Real Time Quality Assessment of General Purpose Polystyrene (GPPS) by means of Multiblock-PLS Applied on On-line Sensors Data

Abstract

In the petrochemical industry, in order to control the final product quality over time and to detect potential plant failures, the amount of lab (off-line) analysis performed every day is very demanding in terms of resources and time. Hence, at/in-line monitoring can be an efficient solution to decrease chemical wastes and operators’ efforts and to perform a fast detection of deviations from normal operative conditions. Moving toward this implementation requires both installation of analytical sensors and the development of models capable to predict in real time the quality parameters of the polymers based on both process and analytical sensors. The primary aim of the current work has been the development of real time monitoring models by advanced chemometric tools for the prediction of a General Purpose PolyStyrene (GPPS) quality property, fusing Near Infrared (NIR) and process sensors data. In the plant considered, in addition to standard process sensors, along the GPPS production line, operating in continuous, two NIR probes are installed in-line. After the arrangement of the available data in different blocks, aiming at studying the specific contribution of the two types of sensors and of the main phases of the process, Multiblock-PLS (MB-PLS) method was employed to fuse the different blocks and to assess which were the most relevant sensors and plant phases for the prediction of the two quality parameters. Good prediction performances were achieved, allowing identifying the most significant data blocks for the GPPS quality prediction. Moreover, prediction errors obtained by models computed without considering blocks of data belonging to the final stages of the process were similar to those involving all the available data blocks. Therefore, a good real time assessment of the GPPS quality can be obtained even before the production is completed, which is very promising in view of minimizing the number of off-line laboratory analyses.