Prediction method for crystal plasticity constitutive parameters in single-crystal superalloys via two-stage tiered mapping neural network based on nanoindentation

Crystal Plasticity Finite Element (CPFE) demonstrates significant advantages in characterizing anisotropic plastic deformation of crystalline materials, yet the efficient and accurate acquisition of its constitutive parameters remains a critical challenge. The nanoindentation technique, as a pivotal method for microscale mechanical measurements, provides an effective pathway for local mechanical property characterization. This study proposes a two-stage neural network prediction framework driven by single-crystal nanoindentation data. By generating nanoindentation and uniaxial tensile simulation datasets through CPFE, a tiered mapping relationship of ‘nanoindentation curve → stress-strain curve → crystal plasticity constitutive parameters’ is established using a multilayer perceptron (MLP).The results demonstrate that the constitutive parameters predicted by the two-stage neural network framework can accurately reconstruct nanoindentation curves, with the stress-strain curve prediction achieving a coefficient of determination R 2 > 0.97 and a mean absolute percentage error (MAPE) of merely 4.80 % for the plastic flow stress. Cross-model validation demonstrates that the mapping between nanoindentation load–displacement curves and stress–strain responses is universally applicable when identifying parameters of different crystal plasticity constitutive models. This overcomes the dependence of traditional direct inversion approaches on specific constitutive formulations. Using a nickel-based single-crystal superalloy as a case study, the constitutive parameters inverted from experimental nanoindentation data show excellent agreement with simulation results, confirming the engineering applicability of the proposed method. Once the neural network is trained, the framework reduces the time required for a single parameter prediction from several hours—typical of conventional optimization-based methods—to only a few seconds, providing an efficient solution for characterizing local mechanical properties of multiphase or heterogeneous materials and enabling high-throughput identification of constitutive parameters.