Process Monitoring Using Canonical Correlation Analysis

Modern industrial systems are growing day by day and their complexity is increased due to increasing demands on production quality system performance and economic operation.To cope with these issues data-driven techniques like principal component analysis(PCA)and partial least square(PLS)and canonical correlation analysis(CCA)used for fault diagnosis and process monitoring for systems.They assume that data to be investigated are not self-correlated.However,most large-scale chemical industrial plants are nonlinear in nature so these techniques not cope with them,invalid in nature.To fulfill this gap there is need to develop an algorithm that can manage these abnormalities of the process.The demands of industrial products are increasing rapidly so different adaptable techniques are being proposed.Canonical Correlation Analysis(CCA)is multivariate data-driven methodology which takes input-output both process data into consideration.This thesis deals with the implementation of data-driven techniques,like principal component analysis(PCA)partial least square(PLS)and canonical correlation Analysis,for process monitoring in Tennessee Eastman(TE).Principal component analysis(PCA)is the most commonly used dimensionality reduction technique for detecting and diagnosing faults in chemical processes.Although PCA contains certain optimality properties in terms of fault detection and has been widely applied for fault diagnosis,it is not best suited for fault diagnosis.Canonical correlation analysis(CCA)has been shown to improve fault diagnosis in the chemical process as compared to PCA and PLS.Multiple faults are detected simultaneously by using T~2–statistics and Q–statistics(SPE).In this study,we compare these techniques and find CCA is more efficient than PCA and PLS.Performance evolution is illustrated by fault detection rate.In this work,it is shown that the FDR rate of CCA is higher than PCA and PLS,and further compares their performance from the application viewpoint,an industrial benchmark of Tennessee Eastman process is utilized to analyse the effectiveness of all the discussed methods.The study results are dedicated to providing a reference for achieving successful PM-FD on large-scale industrial processes.