Plant Performance Optimization Using Smartops

Abstract

The implementation of machine learning (ML) leads to the capability of identifying the non-linear relationship of input and output data. The growing availability and frequency of data collection has resulted in the widespread implementation of ML techniques for predictive modelling. The SmartOps^TM approach aims at providing ‘early warning’ in feedwater variations and optimizing the cleaning schedules using ML. This study was conducted at a water treatment plant, situated beside a main industrial plant in Malaysia. Taking the source from the pond water, the treatment plant is equipped with ultrafiltration (UF) and reverse osmosis (RO) membranes, and the product water was sent to the main plant for manufacturing purpose. The SmartOps^TM model outperforms other ML model in the prediction of pond water quality. The SmartOps^TM model shows relatively high r2 values at 0.97 in the prediction of the conductivity as compared to other ML model (r2 = 0.80). The process optimization using the SmartOps^TM is expected to save and minimize the operational costs, which include the chemical and electrical consumption of the treatment plant. The methodology offers a great potential for implementation in the operations of treatment plant, which could improve energy consumption and hence, making the plant more capable and optimized over time.