The paper discusses the possibility of detecting damage to the bearing of an asynchronous motor operating as part of a variable-frequency drive using machine learning algorithms and the use of multi-band filters during processing information about the stator current. Detecting a fault in an asynchronous motor when powered by an inverter is more difficult than when powered directly from the network due to the masking of a fault signs by higher harmonics from the inverter in the stator current. The effectiveness of using multiband filters (in particular, a type II Chebyshev filter) to solve the problem of extracting fault symptoms is shown. The work used an approach based on a type II Chebyshev filter, which has a flat amplitude-frequency characteristic in the passband, which is important for preserving the signal amplitude and, in comparison with the Butterworth filter, which also has a flat amplitude-frequency characteristic in the passband, has a faster decline in the suppression band. Multi-band filtering was used at the stage of extracting fault features from stator current signals and then feeding them into machine learning classifiers. To recognize the fault, after using multi-band filtering, four machine learning methods were applied, namely: artificial neural network, K-nearest neighbors, support vector machine and naive Bayes classifier. All methods showed a very high probability of correct fault recognition. The highest efficiency (more than 90% probability of correct fault recognition at various operating frequencies and engine loads) was shown by classifiers based on an artificial neural network and the K-nearest neighbors method.