Development of a new method for measuring multi-parameter of NaCl solution film based on absorption spectroscopy and interpretable convolutional neural network

Abstract

Seawater desalination is an important approach to addressing water scarcity. In its core process, falling film evaporation, simultaneous measurement of film thickness (L), concentration (C), and temperature (T) is essential for improving desalination efficiency, but it remains challenging. In this study, a new method for measuring multi-parameter of NaCl solution film was proposed, integrating absorption spectroscopy and interpretable one-dimensional convolutional neural network (1D-CNN). First, single-channel (SC), dual-channel (DC), and three-channel (TC) 1D-CNN models were constructed based on absorption spectra of NaCl solution under various conditions. An optimal TC-1D-CNN (Opt-TC-1D-CNN) model was obtained by introducing noise-based data augmentation. It was found that the normalized root mean squared errors (nRMSE) of the predicted L, C, and T are 2.26 %, 3.80 %, and 6.52 %, respectively. Model interpretability was then analyzed using the SHapley Additive exPlanations (SHAP) method to explore the physical relevance between key wavelengths and characteristic absorption features of NaCl solution. In addition, the proposed approach was compared with the previously developed inversion algorithm and applied to investigate NaCl solution film evaporation processes under different conditions. It was found that the Opt-TC-1D-CNN model shows good agreement with the imaging method and infrared thermal imager throughout the entire evaporation processes. This study provides a new method for simultaneous and accurate measurement of film parameters, with potential to support the design and optimization of desalination equipment. It can also provide theoretical guidance for future studies of other falling film processes in fields such as energy, chemical engineering, environmental monitoring, and biomedicine.