work in. It is also one where getting the inspection right pays back fastest, because a defect that escapes the line is expensive and a recall is far worse.

Four jobs account for most automotive vision, weld inspection, surface defect detection, dimensional measurement and assembly verification. Each one needs a different technology. The common mistake is to pick a camera first and work backwards.

Weld inspection

A weld has to be judged on its shape as much as its surface. Is the bead the right height and width, is there enough of it, has it wandered off the seam, are there pores or spatter where there should be none. Two-dimensional imaging catches the surface faults. The geometry needs 3D. This is where a laser profiler such as Zebra's AltiZ or Teledyne's Z-Trak earns its place, measuring the bead cross-section in microns as the part moves past. Shiny metal and dark burn marks in the same frame make this a high dynamic range problem, which means the camera and the lighting matter as much as the software. For the variable faults that resist a fixed rule, deep learning in Zebra's Aurora tools adds the judgement a rules-based check cannot. Where the concern is the weld process itself rather than the finished bead, thermal imaging can read the heat signature as the weld is laid down.

Surface defect detection

Painted panels and body surfaces are the hardest thing to inspect in the plant, because the defects are subtle and the surface is specular. A scratch, a dent, a run in the paint or an orange-peel texture shows up only when the light is controlled precisely. Deflectometry, which projects a known pattern onto the surface and reads how it distorts, is the technique built for this. It finds dents and waviness the eye would miss. For continuous surfaces and coil, a line scan camera such as Teledyne's Linea builds a seamless image as the material moves. High-resolution area scan covers discrete panels. Across all of them, deep learning has become the practical way to separate a real defect from harmless variation, which on a paint line is the difference between a useful system and one that cries wolf.

Dimensional measurement

Automotive lives on tolerances. Gap and flush between panels, hole positions on a stamped bracket, the length of a shaft, the fit of a trim piece, all have to sit within tight limits and be proven. For measurement in a plane, a telecentric lens removes the perspective error that trips up a normal lens. A Computar or Kowa telecentric in front of a high-resolution camera measures a feature the same whether it sits near or far in the field. For form and position in three dimensions, the same laser profilers used on welds measure gap and flush and stamped geometry directly. The rule worth keeping is to match the measurement method to the tolerance. Asking a 2D check to prove a three-dimensional tolerance is where measurement systems quietly go wrong.

Assembly verification

The last job is confirming that the right parts are present, correct and in the right place, the clip that is often missing, the connector that is not fully seated, the label that has to match the vehicle. For a single clear check, a smart camera or vision sensor such as Zebra's FS range or the Iris GTX does the whole job in one compact unit at the point of assembly. Where the parts vary or the check is subtle, deep learning classification is more forgiving than a rigid template. And because every part in automotive has to be traceable, code reading sits alongside the inspection, reading the dot-peen, laser-etched or printed codes that follow a component through the plant and out to the field. Traceability is not a bolt-on in automotive. It is usually the reason the camera is there at all.

What automotive adds to every one of these

The technology is only half the problem. Automotive adds conditions that decide whether a system survives contact with the line. It runs at full production speed, which means the camera, the lighting and the processing have to keep up without slowing takt time. It runs around the clock, where reliability and a spares plan matter more than a clever demo. The environment is unkind, with weld fume, oil mist, vibration and wash-down, which makes enclosures and ingress ratings earn their keep. Every result usually has to be logged against a part for IATF traceability, making the inspection and the record one job, not two. A system specified for a benign lab rarely survives a body shop.

How Clearview approaches an automotive line

We are engineering-led rather than catalogue-led, which matters more in automotive than almost anywhere. The four jobs above each need a different technology. Most of the value is in matching the right one to your tolerance, your speed and your environment before anything is bought. We prove the approach on your own parts in our Insights Lab first, with your welds, your painted panels, your actual components, because a paint defect or a weld bead behaves nothing like a test target. We have local engineers in the UK, Spain, Germany and France who can stand on the line with you. We carry the 3D, deep learning, lens and lighting pieces that automotive inspection actually calls for, rather than a single camera range we have to make fit.

Read the Automotive Vision Guide

We have put the detail behind all four inspections into one place, the cameras, lenses, lighting, 3D and software that suit each automotive job, with the trade-offs for high-speed lines and the traceability automotive demands. Read the full Automotive Vision Guide. To talk through a specific line or prove a check on your own parts, get in touch with our team.

Get in touch at info@clearview-imaging.com or call +44 (0)1844 217270

Related reading   2D vs 3D Machine Vision   |   Area Scan vs Line Scan   |   Why Your Lighting is Probably Wrong   |   Machine Vision and AI   |   Contact