Machine Learning-assisted Test Records Analysis during the Real-Time Validation of Automotive Software Systems based on HIL Simulation

No ratings

Presented at HICSS 2025 by

In the automotive industry, according to ISO 26262, comprehensive testing is conducted to ensure software systems quality over various phases of the V-model. However, at the system integration and testing phase, a significant amount of time series data is analyzed manually based on expert knowledge to identify the nature of the failure occurring, which is a costly process in terms of time, effort, and difficulty. This study proposes a novel Machine Learning (ML)-assisted failure analysis approach for the real-time validation of automotive software systems (ASSs). Specifically, based on a representative critical faults dataset, intelligent data-driven ML-based Fault Detection and Diagnosis (FDD) models are developed, including LSTM and Denoising Autoencoder (DAE)+k-means techniques for known and unknown sensor-related faults, respectively. The novelty of these models lies in their ability to identify single and simultaneous faults in noisy conditions. To assess the efficacy of the developed model, a dataset from a virtual test drive platform with a high-fidelity entire vehicle model was employed. The evaluation outcomes illustrate the superiority of the proposed LSTM model for known faults classification in comparison to other state-of-the-art methods, with an average F1-score of 91.85%. Furthermore, the integration of DAE with k-means exhibited a high clustering performance against noise with a low MSE and DB, i.e., 0.044 and 0.68, respectively. Consequently, the proposed approach facilitates the real-time validation of the ASSs in an efficient manner, thereby enhancing the safety and reliability of engineering.