Multiple-fault diagnosis using dynamic PLS built on qualitative relations

Abstract This study suggests the hybrid method of qualitative model and quantitative data-based method to improve diagnosis accuracy and resolution, and promote reliable diagnosis of multiple-fault. The proposed method is based on signed digraph offering a simple and graphical representation for the qualitative relationships between process variables. On local causal relationships of each variable in signed digraph, PLS model is built to predict the target variable of the causal relationships. In order to handle accurately process dynamics, the input of PLS uses the current value of source variables as well as the past values of source and target variables. The measured value and predicted one are compared to diagnose fault.