Modelling, Sensitivity Analysis and Optimization of System Parameters of Edible Oil Refinery

Abstract

This research paper discusses the use of machine learning for sensitivity analysis in the edible oil refinery industry. The edible oil refinery consists of four units, and subsystem A (Cleaning) contains parallel subcomponents that enable it to function in reduced capacity when some of its components fail. As a result, the system operates in reduced capacity and fails entirely when unit A fails. There is single repairman who is available for 24x7. The failure and repair time distributions are distinct and constant for each unit. Subunits B (Shelling), D (Crushing/Ribbing), and E (Expeller Pressing) are connected in series, so, if any of them malfunctions, the entire system fail. System parameters' behavior for various repair and failure rates is discussed graphically and in tabular form. Comparison of the Performance of system parameters is carried out using linear Classifier and Logistic Regression and graphs are drawn. From the comparative analysis of models, it is concluded that the linear classifier is better than logistic regression for MTSF, Busy Period of server and Availability.