Real-time estimation of low-density polyethylene product quality with machine learning algorithms

Plastics industry is one of the major actors of the economy. Plastic, which is the raw material of many products, has different derivatives. This study deals with to estimate the product quality of LDPE-T (Low Density Polyethylene) in real time which is one of the raw materials of plastic, which is the most frequently used material in daily life. LDPE-T, which is obtained by high pressure process of polyethylene, is produced in the factory in a completely insulated and continuous system. The quality (first, second, third class products) of the products like LDPE-T is determined by the multiple chemical and physical lab analysis. From production to quality estimation, it may take 5-7 hours. In order to predict the real time quality of the manufacturing product, process data from the reactor section of the plant and previous years laboratory results consolidated and machine learning algorithm built. The objective function of the machine learning algorithm is to predict the quality of the real time production and show the quality to plants technicians. To tackle this problem, ensemble boosting model is proposed in this paper, and the best prediction is obtained with XGBoost algorithm which is 97% overall accuracy rate. XGBoost is an optimized distributed gradient boosting library designed to be highly efficient, flexible and portable. The results indicated that quality of the LDPE-T production can be successfully predicted using this relatively straightforward machine learning tool.