Fault detection method with adaptive fusion of dual-space features
LIU Mei-zhi
KONG Xiang-yu
AN Qiu-sheng
LUO Jia-yu
Abstract：Due to the complex structure of the large complex industrial processes,the process variables often exhibit hybrid correlations.A single model cannot accurately represent the hybrid correlations between variables,resulting in a large number of missed alarms or false alarms in the fault detection.To address this problem,a fault detection method with adaptive fusion of dual-space features is proposed.Firstly,the Gaussian linear features and non-Gaussian nonlinear features are extracted in the original data space and the residual kernel space,respectively,using a hierarchical feature extraction strategy.Then,the Bayesian inference is utilized to convert the monitoring statistics from different spaces into failure probabilities,and an adaptive probabilistic weighting strategy is designed to construct the total probabilistic statistical indices for monitoring the process operation status.Finally,several experiments on a numerical simulation and the Tennessee Eastman benchmark process are presented to demonstrate the feasibility and effectiveness of the proposed method.
Keywords：fault detectionfeature extractionhybrid correlationsBayesian inferencestatistical index
Publication Date：2025-09-30
Online Publishing Date：2025-10-28(First online date of this platform, not the publication date of the document)
Pages：12( 1721-1732 )