Search for a command to run...
The development in automotive industries towards electrification and automation raises special attention to Safety-Related Availability (SaRA) in onboard power supply systems (powernets), where reliable battery diagnosis becomes the crucial enabler. The Safety of the Intended Functionality (SOTIF, ISO 21448) complements traditional Functional Safety (FuSa, ISO 26262) by addressing functional insufficiencies of Advanced Driver Assistance Systems (ADAS). However, applying SOTIF to other use cases, such as the battery diagnosis, remains underexplored.This work presents a SOTIF-oriented probabilistic validation framework that systematically integrates uncertainty quantification (UQ) and reliability analysis (RA) techniques. This framework adopts a Bayesian neural network (BNN) with heteroscedastic noise modeling as the core modeling tools. We jointly quantify epistemic and aleatory uncertainties to guide risk estimation, active test selection, and hazardous scenario discovery. To improve explainability, we applied SHAP-based sensitivity analysis with additional care given to input dependencies.The framework is validated using battery test data generated from a high-fidelity electrochemical battery model, which serves as a representative proxy for real-world batteries. This work aims to provide a practical extension to SOTIF for real-world deployment of complex safety functions and encourage further research in this direction.