Self-Diagnosis Methods for Industrial Sensors With No Reference Values

In industrial environments, sensor fault diagnosis is often hindered by a lack of reference values. To address this issue, this paper proposes a reference-free self-diagnostic method based on variational Bayesian inference. The proposed approach integrates a causal discovery–based quantitative evaluation framework for the redundancy relationship λi→jwith a physically interpretable β -binomial mixture model, thus effectively combining the structural interpretability of causal analysis with the precision of statistical modeling. The redundant relationships among sensors are first quantitatively evaluated using causal discovery techniques, and a fault statistical model is constructed based on the β -binomial distribution. Variational Bayesian inference is then employed to design a self-diagnostic algorithm for sensor measurement systems with and without reference values. The algorithm enables self-learning of sensor health states by leveraging observed data in the absence of reference values. Finally, a temperature measurement system in a nickel flash smelting furnace is used as a case study to validate the feasibility and effectiveness of the proposed method, with a 10% improvement in robustness.