Multi-objective optimization of cylinder liner honing surface parameters for lubrication and wear properties based on BPNN–NSGA-II

Simultaneously enhancing oil film thickness while controlling wear load remains a key challenge in cylinder liner honing texture optimization, yet these performance indicators are typically optimized separately. This study proposes a simulation-driven multi-objective framework that extends conventional finite-level discrete screening to continuous-domain parameter optimization. A piston ring pack-cylinder liner model is established with crosshatch angle, groove depth, groove density, and groove width as design variables. A preliminary discrete solution is identified from 49 orthogonal experimental schemes using entropy-weighted composite scoring. Subsequently, 650 Latin hypercube samples are employed to train a backpropagation neural network (BPNN) surrogate, which achieves test-set coefficients of determination of 0.999 for oil film thickness and 0.9332 for wear load, confirming high predictive accuracy. The validated BPNN is then coupled with the non-dominated sorting genetic algorithm II (NSGA-II) for continuous multi-objective optimization, and the final Pareto compromise is selected via the entropy-weighted technique for order preference by similarity to an ideal solution (TOPSIS). The trained weights, biases, and activation functions are explicitly formulated into closed-form response functions, enabling rapid and reproducible performance evaluation without repeated simulations. The optimal parameter combination, crosshatch angle 30.018°, groove depth 3.485 μm, groove density3.993 mm −1 , and groove width 39.875 μm, yields a 7.23% increase in oil film thickness relative to the discrete optimum, accompanied by only a 0.02% rise in wear load. The proximity of the two optimized solutions in the parameter space validates the proposed BPNN-NSGA-II method as a practical and reliable approach for honing texture design.