A Search-Based Domain Adaptation Network for Fault Diagnosis of Rotating Machinery Under Cross-Operating Conditions

Most domain adaptation diagnostic methods focus only on minimizing the distribution discrepancy between domains but fail to learn discriminative features among different classes of data. This often results in data being statistically similar but lacking discriminative power. To solve this problem, this paper proposes a Searcher Mechanism and a Supervisor Regularization Mechanism. The Searcher Mechanism learns domain-invariant features and inter-class discrepancy by round trip search for homogeneous samples between different domains. The Supervisor regularization mechanism improves generalization and prevents overfitting by ensuring that the searcher explores all target domain samples with equal probability. First, we employ a wide-kernel deep residual network as a feature extractor to extract features from the data of different domains. Subsequently, the Searcher Mechanism continuously aligns the distributions between different domains while learning discriminative features for different classes. Additionally, to ensure that the Searcher Mechanism searches all target domain samples fairly, a supervisor regularization mechanism is introduced to supervise the search process. Through learning discriminative features, the classifier can form more accurate decision boundaries. Experimental results on two datasets under varying operating conditions validate the cross-domain diagnostic performance of the proposed method. The related code is on https://github.com/jiaqilearning/searcherFD.