Assessment of fouling in plate heat exchangers using classification machine learning algorithms
Keywords:Fouling, plate heat exchanger, machine learning, classification, k-nearest neighbours, decision tree, Naïve Bayes
Plate heat exchangers (PHE) used in combi-boilers are continuously affected by small particles, both from the heating circuit and other components of the heating system. These particles can accumulate in the heat exchanger and create clogging that affect the performance of the heat exchanger over time by generating a insulating layer. To avoid unexpected blockage and other kinds of mechanical failure caused by unintended particles that originate from the pipeline and other components, it is crucial to design an effective predictive maintenance system for PHE used in the combi-boiler. In this study, the early stage of blockage in a PHE is investigated experimentally to minimize the field failure rate. The data is acquired from an experimental set-up in which just the PHEs are tested. The PHEs with the same plate pattern and different plate numbers are tested using varied flow rate and inlet temperatures as parameters. The overall heat transfer coefficient and fouling resistance are calculated to associate with the functionality of PHE. A comparison study of multi-classification algorithms has been investigated to present an algorithm which gives the most accurate model trained by experimental data. K-folds cross validation are studied using Naïve Bayes, k-nearest neighbours (kNN) and decision tree machine learning algorithms. As a result, the behaviour of overall heat transfer coefficient and fouling resistance in normalized time scale show the expected trends. The attempted models of machine learning algorithms result in Naïve Bayes predicting the classes of test data perfectly and it is followed by decision tree algorithm with an accuracy of 99.3% and kNN algorithm with 96.3%.
How to Cite
This work is licensed under a Creative Commons Attribution 4.0 International License.