Knowledge-guided multi-objective deep learning enables robust LIBS analysis for industrial magnesium alloy identification.

Laser-induced breakdown spectroscopy (LIBS) has emerged as a promising technique for rapid elemental analysis in industrial manufacturing owing to its capability for real-time, non-contact, and multi-element detection. However, severe spectral noise, matrix effects, and spectral overlap substantially limit the robustness and reliability of conventional LIBS data analysis under complex industrial conditions. To address these challenges, a knowledge-guided multi-objective deep learning framework is proposed for intelligent LIBS-based magnesium alloy identification. The proposed approach integrates hierarchical spectral feature extraction with Grey Wolf Optimizer (GWO)-based adaptive hyperparameter optimization while incorporating spectroscopy-informed prior knowledge into the feature learning process to improve interpretability and generalization. Unlike conventional optimization strategies that focus solely on classification accuracy, the proposed framework simultaneously optimizes classification accuracy, noise robustness, and computational efficiency. Extensive experiments, including multi-level noise evaluation, ablation analysis, Grad-CAM visualization, inference latency measurement, and stratified five-fold cross-validation, demonstrate the superiority of the proposed method. The optimized model achieves an average classification accuracy of 99.90%, exhibits exceptional robustness under severe spectral interference, and reaches saturated performance within 150 training epochs under a maximum training setting of 200 epochs with early-stopping strategy, and maintains an inference latency of approximately 45 ms. Furthermore, the extremely low cross-validation standard deviation (0.0002) confirms excellent model stability and generalization across different data partitions. These results demonstrate that the proposed strategy provides a robust, interpretable, and practically deployable solution for intelligent LIBS analysis in industrial material identification.