Failure mechanism-driven multi-adversarial domain transfer learning for rolling bearing fault diagnosis

The fault diagnosis of rolling bearings is crucial for ensuring the safe operation of mechanical equipment. However, existing data-driven methods often face performance bottlenecks in cross-condition diagnostic tasks due to a lack of understanding of physical failure mechanisms. Furthermore, they are prone to negative transfer, which reduces diagnostic accuracy when significant discrepancies exist between the source and target domains. To address these challenges, this paper proposes a Failure Mechanism-Driven Multi-Adversarial Domain Transfer Learning algorithm. The core of this method is the deep integration of physical prior knowledge with data-driven models. It first pre-trains a deep network using simulated vibration signals derived from the dynamic equations of bearing failures to establish a robust initial knowledge base. Subsequently, a multi-adversarial network framework is designed that includes both global and fine-grained class-level alignment, and introduces a knowledge loss function guided by physical principles, aiming to minimize inter-domain discrepancies while effectively suppressing negative transfer. Experimental results on two public bearing datasets show that the proposed method achieves average diagnostic accuracies of 88.15% and 96.74% on different transfer tasks, representing an improvement of up to 4.55 percentage points compared to existing mainstream domain adaptation methods. This research provides a more robust and effective technical pathway for the intelligent diagnosis of bearings under complex operational conditions. • We address cross-domain bearing fault diagnosis via simulation-guided transfer learning. • We pre-train a network on simulated vibration signals from dynamic equations. • Our method improves accuracy and reduces errors in fault diagnosis. • Integrating failure mechanism insights boosts transfer learning performance.