Research on sensor fault diagnosis based on multisource information fusion decisions considering the characteristics of arch dams

Because the operating conditions of high dams are becoming increasingly complex, these dams are prone to the generation of anomalous monitoring data caused by the coupling of structural damage and sensor faults. If a fault is not found and eliminated, it will induce monitoring data distortion and compromise data reliability, which will affect the assessment of dam safety in a data-driven manner. Therefore, this study presents a multisource information fusion method for sensor fault diagnosis and faulty data reconstruction that considers the structural characteristics of arch dams. By utilizing the correlations of the deformation response between the same elevation measurement points of the arch dam, the similarity characteristics of the sensor group are established to obtain threshold lines for sensor faults. The probability of a sensor fault is determined by whether the temporally similar features exceed the threshold lines. A new Tanimoto evidence similarity measurement is subsequently introduced to resolve evidence conflicts to comprehensively evaluate the probability of structural damage to a dam. Once the faulty sensor is identified, Bayesian–Markov Chain Monte Carlo (MCMC)–Gibbs sampling is used to quickly identify the corresponding fault types, and the faulty monitoring data are more accurately reconstructed through the posterior distributions of the parameters obtained by the model. A case study of a proposed DGS project shows that there is a 91.41% probability that the project has a sensor fault, and it is determined that sensor PL 10-1 has a fault based on multisource information. By comparing the Bayesian–MCMC model with the traditional statistical regression model for reconstruction inversion, the effectiveness of the proposed method is demonstrated.