F-20 Neural Network Analysis of HTFU Crude Oil Fouling Database
Traditional parametric methods do not fully account for underlying fouling mechanisms, limiting analysis of data exhibiting high scatter. In this study, artificial neural network (ANN) methods were used to correlate a select database of 98752 data points from test runs with higher fouling propensity from the annular test section of the HTRI High Temperature Fouling Unit (HTFU). The resulting networks predicted (1) fouling resistance within an overall average error of 24.7 percent using back propagation (BP) with all scattered data, (2) fouling resistance within an overall average error of 15.6 percent using BP with smoothed data, and (3) fouling rate within an overall average error of 11.7 percent using a radial basis function (RBF) network.
These results can be used to establish threshold fouling parameters, as well as minimum no-fouling velocities as a function of surface temperature for different colloidal instability indices. A simple calculation method was developed to predict fouling resistance based on Reynolds number, Prandtl number, surface temperature, colloidal instability index, and time. The approach of this study should be expanded to a larger set of data and input variables.