Thermal analysis of shell and tube heat exchangers using artificial neural networks

Abstract

The shell and tube heat exchangers are most commonly used industrial equipment to transfer heat from one fluid to another. The design process is specified by Tubular Exchanger Manufacturers Association (TEMA). The initial design has to be iteratively optimized for increasing the heat transfer, minimize the pressure drop and reduce the fluid pumping power. LMTD method is chosen for its simplicity and quick analysis. The design procedure if approached by Finite element method, needs high computing power and tedious to converge. A feed-forward artificial neural network is set up to simplify the iterative nature of the design process. This also gives the necessary quick design iteration cycles to reach the optimized design. A design space is created with heat exchanger parameters, and the feed forward network is trained with semi-empirical data. The trained network is used in the design performance evaluation. This approach shows promise of quick design changes and can accommodate variable thermo-physical properties of fluids, and can be trained for different fouling patterns in the heat exchangers from real time data. The neural network can predict the steady state performance within the design space and results match well with LMTD calculations. Subsequent to the steady state analysis, dynamic modeling is attempted. A neural network method is used to reduce the complication of the model. Simplified mathematical model is used initially to train the network. It is found that the feed forward networks can predict the dynamic behavior, but it needs additional parameters to improve its predictions.

Keywords: Shell and tube heat exchanger, feed forward, artificial neural network