Multimodal Data-Driven Prediction of PEMFC Performance and Process Conditions Using Deep Learning

The proton-exchange membrane fuel cell (PEMFC) is one of the important technologies advancing sustainable energy. However, predicting its performance and optimizing processes is challenging due to the complexity of integrating various types of data with interdependent variables. This study introduces a novel deep learning model using multimodal data that integrated convolutional neural networks (CNN) and deep neural networks (DNN) to address these challenges. The proposed model predicts the performance through the CNN model using cell images taken from the optical microscope, and based on this, generates multimodal data to predict the optimal process conditions for each performance through the DNN model. Trained on a diverse array of experimental data under various conditions, our model significantly enhances the reliability of performance predictions and optimal process determinations, evidenced by an R2 value of 0.83. Unique to this research, the AI model utilizes both PEMFC cell images and performance data, enabling automatic performance prediction and substantially reducing the need for individual cell measurements. By analyzing both morphological images and experimental data, our model accurately predicts optimal process conditions, overcoming previous integration challenges. This method not only facilitates the performance assessment process but also optimizes manufacturing operations, thereby increasing efficiency and production rates in PEMFC manufacturing.