الفهرس 
الفصل الاولOnly 14 pages are availabe for public view
 |  | from | 156 |  |  |
| | | from | 156 | | |
Abstract
This work presents a predictive method for industrial heat exchanger efficiency using artificial neural networks (ANN). The model developed by ANN can predict the maintenance time before the critical fouling time is reached. In fact, fouling affects the performance of the heat exchangers
and causes additional costs due to sudden mechanical faults. Therefore, the fouling behavior has to be studied, which is very complex for heat exchangers because of the difficulty in monitoring
the growth of fouling. Many methods used to estimate the efficiency and evaluate the performance of industrial heat exchangers. This work introduced important approaches, C factor (experimental
method) based on pressure DROP and volumetric flow rate. The C-factor approach gives relatively accurate results but takes a lot of long reading. In addition, there is Another approach; which is the
traditional methods of thermal analysis, Wilson plot, and modification of Wilson plot are complex mathematical models because it needs many assumptions, design aspects, and complexity of the heat exchanger’s geometric configuration, which gives approximate results. Finally, the approach, ANN (Modern Method) is a highly sensitive technique for evaluating the performance of industrial heat exchangers. This ANN model uses a Feed-Forward Neural Network configuration with Bayesian Regularization algorithm. Using a span of time of running the heat exchanger to taking readouts and measurement variables for training and test processing to build an architectural neural network model. It was found that maximum deviation between the ANN and thermal analysis
results and the comparison with the C-factor experimental results is 9.8% and 33.6%, respectively. Based on, the good results this assisted model reference strategy of ANN can be used to predict the efficiency of the heat exchangers. Additionally The ANN is flexible and capable to update in
terms of new sets of weights and biases when the validity range changes in the same network.