Monitoring of heat exchanger fouling by integrating dimensionless entropy generation features and cost-sensitive neural network

Surface fouling in shell-and-tube heat exchangers severely degrades heat transfer efficiency and reinforces flow resistance. Traditional macroscopic monitoring indicators based on temperature or pressure drop are susceptible to operational fluctuations and suffer from hysteresis defects due to the asynchronous evolution of flow and thermal resistances. Given this challenge, a high-precision early fouling monitoring framework integrating thermodynamic entropy generation mechanisms with a data-driven approach is constructed in this study. A helium-water heat exchanger is employed as the test case. First, the accuracy of the helium-water heat transfer numerical model is validated using experimental data. Following the second law of thermodynamics, two dimensionless features, the Comprehensive Fouling Index (CFI) and the Flow-Thermal Degradation Ratio (FTDR), are established to accurately quantify the dominant mechanisms of system performance degradation. Furthermore, a cost-sensitive genetic algorithm-optimized multilayer perceptron prediction model is developed, with a third-order response surface methodology employed to augment high-fidelity samples. The results reveal that CFI exhibits strong temperature-independence with a maximum cross-condition deviation of only 1.8%, demonstrating robust decoupling from operational fluctuations. Concurrently, FTDR successfully identifies a critical mechanism transition threshold at approximately 2.25 mm, enabling a physics-interpretable distinction between flow-dominated and thermal-resistance-dominated degradation regimes. The proposed framework integrates these entropy-generation features with the cost-sensitive GA-optimized MLP. Consequently, on the independent physical blind test set, the proposed framework achieves R 2 = 0.9978 and RMSE = 0.045 mm, outperforming the standard BPNN (R 2 = 0.9898, RMSE = 0.097 mm), demonstrating considerable engineering value for high-precision fouling monitoring applications.