Using a Neural Network Approach to Predict Deposits on the Surfaces of Heat Exchange Equipment

Abstract

This work proposes a neural network (NN) approach for predicting the following values: the heat transfer coefficient at the point of interest in the operational period of plate heat exchangers (PHEs), and the time-point to reach the lower allowable limit of the heat transfer coefficient. In this approach, neural network models replace complex mathematical modelling that used systems of differential equations and matrices of heuristic coefficients to calculate the flow rate of deposits on PHE plates, which required the involvement of serious computing resources. Training a feed-forward neural network (FFNN) on a small dataset simulated in the vicinity of reference points obtained by industrial measurements showed the proper coefficient of determination R² = 0.99 (accuracy) of the short-term prediction forecasts and for operational evaluation of the heat transfer coefficient due to the static type of NN.