Development of Multi-Layer Feed Forward Back Propagation Neural Network Model for Shell Side Thermal Rating of Shell & Tube Heat Exchanger

ABSTRACT

Thermal rating relates to prediction of the performance of the heat exchanger for given working condition. The shell side sizing / rating problem is particularly complex with lot of correlations involved for a reasonably accurate design or rating irrespective of the approach followed. There are a good deal of analytical approaches for design & sizing of shell side viz. Kern method, Tinker method, Palen & Taborek approach, Bell or Delaware method etc. But each of these methods has certain assumptions & associated shortcomings. Further, each method involves understanding of lot of correlations & large experimental work for establishing the constants used in these correlations. Typical analytical procedure for shell side thermal rating of a shell & tube heat exchanger accepts flow rates of fluids, inlet temperatures of the fluids, flow pressures, fluid properties and various practical fouling factors and using various correlations, tables and/or charts, evaluates either the heat duty possible or the outlet temperatures under given operating conditions along with the associated pressure drops.

For such systems which involve complex equations but for which significant data is available, model free methods like artificial neural networks provide powerful & robust means to reduce uncertainty using pattern based learning. Use of artificial neural networks and simulated artificial neural networks for design & rating of heat exchangers has received great impetus presently. A neural network model fits into the scheme of things by replacing the complex calculations & presenting an equation free model. Development of an artificial neural network model basically requires the optimization of network architecture and selection of proper network functions. Beginning with a simple 3-layered neural network model with six cells in the hidden layer and random initialization of weights & bias, an extensive study is conducted to develop a predictive artificial neural network for the thermal rating of the shell side of a single shell single pass type of shell and tube heat exchanger. Such a neural network can predict the output parameters achievable with a set of input parameters. The results of the neural network are compared with the analytical results.

---

Full Text (PDF)

© 2010 Research Publishing Services