Deep Learning-based Fault Diagnosis for Three Phase Induction Motors in Ventilation Systems

Induction motors are critical components in HVAC and ventilation systems within radiopharmaceutical facilities, where even minor electrical faults can lead to airflow disruptions, compromise contaminant containment, and elevate safety risks for both personnel and the environment. This study presents a hybrid 1D-CNN + LSTM model for real-time fault diagnosis of three-phase induction motors. A custom MATLAB/Simulink dataset simulating healthy operation and fifteen distinct fault conditions, including high-amplitude and zero-voltage scenarios, provides features such as stator currents, rotor speed, torque, and three-phase voltages. The hybrid model combines 1D-CNN's spatial filtering with LSTM's temporal modeling to achieve 99.7 % classification accuracy, a testing loss of 0.004, and cross-validation accuracy between 99.6 % and 99.8 %. Compared to KNN, decision tree, and ensemble baselines, our approach reduces inference time to 10 s for 50 000 observations through quantization and pruning, making it viable for predictive maintenance in high-risk environments. By detecting incipient faults that could destabilize ventilation, this method enhances equipment reliability, minimizes downtime, and supports compliance with stringent regulatory standards.