Three-phase induction motors constitute the backbone of modern industrial drive systems due to their structural simplicity, reliability, and cost efficiency. However, mechanical and electrical faults significantly reduce operational reliability and may lead to unplanned downtime, energy losses, and safety risks. This study proposes an integrated intelligent monitoring and diagnostic framework based on current, temperature, and vibration signal analysis combined with a Nonlinear Auto Regressive model with eXogenous inputs (NARX) artificial neural network. Experimental investigations were conducted using a laboratory-scale test bench under controlled fault conditions including stator unbalance, bearing damage, and shaft misalignment. Multi-sensor data acquisition enabled time-domain and frequency-domain feature extraction for dynamic fault characterization. The collected dataset was used to train and optimize a NARX neural network capable of modeling nonlinear temporal dependencies inherent in induction motor behavior. The developed model demonstrated high classification performance with accuracy rates of 94.2% for general faulty motor detection, 95% for shaft misalignment, 98% for bearing defects, and 95% for stator-related faults. The proposed methodology provides a robust and scalable solution for early fault detection and predictive maintenance in industrial applications.
Three-phase induction motors constitute the backbone of modern industrial drive systems due to their structural simplicity, reliability, and cost efficiency. However, mechanical and electrical faults significantly reduce operational reliability and may lead to unplanned downtime, energy losses, and safety risks. This study proposes an integrated intelligent monitoring and diagnostic framework based on current, temperature, and vibration signal analysis combined with a Nonlinear Auto Regressive model with eXogenous inputs (NARX) artificial neural network. Experimental investigations were conducted using a laboratory-scale test bench under controlled fault conditions including stator unbalance, bearing damage, and shaft misalignment. Multi-sensor data acquisition enabled time-domain and frequency-domain feature extraction for dynamic fault characterization. The collected dataset was used to train and optimize a NARX neural network capable of modeling nonlinear temporal dependencies inherent in induction motor behavior. The developed model demonstrated high classification performance with accuracy rates of 94.2% for general faulty motor detection, 95% for shaft misalignment, 98% for bearing defects, and 95% for stator-related faults. The proposed methodology provides a robust and scalable solution for early fault detection and predictive maintenance in industrial applications.