Scalable Mixed-Criticality Safety Concepts for a Stationary Battery Management System (BMS) With Integrated Machine Learning (ML)

Battery management systems (BMSs) play a crucial role in controlling and supervising the safe operation of lithium-ion batteries in stationary energy storage systems. The integration of mixed-criticality functions (combining safety functions of different criticality and non-safety-related functions) into these electronic embedded systems offers significant potential benefits but also introduces technical and certification challenges. Additionally, modern BMSs are increasingly integrating battery models based on machine learning (ML). This publication contributes with the example-based design and analysis of scalable computing architectures for the development of a mixed-criticality BMS that integrates ML battery models in stationary applications. The safety concept focuses on the computing architecture, where mixed-criticality functions and ML models are integrated, and on analyzing the advantages and disadvantages of the proposed approaches. The computing architectures discussed range from widely used lockstep and single-core processor solutions to high-performance computing platforms such as multicore devices. The system-on-a-chip with hypervisor is identified as the preferred architecture for a BMS that is flexible to efficiently adapt and integrate non-safety-related functions that require intensive computing resources without compromising safety requirements.