Stereo and Laser Measure a Car Body in 90 Seconds
A car body is made of stampings, closures and structural parts, and its dimensional chain is long: a deviation anywhere can surface as an assembly gap or a flushness error. This project pairs an industrial robot with optical tracking, switches between stereo vision and laser sensing according to the feature being measured, and captures body point clouds fast enough to complete dimensional measurement, dimensional tolerance and form evaluation, and surface defect detection within a 90-second cycle, subject to site acceptance results.
Three jobs in one system, and why
Surface defect detection and dimensional measurement sit in different stations in a traditional process, with separate equipment, separate positions and separate data. Three systems mean three loadings, three alignments and three data sets that do not correspond. Combining them saves more than floor space: the three results land in one quality file, so a defect location and a dimensional deviation can be compared against each other.
When laser, when stereo
Large surfaces suit laser scanning: fast point capture, wide coverage, continuous acquisition across curved and long profiles. Holes and edges suit stereo vision: a smaller field with higher resolution gives more stable edge localisation. Switching by feature lets each sensor do what it is best at instead of forcing one method through every task. The mapping from feature type to sensor is fixed in the measurement plan, not decided by the operator on the floor.
Optical tracking fixes accumulation, not accuracy
Over large motions the robot's positioning error accumulates, and the more stations and the longer the travel, the more visible it becomes. Optical tracking measures the end-effector pose in real time and gives each station its own datum, separating robot motion error from the measurement chain so multi-station clouds remain usable in one frame. The larger the workpiece, the more this matters.
Switching is planned in advance
Sensor changeover is not decided on the spot. A feature library classifies the body's elements ahead of time, and the system generates the sensor choice and the motion path from it, which protects accuracy and cuts manual teaching time. With a good plan, switching is incidental; with a rough one, it becomes trial and error at the station.
90 seconds and one archive
Once the cloud is in the workpiece frame, the algorithm fits planes, circles and cylinders, compares measured against nominal values, and evaluates straightness, flatness and position. Local form anomalies in the same cloud feed surface defect recognition. The measurement report and the form table bind to the body serial number, so they can be traced and counted. The robot follows its planned path while stereo and laser alternate and the cloud is stitched and processed in parallel inside the 90 seconds.
Where this is overkill
If the requirement is only whether a feature is present, a lighter 3D POKA-YOKE scheme is enough. The value here is carrying dimensions, form and defects at once. Configure the system to the inspection goal rather than to the brochure.
Review questions
Why switch between stereo and laser automatically?
Laser captures large surfaces quickly, stereo measures holes and edges precisely. Switching lets one task hold both efficiency and accuracy.
What does optical tracking contribute?
It measures the robot pose in real time, corrects positioning error over long travels, and supplies the datum for stitching stations together.
Do defects and dimensions compete for cycle time?
They share the capture. Dimensional evaluation fits reference geometry while defect recognition looks for local form anomalies, so the acquisition step is not duplicated.