SDP-Based Quality Monitoring with Application to the Tennessee Eastman Process (TEP)

Development and implementation of advanced monitoring and control techniques requires measurement of variables which cannot be determined physically or difficult to measure. Soft sensors can be used as a relatively inexpensive alternative for hardware sensors as a suitable solution in the process industries by estimation of easy-to-measure variables using hard-to-measure variables. In this study, design of data-driven soft sensor based on state-dependent parameter modeling method by using local instrumental variable (LIV) have been presented to predict quality variables in Tennessee Eastman (TE) process. Unlike other soft sensor modeling methods, the state dependent parameter modeling method has simple structure and often requires fewer input variables. Moreover, state dependent modeling method using local instrumental variable can identify influencing variables which have been affected the target variables. The performance of identifying technique and proposed soft sensors has been investigated on Tennessee Eastman process. In the present study, LIV based Soft sensor models have been developed using MATLAB software to predict concentration of A and E components. The evaluation results of the proposed models on the test data set report that the root mean squared error (RMSE) for concentration of components A and E are 0.3191 and 0.0174, respectively. The proposed LIV model reduced prediction error (RMSE) for the concentration of component E by 98.18% and 97.6% as compared to Partial Least Squares (PLS) and Dynamic Inner Partial least squares (DiPLS) methods, respectively.