Quality Prediction Improvement through Adaptive Nonlinear Principal Component Regression

The purpose of this paper is to present the predictor improvement for the refined palm oil quality based on the adaptive multivariate statistical process control. The time-varying behaviour of the palm oil refinery process has made it difficult for the industrial personnel to monitor and sustain the production of high quality refined palm oil. It will be very costly for the palm oil industries to repeat the refining process for the low quality refined palm oil, to meet the customer's preference in the market. Alternatively, the quality of the refined palm oil can be measured before the process through a systematic quality monitoring, where the information from the quality analysis and process condition is integrated to develop an efficient quality prediction tool. The hybrid of timeseries expansion methods namely recursive window (RW) analysis and exponentially weighted recursive window (EWRW) analysis along with the nonlinear principal component regression based on the nonlinear iterative partial least squares algorithm (NIPALS-PCR) are proposed to develop the adaptive prediction model. The predictor coefficient is then used to predict the refined palm oil quality based on the input quality and process variables. Through the validation with the online data, both NIPALS-PCR RW and NIPALS-PCR EWRW perform better than the NIPALS-PCR static, where the prediction is improved by 95 %.