In industrial production systems, motors are widely used across sectors such as transportation and energy production.Current diagnosis methods primarily rely on signal analysis and single-model recognition, resulting in low precision.The current diagnostic methods mainly rely on signal analysis and single-model recognition, which have deficiencies in capturing complex noise environments and longsequence fault features, resulting in diagnostic accuracy often below 90% in actual industrial scenarios.This paper proposes a novel fault diagnosis model that deeply integrates a Belief Rule Base with a Bidirectional Long Short-Term Memory network.Diverging from conventional sequential or parallel structures, the proposed model introduces a "tunable input" mechanism, where the Belief Rule Base's real-time inference confidence directly gates and modulates the input channels of the Bidirectional Long Short-Term Memory, enabling dynamic, evidence-driven feature prioritization.The model processes uncertainty-embedded diagnostic signals and extracts longrange temporal dependencies to achieve accurate motor fault diagnosis.The test results indicate that the model achieves 99.82% precision, 99.45% recall, and 99.34% F1 score on the training set.In the loss value tests, the loss converges to 0.064 after 60 iterations, outperforming the comparison models.Overall, the proposed model effectively overcomes the deficiencies in reference and accuracy found in existing methods.It applies to multiple motor types and offers a new approach to motor fault diagnosis, enabling intelligent detection in industrial equipment.
Feiyu Liu (Thu,) studied this question.