A Robust Fault Detection and Classification Approach for Wireless Sensor Networks Using Stabilized Model Predictive Control and Optimized SVM Classifier

The industrial and environmental systems of wireless sensor networks are very susceptible to sensing errors, packet corruption, and noisy measurements, which may cause the control loop to be destabilized and result in the unsafe actuator behavior.Real-time fault detection, fault classification, and resilient predictive control are made possible by means of a fully integrated FPGA-based fault-aware Model Predictive Control (MPC) framework presented in this study. The suggested architecture integrates adaptive preprocessing, statistical fault detection, a hardware-optimized Tiny-SVM classifier, with an MPC controller which applies a built-in fault probability in its optimization operation. Placed on a Xilinx ZynqUltraScale+ MPSoC and tested with a publicly available WSN dataset, the system can achieve microsecond scale latency, high classification accuracy as well as high resistance to severe fault injection. Co-designing hardware and software reduces the amount of FPGA resources used while preserving stable closed-loop operation even in the event that several sensors become unreliable. Findings prove the framework can offer high-performance, resource efficient, and fault resilient control solution that can be used in industrial IoT, environmental monitoring, and safety critical cyber-physical systems.