Evaluation of the Effectiveness of Heat Exchangers Using Machine Learning

Abstract

This study reports on the analysis of the operational performance of four plate heat exchangers, which are part of a cooling system in a glass factory. The overall objective of this work is to develop a diagnostic methodology that can be utilized by the partner company. The energy balance between the hot and cold fluids, the method of the logarithmic mean temperature difference, and the effectiveness-number of transfer units method were used to correlate the pressures, flow rates, and temperatures recorded in the process over a defined period. These techniques helped consolidate a statistical database with heat transfer rates, the amount of heat transferred, operation effectiveness, pressure drop, and other thermophysical parameters. To investigate the influence that pressure drop and operating equipment have on the system’s effectiveness, supervised machine learning models were developed. An inferential sensor was modeled using linear regression to correlate effectiveness with total pressure drop, yielding a coefficient of determination of 0.8663. A classifier was modeled using the K-Nearest Neighbors algorithm to identify which equipment is in operation, and its predictions achieved an accuracy of 0.9426 in relation to the observed operational classes. It is concluded that the system’s effectiveness decreases as the pressure drop increases and that each set of operating equipment can be identified based on the recorded signals.