Fault Diagnosis of Rotating Machinery based on Domain Adversarial Training of Neural Networks

With the increased requirement of reliable facility operations of rotating machinery, the prediction and diagnosis of fault signals are crucial to improve the safety of equipment. Fault diagnosis with artificial intelligence is an effective method to classify the machinery failure rapidly and automatically. However, the training process requires mass of labeled data which is impractical to obtain. Transfer learning are promoted to overcome the shortage of data by transferring the results of related study and combining current resources to diagnosis. Domain adversarial training of neural networks (DANN) as a typical model of transfer learning efficiently solves this problem. In addition, cohesion evaluation technique is used in the data preprocessing to establish low-dimensional sensitivity feature vectors. In order to verify the effectiveness of the methods, experiments are conducted on two different platforms for transfer learning. The experiment reveals that the proposed method can achieve better results than conventional methods under several evaluation metrics.