TEM-SDRF: A Novel Approach for Machinery Fault Diagnosis Amid Dual Imbalances in Fault and Noise Samples

Fault diagnosis model for imbalanced samples plays a critical role in ensuring the reliability and efficiency of engineering system. However, existing methods usually consider only singular imbalances, such as imbalanced samples, making it difficult to tackle the occurrence of multiple imbalanced phenomena in actual applications. To address this issue, we propose a fault fusion learning method based on adaptive dynamic receptive fields. Specifically, a trilinear extraction mechanism (TEM) with the selective dynamic receptive field (SDRF) is designed to learn linear and nonlinear features in intrinsic mode function, which composed by the complementary ensemble empirical mode decomposition with adaptive noise (CEEMDAN) from imbalanced samples with imbalanced noise. The effectiveness of the proposed method has been validated by two fault case studies. Comparative results indicate its promise for fault diagnosis, as it significantly reduces training time while achieving more than 99% prediction accuracy in various tasks.