The traditional approach for collision prediction in suspension systems uses nominal kinematics and part shapes. After simulating various combinations of wheel travels and steering rack strokes, general offsets are added to the part hulls. Afterwards the clearance between parts is rated based on experience. The main idea of the new approach is to split the problem into two tasks: the variation in motion due to kinematic effecting tolerances and the variation in outer contours. To handle the effort, the multibody simulation is substituted by neural networks to predict the trajectory based on hardpoint tolerances. The second task can be solved by the approach of summarized voxelization, which is used to describe parts in voxel representation including the probability, that the part is in that position. To predict the collision probability, the potentially involved voxels in a certain wheel position are relatively moved to the other part for an arbitrary number of variations calculated by the meta-model. Afterwards, the moved elements can be revoxelized in the relative grid. This leads to a voxel representation which includes variation in motion and in shape. A collision calculation can be done afterwards and yields the overall probability of part collisions. In the first place, the methodology offers the opportunity to take tolerance computationally into account and also provides statistical collision information. This revolutionizes the existing clearance evaluation methods.
Session:
Poster
|