Three-phase induction motors are the primary actuators for converting electrical energy into mechanical energy in the productive sector, constituting key assets due to their widespread use and critical function. Reducing maintenance costs and implementing predictive techniques incentivize the development of systems to identify intrinsic defects. The increasing demand for customization in manufacturing affects maintenance due to fast production line adaptations. This leads to unforeseen failures that compromise reliability. There is a lack of research on detecting and diagnosing faults in induction motors under intermittent drives or varying operating conditions. To fill this gap, the present research proposes a methodology for recommending algorithms to diagnose and detect broken bar defects in three-phase induction motors during transient operation based on a cognitive system. The framework explains and detects fault causality. Using experimental data (current, voltage, vibration), three-phase induction motors were tested under normal conditions, applying various severities of broken bar faults with load torque variations. Features were extracted from each signal, and feature selection algorithms of different mathematical natures were applied. Machine learning models were built, validated, and tested with multicriteria measures. To assess robustness, white noise was inserted into the experimental signals. The Consistency-Based Filter algorithm emerged as the most suitable for feature selection combined with Random Forest and Multilayer Perceptron models. The best results were achieved with up to 80 % noise tolerance without compromising predictive capacity for diagnosing defect severity. Features following a Gaussian distribution showed better predictive capacity, resulting in a reliable framework for fault diagnosis in induction motors.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号