PREDICTIVE DIAGNOSTICS OF ELECTRIC MOTOR VIBRATION SIGNAL BASED ON MSCNN-LSTM TECHNOLOGY

Timely diagnosis of electric motor faults is a key task to ensure their uninterrupted operation and reduce repair and maintenance costs.Traditional methods of vibration signal analysis often do not allow achieving high accuracy of defect recognition due to the complexity of processing a large amount of data.The aim of this work is to develop an effective approach for automated detection of electric motor faults based on vibration signal analysis using deep neural networks.To achieve this goal, a method based on the combination of multi-scale one-dimensional convolutional neural network (MSCNN) and long short-term memory network (LSTM) is proposed.The research methodology included preprocessing of vibration signals, development of neural network architecture, selection of hyperparameters, training of the model based on labeled data and its testing.The following types of defects were diagnosed: broken rod, displacement, short circuit, and defects of the inner and outer rings of the bearing.The results of the study confirmed the high efficiency of the developed approach: the model achieved 100% classification accuracy on test data, which was confirmed by error matrix analysis.Additionally, the t-SNE method was used to visualize the feature space, which demonstrated high-quality class separation and confirmed the model's ability to correctly identify different types of faults.The scientific novelty of the work lies in the integration of a multi-scale CNN with a recurrent LSTM network for electric motor diagnostics, which allows for more effective analysis of time dependences of vibration signals and increased accuracy of fault detection.The practical significance of the proposed approach is determined by its application in modern electric motor condition monitoring systems, which will contribute to increasing equipment reliability, reducing operating costs, and preventing unplanned failures in industrial conditions.