Physical characterization and hybrid modeling of erosion wear in industrial waste-filled glass/epoxy composites using machine learning and computational fluid dynamics

Hybrid epoxy-short glass fiber (SGF) composites with Linz-Donawitz slag (LDS) were manufactured and the physical and erosion properties of the produced composites were investigated. As the LDS content increased from 0 to 22.5 wt.%, the measured density increased from 1.223 to 1.468 g/cm 3 , and the void fraction rose from 3.484% to 4.880%. In addition, the microhardness improved by 40.9%, increasing from 22.53 to 31.74 Hv. High inter-property correlation (r > 0.93) was found, and the theoretical density and void fraction showed an almost perfect correlation (r = 0.99). The LDS content was the most significant factor in the L 16 orthogonal array tests for erosion wear, and the average erosion rate decreased to 158.41 mg/kg with an increase in LDS content up to 22.5 wt.%.The highest efficiency condition (32 m/s, 90°, 22.5 wt. % LDS) showed a 73.8% reduction in the erosion wear rate. As LDS content increased, the erosion changed from brittle (peak at 90°) to semi-ductile (peak at 60°) and velocity sensitivity decreased significantly. The experimental erosion data was compared with the results obtained from computational fluid dynamics (CFD) simulations using the ANSYS Fluent software and good agreement was achieved showing slightly higher prediction in the simulation as a result of ideal assumptions. Three regression models were evaluated including Polynomial Regression (R 2 = 0.841), Random Forest (R 2 = 0.923), and Neural Network (R 2 = 0.956), where the Neural Network model demonstrated an accuracy that was 13.7% higher than the Polynomial Regression model.