Bayesian networks approach on intelligent system design for the diagnosis of heat exchanger

Abstract

The heat exchanger highly influences the series of cooling processes. Therefore, it is required to have maximum performance. Some of the factors causing a decrease in its performance are increased pressure drop in the Plate Heat Exchanger (PHE), decreased output flow, leakage, flow obstruction, and mixing of fluids. Furthermore, it takes a long time to conclude the diagnosis of the performance and locate the fault. Therefore, this study aims to design an intelligent system for the performance diagnosis of the PHE using the Bayesian Networks (BNs) method approach. BNs are applied to new problems that require a new BNs network model. The system was designed using MSBNX and MATLAB software, comprising several implementation stages. It starts by determining the related variables and categories in the network, making a causality diagram, determining the prior probability of the variable, filling in the conditional probability of each variable, and entering evidence to analyze the prediction results. This is followed by carrying out a case test on the maintenance history to display the probability inference that occurs during pressure drop on the PHE. The result showed that the BNs method was successfully applied in diagnosing the PHE. When there is evidence of input in the form of a pressure drop, the probability value of non-conforming pressure-flow becomes 61.12%, PHE clogged at 73.59%, and actions to clean pipes of 70.18%. In conclusion, the diagnosis carried out by the system showed accurate results.