Battery Housing: Stereo and Laser Check Every Face in 90 s
A new-energy battery housing is a large structural part of the battery pack, with many holes and mounting faces on the frame, bottom and side walls. The machining accuracy of these features directly affects cell assembly and vehicle fit. This cell uses an industrial robot carrying stereo vision and a laser scanner, working with optical tracking, switching acquisition by inspection area, and completing dimensional tolerance, form-position evaluation and surface defect detection within a 90 second cycle, subject to site acceptance results.
A multi-face part rules out 2D before accuracy is discussed
The housing is not a flat plate. Frame, bottom and side walls face different directions, and holes and mounting faces are distributed across several planes. 2D imaging expresses only a projection and cannot capture position and depth of 3D features. To evaluate dimensions, form-position relationships and surface quality in one data system, measurement has to be 3D.
Large area and local detail need different sensors
Stereo vision covers the external point cloud quickly, while laser scanning captures local detail at high accuracy. A single sensor has to trade speed against precision; combined, each does its own job. Broad surfaces are covered by stereo, critical areas are refined by laser, and the method avoids sacrificing accuracy for speed or slowing the cycle for precision.
The coordinate chain: hand-eye calibration plus optical tracking
Calibration establishes sensor parameters and the hand-eye relationship with the robot, and optical tracking unifies measurement coordinates into the workpiece frame. Robot positioning error accumulates as the arm moves; optical tracking localizes in real time and keeps the moving trajectory accurate. Without this link, data from different faces cannot be assembled.
Path planning owns most of the 90 seconds
The housing arrives by line conveyor, and the robot follows a planned trajectory to measure each face. Which face comes first, where the laser takes over, and how long the arm dwells at each position all affect the cycle. The 90 seconds is not the algorithm computing quickly; it is the path removing wasted motion.
How results become a release decision
Feature extraction locates holes, planes and edges in the point cloud. Dimensional calculation compares features with the design model, evaluates dimensional tolerance and form-position relationships, identifies surface defects in the same pass, and outputs results that drive release of passing units. Data is retained for later process analysis.
Where the sensor mix stops paying
When the workpiece is dominated by large-area coverage, a stereo-focused configuration is enough. Where local detail demands high accuracy, laser scanning joins the mix, with optical tracking to hold moving-measurement accuracy. More sensors are not automatically better; the mix follows the feature composition.
Four questions from selection
Why not use a single sensor for the housing?
The external surface is large and the points are many. Stereo covers area quickly and laser handles local detail; switching as needed is the only way to keep both speed and accuracy.
What problem does optical tracking solve?
It unifies stereo and laser data in the workpiece frame, reduces the effect of robot positioning error during moving measurement and keeps large-range trajectories accurate.
What does dimensional and form-position evaluation include?
The external point cloud is compared with the design model to evaluate dimensional tolerance and form-position relationships, with surface defects identified in the same pass. All three results share one source and cannot contradict each other.
How was the 90 second cycle set?
It matches the transport rhythm of the new-energy line. Path planning, sensor switching and parallel computation set it, with final figures subject to site acceptance results.