(229g) A Machine Learning Approach for Examining Flushing Operations in Packaging Plants

A typical oil manufacturing industry formulates over a thousand unique oil products by blending different base oils with numerous groups of additives. Each of these products is packaged in a container at the industry and it is not practical to have a separate production line for each product. So, the system of blending vessels, ancillary equipment, and associated pipeline network must be reused multiple times for numerous batch productions. Each product has a given set of product quality requirements and the blending of two oils must be minimized. In general, the use of external agents to clean the line between filling operations is not used since this introduces additional contaminants to the oil. So, the equipment is cleaned between filling operations by using the upcoming product batch to clean or flush the filling equipment and associated pipelines. This results in the formation of commingled/mixed oil that is regarded as a low-value product since it doesn’t meet product specifications. This project reports on our work in identifying optimization opportunities in the filling operations. To achieve desired product specifications, attain quality targets, and maximize the productivity of assets, it is of paramount importance to enhance process control, minimize human errors and improve the resource management footprint of these industries. To this end, we present a machine learning approach to learn from the data of existing operations and strategically optimize the flushing operations.