Toward Online Explore of Concept Drift for Fault Detection of Chemical Processes

Abstract Conventional fault detection and diagnosis (FDD) systems of chemical process ignore inevitable effects of Concept Drift (CD) on the FDD performance. Updating methods of FDD, generally, do not provide information for monitoring severity changes of the CD in the processes. Hence, this study explores CD monitoring using Hoteling’s T 2 and Q statistic. Two Transformation Functions (TF) (Hotelling’s T 2 and Q statistic) are obtained for normal and faulty conditions of the initial dataset. These TFs are implemented on CD-affected online datasets which have been classified with the incremental learned classifier. Comparison of transformed components of each class between initial dataset and online datasets based on Root Mean Square Error, RMSE, leads to monitoring of the CD in the process. Results present promising performance of the suggested framework.