PINN-based Oka reinforced modelling and prediction of erosion wear in Al-Mg-MoS₂ powder metallurgy composites

The solid-particle erosion is a crucial degrading mechanism in engineering components, although dependable predictions are challenging, especially when experimental data is limited. This analysis explores the erosion characteristics of Al–Mg–MoS₂ powder-metallurgy composites and proposes a multi-fidelity Physics-Informed Neural Network (MF-PINN) using a Modified Oka erosion formula. Erosion evaluations were conducted using ASTM G76 standards with L27 design across three impingement angles, velocity, and composite composition, with four repetitions for every condition. The analysis of variance revealed that the foremost element was the erodent velocity (53.29%), followed by the composite composition (14.84%) and the impingement angle (7.00%). Each of the three principal effects demonstrated statistical significance (p < 0.05). The MF-PINN combines the Oka-based physical trend with experimental measurements and a neural correction for material-specific deviations. Training, validation, and testing were grouped to prevent leakage between repeated measurements. The model demonstrated R² values of 0.966, 0.977, and 0.954 for training, validation, and testing, respectively, with testing MAE and RMSE recorded at 2.05 × 10⁻⁶ g/min and 3.44 × 10⁻⁶ g/min, respectively. The interpretation of the model reproduced the empirically identified factor hierarchy and indicated that a physically consistent and data-efficient framework for erosion forecasting in the investigated domain.