Modeling and application of industrial process fault detection based on pruning vine copula

Abstract

Industrial processes are usually nonlinear multivariate stochastic systems. Describing the distribution characteristics of variables through a vine copula model can well define complex correlation information. However, modeling based on vine copula involves computational complexity. This study proposes a method to prune vine copula and introduces an indicator to conduct pruning process. This technique constructs a simplified model by removing the weakly correlated component from the copula structure without reducing the accuracy of model. Lastly, fault detection for industrial processes based on pruning vine copula is conducted based on a generalized local probability monitoring index. Experimental results of the monitoring process show that the method can reduce the time of modeling and improve the effect of fault detection.