Research on Few-Shot Sample Fault Prediction Method for Electric Drive Systems Based on Transfer Learning

This paper introduces a method for fault prediction in electric drive systems, specifically designed to address the challenge of few-shot fault sample availability. Utilizing operational data under normal conditions, the method involves the pre-training of a Long Short-Term Memory (LSTM) network model, followed by the augmentation of scarce fault data through time sliding windows. Transfer learning is then applied to adapt the pre-trained model to fault detection tasks. This innovative approach not only significantly enhances the accuracy of fault prediction but also leverages minimal data to achieve high reliability and efficiency in electric drive assemblies. The methodology demonstrates a substantial improvement over traditional fault prediction models, offering a practical solution to the often resource-intensive field of electric drive system maintenance.