Predicting the Vibroacoustic Quality of Steering Gears

Abstract

In the daily operations of ThyssenKrupp Presta AG, ball nut assemblies (BNA) undergo a vibroacoustical quality test and are binary classified based on their order spectra. In this work we formulate a multiple change point problem and derive optimized quality intervals and thresholds for the order spectra that minimize the number of incorrectly classified BNA. We pursue a multiobjective goal: the first objective function maximizes the Cohen Kappa metric, while the second objective function reduces the number of employed order intervals. The proposed approach is based on a genetic algorithm and incorporates prior information on the correlation structure of BNA and steering gear vibroacoustics, gained via canonical correlation analysis. The computational experiments show a reduction of both the number of employed order intervals and the costs arising from falsely classified BNA parts with respect to the current production setting, ensuring thus a high practical relevance of our suggested approach.