A physics-constrained GA-BP neural network for contact fatigue life prediction of surface-strengthened transmission gears

Helicopter transmission gears operate under harsh high-speed, heavy-load, high-frequency alternating stress conditions, where contact fatigue failure accounts for most gear failures, critically limiting the reliability and service life of transmission systems. The shot peening (SP)-finishing composite process is a recognized high-efficiency synergistic surface strengthening technology for gear anti-fatigue manufacturing, yet the coupling mechanism between its multi-process parameters and surface integrity gradient characteristics has not been systematically clarified. Traditional fatigue life prediction methods fail to fully account for the gradient evolution of surface integrity induced by composite strengthening, causing significant prediction deviations for surface-treated gears. Meanwhile, pure data-driven intelligent models rely heavily on high-quality experimental data, are prone to local optima, and lack sufficient physical interpretability, severely restricting their engineering application in aviation gear life assessment. To address these gaps, this work develops an integrated framework combining multi-process experimental characterization, multi-scale coupled simulation integrating Johnson-Cook constitutive and ETMB dislocation density models, and a physics-constrained GA-BP neural network for contact fatigue life prediction of surface-treated 9310 steel gears. We quantitatively reveal the synergistic evolution of key surface integrity parameters under the SP-finishing process. The multi-scale model has high accuracy, with relative errors below 8% for residual stress and below 12% for surface roughness. The proposed GA-BP model yields life prediction error within scatter band of 1.7, outperforming traditional BP networks. This study provides a framework for aviation gear process optimization, surface integrity regulation and high-precision life assessment, with important engineering value for anti-fatigue design of high-reliability gear components.