Predicting stress-strain properties of hybrid DLP composites: the role of photovoltaic waste granularity and glass powder content via long short-term memory modeling

The exponential accumulation of end-of-life photovoltaic (PV) waste necessitates sustainable recycling strategies beyond conventional disposal. This study proposes a novel framework integrating PV waste valorization into Digital Light Processing (DLP) composites with data-driven deep learning, bridging the gap between material upcycling and advanced characterization. Commercial photopolymer resins typically exhibit limited mechanical performance, motivating the incorporation of waste-derived reinforcing fillers. Experimental results show that coarse PV particles (sieve 200–400) combined with Glass Powder significantly enhance mechanical properties, achieving a maximum tensile strength of 27.77 MPa and flexural strength of 33.75 MPa through synergistic mechanical interlocking and crack-bridging effects. In contrast, finer particles (sieve 600) provide superior shear stability across filler variations by minimizing UV attenuation and preventing interfacial debonding at higher concentrations. To reduce reliance on costly destructive testing, Long Short-Term Memory (LSTM) networks were developed to predict full non-linear stress–strain behavior. The model effectively captures path-dependent material responses while filtering experimental noise. Computational analysis indicates that deformation-specific datasets with the Adamax optimizer yield optimal convergence. The optimized architectures (1F, 3B, and 2C) achieve strong predictive accuracy with R² values exceeding 0.9, demonstrating robustness despite limited data. This study enables efficient material design, supporting sustainable manufacturing and digital twin development.