Work

Experimental analysis and explainable machine learning modeling of heat and mass transfer in an ionic-liquid-based adiabatic dehumidification system

Year: 2026

Type: article

Abstract: To overcome the poor generalization and non-linear limitations of traditional correlations, this study investigated the heat and mass transfer characteristics of a counterflow adiabatic dehumidifier utilizing a novel ionic liquid. A finite-difference numerical model was developed to quantify the mass and heat transfer coefficients and the number of transfer units from experimental data. Subsequently, a comprehensive database combining experimental and literature values was constructed to benchmark a traditional least squares method (LSM) correlation against five machine learning (ML) models. The experimental results indicated that optimal system performance was achieved under an airflow rate of 3000 m 3 /h, with the energy factor ( EF ) being 3.42 kg/kWh. Reducing the airflow rate from 7000 to 3000 m 3 /h improved the EF by 70.1% because of an increase in residence time, despite a decrease in the total moisture removal rate. In performance prediction, the baseline LSM failed to capture nonlinear behaviors [coefficient of determination ( R 2 ) < 0.47]. Conversely, the CatBoost model demonstrated high accuracy and generalization ability, achieving R 2 values of 0.97 and 0.88 in predicting the Sherwood number and Nusselt number, respectively, with more than 87% of predictions falling within the ±20% error band. SHapley Additive exPlanations analysis revealed that the Reynolds number has dominant effects on the heat and mass transfer rates, whereas fluid-specific parameters such as Prandtl number also contribute significantly to model decision-making. These findings confirm that interpretable ML models can effectively distinguish the physical properties of ionic liquids and traditional salts, offering a robust framework for optimizing next-generation energy-efficient dehumidification systems.

Source: Thermal Science and Engineering Progress

DOI: 10.1016/j.tsep.2026.104810

Authors: Jia-Wei Zheng, Yu-Zhi Liu, Yu-Lieh Wu

Institutions: National Kaohsiung Normal University, National Chin-Yi University of Technology

Language: English

Topic: Adsorption and Cooling Systems

Subfield: Mechanical Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Affordable and clean energy

Open Access status: hybrid

Estimated APC paid: $3,000