AI-Based Virtual Sensors: Modeling and Implementation in Embedded Systems

Modern embedded systems often face the challenge of estimating parameters that are difficult or expensive to measure directly. Artificial intelligence offers a solution through virtual sensors capable of modeling such data based on existing signals. For example, in battery management systems (BMS), AI can predict the state of charge (SOC) without additional sensors, reducing costs and improving fault tolerance.

The development of such solutions involves several key stages. First, AI models are integrated into Simulink for system-level simulation and validation. Formal verification of neural networks is then conducted to assess their behavior across various scenarios. Special attention is given to optimizing models for the limited resources of embedded processors—minimizing memory usage and accelerating execution.

In the final stage, dependency-free C code is generated, simplifying deployment on target devices. Engineers can analyze trade-offs between model accuracy, performance, and hardware requirements. This approach enables the creation of efficient solutions for industrial and consumer devices where reliability and resource efficiency are critical.