Optimized deep learning using hybrid PSO-GA on raw XRD images for accurate classification of material degradation

Abstract Accurate damage classification in materials using X-ray Diffraction (XRD) patterns is essential for ensuring structural integrity. This study introduces a novel hybrid deep learning and evolutionary optimization framework, PSO-GA, which integrates Convolutional Neural Network (CNN) with Genetic Algorithms (GA) and Particle Swarm Optimization (PSO) for fine-grained classification of material conditions: fresh, slightly degraded and severely degraded. Unlike conventional methods relying on manual features or tabulated data, our model processes raw XRD image plots, capturing spatial and intensity variations. CNNs automatically extract hierarchical features, while GA and PSO jointly optimize model hyperparameters to improve classification accuracy. The proposed method achieved strong performance with F1-scores of 0.925, 0.865, and 0.775 for Intact, Partially Damaged, and Damaged classes, respectively. Feature visualization using t-SNE and PCA revealed distinct clustering among damage categories, confirming the model’s discriminative strength. Optimization results showed rapid convergence with PSO and stable fitness improvement with GA, reaching a fitness value close to 0.99. The PSO-GAframework is robust, interpretable, and novel in its end-to-end learning from raw XRD plots, making it suitable for real-time, non-destructive testing in industrial applications.