AI with Model-Based Design: Virtual Sensor Modeling

Virtual sensor modeling is a resourceful technique that may be used to mimic the behavior of a physical sensor. Developing a virtual sensor (also known as soft sensor) can help in situations when the signal of interest cannot be measured, or when a physical sensor adds to much cost and complexity to the design. For example, when developing a Battery Management System (BMS) for an Electric Vehicle, laptop, etc., having an accurate value of the Battery State of Charge (SOC) is a critical element in the design. However, it may not be directly measured. Deep learning and machine learning techniques can be used as alternatives or supplements to Kalman filters and other well-known virtual sensing techniques. These AI-based virtual sensor models must integrate with other parts of the embedded system. In the case of a BMS, an AI-based SOC virtual sensor must be integrated with power limitation, fault detection, and cell balancing algorithms. Development of such a large and complex system requires integration, implementation, and testing of different components while minimizing expensive and time-consuming prototyping with actual hardware. Model-Based Design is a proven approach to accomplish this.

In this session, you will learn how to develop virtual sensor models using feedforward neural networks, LSTMs, decision trees, and other AI techniques. Using the example of BMS SOC estimation, you will learn how to integrate AI models into Model-Based Design, so that you can test your design using simulation and implement it on an NXP S32K3xx board using automatic code generation. You will see how to evaluate and manage AI tradeoffs that span from model accuracy to deployment efficiency.