Two Robots Judge BIW Holes, Studs and Edges in 45 s
A body-in-white is welded from many stamped parts, and holes, studs and edges are numerous and widely distributed. A missed punch, a missed weld or a position shift at any one of them propagates down the assembly chain. This cell places robots on both sides of the assembly line, each carrying a 3D vision sensor, completes feature inspection within a 45 second cycle and outputs a measurement report with out-of-tolerance alarms in real time. It runs 7 x 24 unattended, subject to site acceptance results.
Three feature types, three extraction problems
A hole is a closed edge contour; if it is missing, it is missing. A stud is a local high or low point, small in form and easily lost in reflection. An edge is a linear structure judged by direction and continuity. Apply one template to all three and boundary samples will fail. This cell assigns a different extraction template to each class and unifies the outputs in one measurement coordinate system.
Why robots on both sides, not one camera seeing all
A body-in-white is large, and a station on one side has limited sight of side panels and roof, where one face is always hidden. Robots on both sides cover the left and right zones and avoid the long reciprocating travel of a single machine that would drag the cycle. Zoning also keeps each station's field and working distance controllable, so point-cloud quality is steadier than one camera trying to see everything.
How 45 seconds is split between capture and compute
The body triggers the inspection program when it arrives in position, and the robots on both sides move their sensors through feature capture while acquisition and computation proceed in parallel by zone. The cycle holds because of zone parallelism and targeted algorithms, not because the sensor frame rate was raised. Passing bodies move on; out-of-tolerance results alarm and are intercepted by station staff.
Arc light, fume and reflection on the welding line
A welding line brings arc light, fume and vibration, and metal surfaces reflect. The imaging end applies filtering and exposure control, while the algorithm end preprocesses point clouds and applies robust handling to suppress reflection and edge noise. These steps do not change the criteria, but they decide whether judgment stays stable across 7 x 24 operation.
What the report is for after it leaves the station
The measurement report outputs feature pass status and position deviation for the line and the quality data system. Its value is not only whether this body passes; it turns welding dimensional state into statistics. Deviation distribution, trend and station-to-station differences can all be read from the report.
What this cell does not do
It judges geometric feature presence and position. It does not evaluate internal defects or precision dimensions. When requirements extend to dimensional tolerance and form-position evaluation, a composite route of stereo vision plus laser is needed. Stating that boundary at selection is cheaper than adding equipment later.
Questions raised in design reviews
How does 3D vision decide a hole is missing?
It captures point cloud or contour in the feature region, extracts hole edges, stud positions and edge direction, and compares them with theoretical features. Height data makes a missed punch or weld reliably detectable without relying on image contrast.
Is a two-side layout mandatory?
For a body-in-white with this size and feature spread, two sides are the practical answer for coverage and cycle. A smaller workpiece with features on one side can use a single side; sight lines and takt decide.
How much can 45 seconds inspect?
It covers presence and position of body holes, studs and edges and outputs a report. The number of features depends on the model and point plan, confirmed at proposal stage.
What if dimensional tolerance and form-position evaluation are needed?
This cell does not perform them. That requirement calls for stereo plus laser composite measurement rather than adding specifications to this system.