Computational and Experimental Studies to Analyze Pilot-Scale Petroleum Packaging Facility

Petroleum packaging facilities process multiple lubricant oils per day to meet the required market demands. As the number of packaging lines are limited, they need to clean the lines between product change-overs. During the product change-over procedure, the lines are cleared to remove the previous product. However, due to the highly viscous nature of most petroleum products, there remains a nontrivial amount of residual oil in the packaging line, which can contaminate the new product. The solution is to flush the line with new product, which mixes with the residual product to generate commingled oil. Further flushing with the new product will clear the line of commingled oil and completely fill it with new product that meets the desired specifications. While this is a successful solution to the issue, the commingled oil created during this procedure is a downgraded product that does not meet company standards with little economic value.

The goal of this work is to accurately model the industrial packaging plant and apply a well-thought scale-down strategy for the construction of a pilot scale plant that predicts the mixing properties observed at the real plant. To address this, we started with a computer model that is designed to predict how different factors affect the flushing operation and whether the new flushes meet industry specifications. Utilizing the historical data and machine learning approach, our model was able to correctly predict 110 out of 116 flushes. Furthermore, the scale-down evaluations to pilot scale were accomplished using the Reynolds number, and the pipe diameter to length ratio. With the completion of the pilot plant of the packaging line, further experimental studies were conducted to better understand the actual mixing properties for products of varying specifications in the packaging line. While maintaining the same residual and flush product, the factors that were analyzed were flowrate, filter casing volume, and filter type. Residence time distribution curves were created for both plants. Dimensionless analysis was conducted to directly compare the pilot plant experiments to the industrial plant. Through this, we gained insights on how to alter the pilot plant to better mimic the packaging plant. We concluded that the filter volume is an important factor in modeling the packaging line, whereas the type of filter was not as important. Further tests are underway for other factors such as variable flowrates for laminar to turbulent regimes and the type of change-overs based on viscosity gradients.