AI with Model-Based Design: Virtual Sensor Modeling

A recent webinar showcased a comprehensive workflow for designing, training, validating, compressing, and deploying AI-based virtual sensor models directly to embedded processors within a unified environment. The session demonstrated how AI models can be integrated into Simulink for system-level simulation, verification, and testing, with a focus on optimizing performance and resource constraints. Key features included formal verification of neural network behavior, model compression to reduce memory footprint and speed up execution, and generation of library-free C code for embedded deployment. The webinar used the example of estimating battery state of charge (SOC) in Battery Management Systems to illustrate the practical application of AI-based virtual sensors. These sensors provide estimates for signals that are difficult or costly to measure directly, enabling more efficient and accurate system monitoring. Attendees learned how to balance design tradeoffs involving accuracy, computational performance, and deployment requirements, while profiling code performance and conducting processor-in-the-loop (PIL) tests to ensure real-time viability. This approach is significant for industries relying on embedded AI, such as electric vehicle battery management, where precise state estimation is critical for safety and longevity. By integrating AI models into a single development environment, engineers can streamline workflows and improve the reliability of virtual sensors. The use of formal verification and model compression techniques also addresses challenges related to neural network robustness and hardware constraints, enhancing the feasibility of deploying AI on resource-limited embedded platforms. Sponsored by MathWorks and presented in collaboration with IEEE Spectrum, the webinar reflects ongoing advancements in AI modeling, embedded systems, and digital twin technologies. It highlights the growing importance of virtual sensors in engineering applications, particularly in areas requiring real-time estimation and validation under strict performance and safety standards.