Data-based Estimation of Cable Bundle Stiffnesses using Gaussian Process Regression Datenbasierte Schätzung von Kabelbündelsteifigkeiten mittels Gauß-Prozess-Regression

Cable harnesses consisting of several kilometers of differently combined and bundled cables are the nervous system of a modern passenger car. In recent years, the Fraunhofer ITWM has developed methods to include flexible elements such as cable bundles in design and assembly simulations already in early development phases. This requires characteristic model parameters such as the effective bending and torsional stiffnesses of the bundles. Due to the high variability of the bundles, the measurement of these parameters is time-consuming and cost-intensive and often not possible if real bundle prototypes are not available in the early development phase. In this paper, we present a data-based approach from the field of artificial intelligence, or machine learning, to estimate effective cable bundle stiffnesses based on easy to measure cable and bundle characteristics. Using a training dataset consisting of measured stiffnesses and other bundle characteristics with sufficient variability and diversity, we identify a nonlinear model that can subsequently be used to estimate the effective stiffnesses of additional bundles. We use the probabilistic method of Gaussian Process regression for this purpose, which estimates a predictive distribution rather than just one value. We present and discuss this approach and demonstrate its applicability, especially for automotive applications.