Fault diagnosis of the hydraulic valve using a novel semi-supervised learning method based on multi-sensor information fusion

Abstract

Hydraulic systems are usually applied in large and complex engineering fields. For hydraulic systems or components in operation, it is difficult to obtain fault data with fault labels due to the high engineering cost. Therefore, a semi-supervised learning (SSL) method based on multi-sensor information fusion is proposed to obtain valuable pseudo label data to diagnose faults of the hydraulic directional valve in operation. In this method, the classification model is trained from a small amount of data with fault labels, thus generating pseudo labels for a large amounts of unmarked data. The contribution of this article is that a multi-sensor fusion algorithm is designed to obtain pseudo labels with high confidence, and an adaptive threshold model similar to generative countermeasure network is designed to intelligently generate thresholds for selecting pseudo labels instead of human intervention. Theoretical and experimental results show that the multi-sensor information fusion algorithm can obtain high confidence pseudo tags, the adaptive threshold model can screen effective pseudo tag samples by generating appropriate thresholds for accelerating the convergence of the classification model. In the hydraulic valve fault diagnostic test, after five iterations, the average diagnosis accuracy of this method can reach 99.72% and 99.00% respectively for different types of hydraulic valves in different engineering fields. This provides a new idea for developing intelligent hydraulic directional valve with self fault diagnosis function.