A Neural Network Assisted On-Line Cleaning of Heat Exchanger Network

Abstract

On-line cleaning of heat exchangers operating in a heat exchanger network (HEN) is aimed at preventing unnecessary losses of energy that can be recovered in HEN. If time behaviour of fouling in each heat exchanger can be estimated on the basis of past experience, then the optimal schedule of cleaning interventions can be determined by maximizing the objective function expressing the economic value of avoidable reduction in the energy recovery [1]. The crucial assumption for the presented paper is that on-line measurements of the mass flow and inlet and outlet temperature are available for each process stream. That made possible to evaluate fouling-induced reduction in the recovered energy flow using neural network model of HEN based on the measurements. The neural network (NN) model of HEN was applied in diagnostics of deposits influence on heat recovery in HEN by taking into account the time behaviour of fouling approximated by NN. The application of a neural network to the evaluation of changes in the energy flow recovered in a HEN has been tested using a simulated heat exchanger network as a reference. The simulation of HEN (implemented in MATLAB software) was based on a dynamic HEN model employing heat exchanger decomposition into interconnected cells whose overall dynamic behaviour is described by an array of lumped-parameter models. Computational aspects of the approach outlined above were studied on the example of a HEN featuring 26 process streams and 31 heat exchangers, operating in a crude distillation unit of 440 t/h processing capacity. A diagnostics of deposits influence on heat recovery in HEN makes it possible to attain a saving of about 5% of recoverable energy with the annual value of about 0,86 million USD.