Vehicles are increasingly getting more complex to design, because of the large number of design variants, the need for modular designs and the use of highly integrated systems due to e.g. the electrification trend. While Computer-Aided Engineering (CAE) tools are widely adopted, physical measurements remain an integral part of the methodology throughout all stages of the vehicle design process, for design verification and validation purposes but also for model updating in view of future design cycles. The industry is seeking solutions to increase the overall efficiency of physical testing, both in terms of equipment cost and in terms of time. Machine Learning (ML) methodologies are therefore envisioned to reduce the amount of time-intensive and repetitive manual labor and to achieve “first time right” testing (e.g. avoiding measurement mistakes). Also in the later stages of the vehicle lifecycle measurement data is collected, e.g. for the purpose of “end-of-line” production quality inspection. In this case ML can enable an automatic interpretation of the measured data in a quality inspection system.

This presentation will discuss ongoing research activities at Siemens on the application of ML to vehicle testing through different use cases, e.g. data-driven virtual sensors for vehicle dynamics testing, sensor anomaly detection in large channel count measurement campaigns and deep learning based booming noise detection in acoustic end-of-line testing.