Heat exchangers, important components of nuclear power plants, are susceptible to efficiency reduction caused by internal fouling, which triggers significant energy losses, localized overheating, and catastrophic failures such as tube cracking or rupture, thereby threatening operational safety. To meet this challenge, more experimental data sets need to be collected and advanced diagnostic algorithms need to be employed to effectively identify the fault of heat exchanger efficiency degradation. This study introduces an innovative experimental design to simulate and analyze efficiency degradation mechanisms, with qualitative evaluations of temperature differentials, flow rates, and heat transfer performance confirming the validity of the experimental setup. Furthermore, the diagnostic capabilities of multiple machine learning algorithms—logistic regression, random forest, and support vector machine (SVM)—were rigorously assessed. Results revealed that the random forest algorithm achieved optimal fault detection accuracy (99.97%), significantly outperforming naive Bayes, multilayer perceptron, and gradient boosting models, all of which exhibited accuracies below 98%, rendering them inadequate for structural fault diagnosis in heat exchangers. This study not only enriches the experimental data, but also provides a valuable reference for the selection of diagnostic algorithms for heat exchanger capacity degradation, and provides a basis for subsequent practical applications of heat exchanger diagnosis.