Deep learning-enhanced multi-modal modeling for electrosorption performance prediction via Nyquist plots.

Area of Science:

• Environmental Science
• Electrochemistry
• Materials Science

Background:

• Flow-through electrosorption systems offer integrated ion recovery and electrochemical redox functions.
• Graphitized nanodiamond electrodes are crucial for these systems.
• Optimizing Cr(VI) removal and Cr(III) regeneration requires accurate predictive models.

Purpose of the Study:

• To develop a novel electrosorption-based multi-modal model for predicting Cr(VI) removal and Cr(III) regeneration efficiency.
• To evaluate the impact of electrode annealing temperature on model performance.
• To compare the predictive accuracy of Artificial Neural Network (ANN) and Convolutional Neural Network (CNN) based models.

Main Methods:

• Electrochemical characterization using Nyquist plots.
• Development of ANN and multi-modal ANN-CNN models incorporating process parameters and annealing temperature.
• Shapley Additive exPlanations (SHAP) value analysis for feature importance.
• 2D simulations for process optimization.

Main Results:

• Simple ANN models showed limited accuracy (R²: 0.344-0.829) without annealing temperature.
• Incorporating annealing temperature improved ANN performance (R²: 0.716-0.814).
• The multi-modal ANN-CNN model achieved significantly higher prediction accuracy (R²: 0.950-0.977).
• SHAP analysis emphasized the importance of electrode durability tests.

Conclusions:

• The multi-modal ANN-CNN model demonstrates superior predictive capability for electrosorption processes.
• Electrode annealing temperature is a critical parameter influencing system performance.
• CNN integration is essential for high-accuracy predictions in electrosorption.
• Electrode durability is vital for the long-term efficiency of flow-through electrosorption systems.