Fault Diagnosis of Motors via Multivariable Time Sequence Features Fusion of Electrical Signals

Motors are widely used in industrial applications and are key equipment to improve production efficiency. To ensure safety and reduce economic losses, an intelligent fault diagnosis system for motors is required. Traditional fault diagnosis methods are mainly based on vibration signal analysis, which are vulnerable to the interference of environment and noise, and lack of analysis of key motor electrical signals. This paper proposes a fault diagnosis method based on multivariable time series features image fusion of motor electrical signals. First, the electrical signals are collected by Hall sensor and multivariable features are extracted by feature engineering technologies. The sliding window strategy is used to divide the feature data into samples. Then, the extracted features are stacked into images to achieve multivariable features fusion. The Smote-Enn strategy is used to balance the proportion of different types of samples. Finally, Mobilenet-v2 is used to classify feature images, and focal-loss is used as a loss function during training to enhance the recognition ability of difficult to classify samples. The experimental results show that the proposed method has achieved a favorable fault diagnosis effect, providing an effective solution for the intelligent fault diagnosis of motor.