Research on Industrial Sensor Self-Diagnosis Method Based on Redundancy Relationship Analysis

As the scale of industrial systems continues to expand, the workload involved with maintaining sensor health has surged. This has led to an increasingly urgent need for autonomous diagnostic capabilities in sensor measurement systems. However, the absence of reference values presents a major challenge in achieving sensor self-diagnosis under unsupervised conditions. Therefore, this study proposes a sensor self-diagnosis design method based on the integration of binary convolutional autoencoder and Bayesian inference (BCAE-BI). First, to extract the redundancy relationships among sensor measurement data, a binary convolutional autoencoder (BCAE) was used to encode the measurement data features into binary form. Second, using the encoded binary data, fault statistical models for sensors were established for both cases with and without reference values, considering both the statistical independence and dependence of sensor measurement data. Under conditions without reference values, a Gaussian Copula is used to construct a multivariate statistical dependency model to reduce the parameter scale, while the maximum mutual information coefficient is applied to evaluate dependencies between parameters. Bayesian inference (BI) is then utilized to achieve fault self-diagnosis and quantify the uncertainty of diagnostic results. Finally, using the nickel flash smelting system as an example, experiments were conducted for the self-diagnosis models in both modes. The experimental results demonstrate the feasibility and effectiveness of the proposed method in both scenarios.