Time-Series-Based Clustering for Failure Analysis in Hardware-in-the-Loop Setups: An Automotive Case Study

Abstract

Testing is an important cost driver in development projects. Especially in the automotive industry, immense efforts are spent to carry out validation facing increasingly complex systems. Hardware-in-the-Loop test benches are essential elements for (functional) validation. Naturally, failures commonly occur, whose analysis is challenging, time-consuming and oftentimes performed manually, making the diagnosis process one decisive cost-driving factor. By experience, many failures happen due to few underlying faults. We discuss our lessons learned when performing similarity-based clustering to identify representative tests for each fault for system-level testing where test execution times are high and the complexity of the system-under-test and also the test setup leads to complicated failure conditions. Results from an industrial automotive case study–a drive train system dataset consisting of 57 test runs–show that utilizing our general, project-agnostic approach can effectively reduce failure analysis time even with a limited set of data points.