Automated fault detection of broken bars in induction motors using kurtosis analysis

Induction motors are commonly affected by damaged rotor bars. To avoid the consequences that this fault may induce, new broken bar fault detection is being proposed, aiming to achieve a high fault detection accuracy in an automatized way. To this end, some approaches rely on neural networks or artificial intelligence algorithms. However, these methods imply a high use of computational resources. Other proposals with lower computational demand are based on a signal processing algorithm, such as the Taylor-Fourier transform, and statistical-based approaches, like kurtosis analysis, for extracting signal data distributions’ characteristics. In this work, the O-Splines of the Taylor-Fourier transform are employed to apply a bank of seven filters to the current signals of an induction motor under different damage levels and two load conditions. The Taylor-Fourier transform is an excellent feature extraction option, suitable for characterizing the current signals’ behavior under different damage and load states. Then, the kurtosis of the seven filtered signals is computed, and based on the separability of data among the healthy case and the faulted ones, one filter is selected for each load case. Finally, based on the selected filter, the amplitude estimations of the filtered signals are displayed as boxplots to identify the motor condition. The main advantage of this method is that it automatically detects the best central frequency of the filter according to the load condition based on the kurtosis, eliminating the need for prior knowledge of the load condition and the motor’s characteristics. Results suggested that this methodology can identify the presence of the broken bar fault starting from a case of 70% of damage in the bar for both 75% and 50% load conditions.