Empirical modelling, control and optimization of lubricant recycling unit

This study consists of two phases. In the first phase, four different forecasting approaches were applied to the chemical processes found in the lubrican recycling unit. The predictive methods used are Multiple Linear Regression, Multiple Nonlinear Regression, Artificial Neural Networks and Ridge Regression. The input variables of models are temperature and pressure, and the output variables of the unit are quantity and quality. Both quantities and qualities of useful end products are classified as light gas oil, medium gas oil, and heavy gas oil and are kept in the daily record of the chemical unit. Four different forecasting models have been compared in term of their performance. The artificial neural network approach showed best predictive model with the least error rate. In the second phase of the study, the non-linear mathematical optimization model generated the optimal temperature and pressure values in order to maximize the quantity and quality of light, medium, and heavy gas oils. Thus, optimum parameters have been obtained to help eliminate at least 4300 litres of wasted oil per week and production has been increased from 57% to over 70%. As a result, it has been observed that controlling the chemical processes by nonlinear mathematical modelling increases the production quantity. Key Words: Empirical modelling, Multiple Linear Regression, Multiple Nonlinear Regression, Artificial Neural Networks, Ridge Regression.