Hygrothermal degradation failure of poly(methyl methacrylate): chemo-mechanical correlations and image-based deep learning prediction

Long-term exposure of Poly(methyl methacrylate) (PMMA) to hygrothermal conditions leads to significant mechanical deterioration driven by physical aging, hydrolysis, and surface defects. This study characterizes the degradation of PMMA aged at 85 °C and 85% relative humidity over 16 weeks using thermal, spectroscopic, and morphological analyses. Although chemical degradation indices derived from FT-IR provided a baseline for property estimation, they exhibited limitations in capturing the localized, stochastic surface damage that initiates fracture. To overcome this, a convolutional neural network (CNN) was developed to predict mechanical properties directly from optical surface micrographs, eliminating the need for destructive testing. The image-based CNN model demonstrated high fidelity in predicting tensile strength, strain at break, and tensile toughness (U_T), significantly outperforming traditional regression methods. This study establishes a practical, nondestructive methodology for evaluating the service life and structural integrity of hygrothermally aged polymers.