Building a 3D Feature Map on a Moving Body Line
A body-in-white is one of the largest assemblies in car manufacturing. The body flows along the welding line, and slots, holes and cylindrical mounting features are spread across it, with position parameters that directly affect final assembly. This project uses a monocular line-laser sensor that swings while it scans: the laser sweeps and the camera captures the light-knife image, and an extraction algorithm computes the feature position parameters. A single body is checked within a 90-second cycle, subject to site acceptance results.
A fixed laser sheet covers one narrow band
A line laser projects a sheet onto the surface, and fixed, it covers one narrow band per shot. On a body the features are widely spread, so a fixed installation would need many units or repeated movement of the body. Swinging the sheet across the workpiece lets one sensor cover a much larger area, which is the trade between coverage and equipment count.
Light-knife cloud extraction is the core
The algorithm extracts the laser stripe centreline from every frame and converts it into a point cloud using the sensor pose and the swing parameters. How accurately the centreline is extracted, and how well pose and swing parameters are matched, decides cloud accuracy. Every subsequent slot or cylindrical feature position is computed on that cloud, so an error here cannot be recovered downstream.
Weld spots and spatter break the stripe
A body surface carries weld spots, spatter, edges and paint steps, and the laser breaks up or distorts where it hits them. The algorithm has to suppress stray reflection and handle stripe breaks so the centreline stays continuous. If a break is mistaken for a feature boundary, the position parameter goes wrong, and that kind of error is very hard to find in the data afterwards.
Frame-to-frame registration sets accuracy over long spans
During a swinging scan the sensor and the body move relative to each other, so frames have to be registered. The scheme combines swing parameters with a motion model to stitch the cloud, keeping position accuracy over a long sweep. If the registration model does not match the actual motion, the further it sweeps the larger the deviation grows.
90 seconds against the line rhythm
The station is integrated into the welding line: the body enters, the sensor starts its swinging scan, and within 90 seconds the slots and cylindrical features are swept and their position parameters computed. Data goes to the quality monitoring system while the body continues down the line. An abnormal result triggers a quality alert rather than a body pulled off the line.
How it differs from image-based checking
Image-based systems convert position from a 2D picture and are sensitive to ambient light and surface reflection; a line laser acquires 3D data directly, so position does not depend on a 2D conversion and resists ambient light better. The price is that capture is a line-by-line sweep, so cycle time and coverage have to be balanced.
Three items are hard requirements at selection: synchronisation between the swing mechanism and sampling, immunity of the light-knife image to interference, and point cloud processing speed. The scheme suits position checking on body-in-white, frame and profile-type large parts, and inline continuous monitoring on welding lines.
Questions before adoption
Why swinging scan on a body?
Slots and cylindrical features are widely distributed, and a fixed sheet covers one band. Swinging lets one sensor cover a wide area continuously.
What does the extraction algorithm actually do?
It pulls the centreline from the stripe image and, with sensor pose and swing parameters, builds the 3D cloud on which feature positions are computed.
What does inline give over offline?
Offline samples one body at a time and takes it out of the line. Inline monitors every body, catches feature position deviation early and feeds live data to weld dimensional quality.