Automation for Quality Assessment of Refined Petroleum Using Neural Network

The selective petroleum derivative refining technology is used extensively in the industrial production of refined pe-troleum. The manufacturing of quality lubricants depends on the stable quality of products obtained with the selective refining unit. The quality of petroleum oil as a product of selective refining of input oil is provided by controlling a number of refining unit operation parameters. These parameters include extraction temper-ature, N-methylpyrrolidone (extraction agent) concentration, etc. The effective values of these parameters are mainly determined by the chemical composition of the input oil. The quality of the obtained refined petroleum is controlled using a flow refractometer with readings confirmed by regular sampling and laboratory tests. One of the problems with such verification is the duration of la-boratory tests when off-spec products can be produced, leading to re-processing costs, etc. This paper suggests using a neural network-based IT solution for the analysis of flow refractometer readings adequacy. The neural network was trained on sets of flow refractometer data obtained at different times and confirmed by laboratory tests during the refining of one material type. The nu-merical experiment showed that the trained neural network can very accurately predict the refraction coefficient for normal refining. Simultaneously, the neural network can easily detect deviations that can be caused by various factors.