A physics-informed neural network-based numerical inverse method for optimization of diffusion coefficients in NiCoFeCr multi principal element alloy

The composition-dependent pseudo-binary (PB) interdiffusion coefficients and the main intrinsic diffusion coefficients of all the components at the near equiatomic composition of NiCoFeCr system are estimated following the PB diffusion couple method. These are otherwise impossible to estimate directly following the conventional method. Subsequently, a physics-informed machine learning based numerical inverse method is used to optimize the diffusion parameters in two steps. Initially, optimization is done by developing a good match with the diffusion profiles and estimated interdiffusion coefficients over whole composition range of the diffusion couples. However, a mismatch was found in the extracted intrinsic diffusion coefficients. Therefore, the second level of optimization is done with estimated intrinsic diffusion coefficients at the Kirkendall plane as constraints demonstrating the need for these diffusion parameters for generating a reliable mobility database. The direct estimation and optimization of diffusion coefficients without using thermodynamic details is an added advantage, especially in multicomponent alloy systems.

Automated summary

This study estimates the composition-dependent pseudo-binary interdiffusion coefficients and the main intrinsic diffusion coefficients of the NiCoFeCr system using the PB diffusion couple method. The diffusion parameters are optimized using a physics-informed machine learning inverse method. The results demonstrate the importance of estimating intrinsic diffusion coefficients and their role in generating a reliable mobility database.