Intelligent Cross-Working Condition Fault Detection and Diagnosis Using Isolation Forest and Adversarial Discriminant Domain Adaptation

The increasing complexity and varying operational conditions of today's rotating machinery present significant challenges for automated fault diagnosis. While data-driven fault diagnosis methods have grown in popularity, they often rely heavily on full-cycle data, making them resource-intensive and less adaptive to diverse working conditions. Addressing this gap, our proposed system avoids the dependence on full-cycle data, employing an efficient two-stage methodology. In the initial stage, an isolation forest (iForest) module operates in an unsupervised mode, isolating operational anomalies indicative of potential faults. These identified anomalies are then channeled into the second stage, where a adversarial discriminant domain adaptation (ADDA) module performs an in-depth fault diagnosis. By streamlining the diagnostic process, our approach not only accelerates fault identification but also reduces the reliance on extensive datasets that are often a staple in conventional diagnostics. Performance evaluations with the XJTU-SY and CWRU bearing datasets show that our system reaches an accuracy of 95.67%, affirming its superiority as a cost-efficient, data-lean solution in machinery fault diagnostics.