Explainable Deep Learning-Guided Screening of LiNO₃-Based Molten-Salt Phase Change Materials for Lithium-Ion Battery...

LiNO₃-based high-temperature phase change materials (PCMs), owing to their specific advantages, e.g., high thermal stability and high latent heat, have been recognized as promising candidates for mitigating or preventing thermal runaway in lithium-ion batteries (LIBs). However, the large variety of possible formulations necessitates the rapid and efficient selection of the optimal PCM to enhance battery safety. In this study, 12 LiNO₃-based PCMs were characterized using laser-induced plasma spectroscopy (LIPS), and an interpretable deep learning framework based on a self-attention-enhanced one-dimensional convolutional neural network (1D CNN) was employed to accurately discriminate among the samples and reveal the relationships between spectral features and thermal properties, namely, onset temperature and latent heat. Comparisons with conventional approaches, including principal component analysis (PCA)-based dimensionality reduction combined with support vector machine (SVM) classification and global variable importance analysis, highlight the superior performance and interpretability of the deep learning approach in linking spectral signatures to material performance. Our results indicate that the high-performing PCM candidates exhibit lower onset temperatures, higher latent heat, and the strongest K/Na spectral intensity ratios. Consequently, the Na-normalized K emission intensity is identified as the rapid spectroscopic probe for screening optimal LiNO₃-based PCMs, providing a simple, fast, and potentially transformative complementary approach to traditional thermal analysis methods for PCM screening.