Industrial equipment fault intelligent diagnosis method, device and equipment and storage medium

Abstract

The application relates to an industrial equipment fault intelligent diagnosis method, device, equipment and storage medium. The method comprises the following steps: acquiring industrial equipment multi-source observation data, generating multiple groups of candidate directed acyclic causal graphs by using a scoring type causal discovery algorithm, and fitting a structural causal model; generating counterfactual queries and calculating counterfactual effects and effect variances of the counterfactual queries under each candidate graph; selecting a diagnostic counterfactual query according to the variances; inputting the diagnostic counterfactual query and the corresponding effect into a large model audit to obtain a structured feedback containing an expected symbol and a confidence degree; constructing a counterfactual consistency regularization term based on the feedback, optimizing an original scoring function, and iteratively optimizing to obtain an optimal causal graph, and analyzing a fault propagation path to realize root cause positioning and diagnosis. The method does not require real intervention data, can improve the reliability of causal structure recovery in small sample and structural ambiguity scenarios, accurately diagnoses faults, and can be integrated in a plug-in mode, and has low engineering landing cost.

Images