TinyML Powered Real Time Rotating Machinery Fault Detection Using Vibration Spectral Analysis on ESP32

This study developed an ESP32-based TinyML system capable of detecting faults in rotating machinery in real-time through vibration analysis. The system operates entirely on the edge device using FFT and an INT8 quantized neural network, eliminating the need for cloud connectivity. Experimental results demonstrated a validation accuracy of 92.4%, a real-world accuracy of 80.27%, and an inference time of just 2 ms, making it suitable for cost-effective predictive maintenance.