A NEURAL NETWORK APPROACH FOR EVALUATION of SURFACE HEAT TRANSFER COEFFICIENT

Abstract

An artificial neural network (ANN) approach for tackling the inverse heat conduction problems was explored - specifically for the determination of surface heat transfer coefficient at the liquid-solid interface using the temperature profile information within the solid. Although the concept is quite generic, the specific cases considered have a particular relevance to food process engineering applications. The concept was tested with two geometric shapes: a sphere and a finite cylinder, the former representing the simplest geometry and the latter representing a cross product of an infinite cylinder and an infinite plate. In developing the ANN model, two approaches were used. In the first one, the ANN model was trained to predict the surface convective heat transfer function, Biot number (Bi) from the slope coefficient (m) of temperature ratio curve under varying boundary conditions. The associated mean relative prediction errors were as high as 5.5% with a standard deviation of 8%. In the second ANN approach, m was related to tan -1 (Bi) which significantly improved the model's predictive performance. The second ANN model could be used with Biot numbers up to 100 with a mean error less than 1.5% for either of the two geometries. Heat transfer coefficients evaluated using the developed ANN model were in agreement (<3% error) with those calculated using conventional numerical/analytical techniques under a range of experimental conditions.