Estimation of lube oil viscosities on a vacuum distillation column

Abstract

The key product specifications on a vacuum distillation column side draw products are viscosity, density, flash and color, although viscosity is usually the most limiting quality. There are online viscosity analyzers available on the market, but their prices are quite high and their reliabilities could be quit low. Possible solution to this problem is development and application of so-called soft sensors, which estimate lube oil viscosities based on available easy-to-measure variables. In this paper we consider the application of neural networks for viscosities estimation of lube oils on the vacuum unit at INA Rijeka lube oil refinery. The proposed soft sensors, based on neural networks, are of nonlinear finite impulse response structure, where their inputs are temperatures and flow-rates of the distillates. Although developed soft sensors do not demonstrate high accuracy they can be used to track trends of the output viscosities, and based on that information plant operators can take proper actions with more certainty. It is to expect some improvements in sensors behavior with arrival of new input/output data.