Fault detection of broken rotor bars in induction motors based on the conservative power theory

Induction motors (IMs) are widely used in energy conversion and industrial drive systems. One of the critical faults in IMs is the broken rotor bar (BRB), which must be promptly identified due to its potential for abrupt damage. Several monitoring techniques have been proposed for the detection process, including motor current signature analysis (MCSA) and model-based fault detection (MBFD) but the challenges of supply unbalance and harmonic conditions persist as a gap. This paper presents a BRB detection framework based on Conservative Power Theory (CPT), employing time-domain power components as physically interpretable features. In contrast to spectral-based approaches, the proposed method extracts statistical features from the CPT power quantities, naturally accommodating nonsinusoidal and unbalanced supply conditions. The fault identification is evaluated using the k-nearest neighbors, support vector machines, random forest, and multilayer perceptron classifiers. Experimental validation was conducted on three induction motors operating at 50 Hz and 60 Hz, under multiple loads and voltage unbalance factors up to 5.77%. The results demonstrate performance metrics of more than 98% in all cases, with maximum accuracy and an F1-score of 99.77% in binary classification. The treatment of incipient fault severity is addressed by assessing the likelihood of a motor being healthy or faulty, allowing the severity and progression analysis of the faults.