Design of a nickel-base superalloy using a neural network (1803.03039v1)

Abstract: A new computational tool has been developed to model, discover, and optimize new alloys that simultaneously satisfy up to eleven physical criteria. An artificial neural network is trained from pre-existing materials data that enables the prediction of individual material properties both as a function of composition and heat treatment routine, which allows it to optimize the material properties to search for the material with properties most likely to exceed a target criteria. We design a new polycrystalline nickel-base superalloy with the optimal combination of cost, density, gamma' phase content and solvus, phase stability, fatigue life, yield stress, ultimate tensile strength, stress rupture, oxidation resistance, and tensile elongation. Experimental data demonstrates that the proposed alloy fulfills the computational predictions, possessing multiple physical properties, particularly oxidation resistance and yield stress, that exceed existing commercially available alloys.

Design of a Nickel-Base Superalloy Using a Neural Network

This paper presents a novel approach to designing nickel-base superalloys through the utilization of a computational tool powered by an artificial neural network. The objective is to develop materials that simultaneously fulfill multiple physical criteria, enhancing the capability of materials design beyond traditional methods. Traditional materials development has largely relied on empirical experiments, which may not achieve the requisite balance of properties necessary for specific engineering applications. The authors propose a method that expedites the design process by computationally predicting, discovering, and optimizing alloy compositions based on a multi-criteria specification.

Methodology Overview

The methodology hinges on integrating pre-existing materials data to train an artificial neural network that predicts individual material properties as a function of composition and heat treatment. This predictive model guides the discovery of alloy compositions that are statistically likely to satisfy specific target criteria. The neural network assists in optimizing properties such as cost, density, $\gamma'$ phase content and solvus, phase stability, fatigue life, yield stress, ultimate tensile strength, stress rupture, oxidation resistance, and tensile elongation.

A significant innovation is the implementation of uncertainty in the predictions, which is quantified using a Bayesian bootstrap approach. This allows a probabilistic analysis of whether a proposed composition will meet the desired criteria, thus enabling the selection of candidates with the highest likelihood of success. Unlike conventional methods such as principal component analysis and robust design, which do not account for uncertainty, this probabilistic framework can evaluate numerous potential compositions, increasing the likelihood of optimizing the alloy's properties.

In terms of optimization, the method employs a simulated annealing approach to navigate the high-dimensional space of design variables efficiently. This process is complemented by computing probabilities of satisfying various criteria, which guides the search for optimal solutions. The approach enables an exploration of $10^8$ potential compositions in approximately one hour, demonstrating the efficiency and practicality of the proposed computational tool.

Results

The paper details the development of a nickel-base superalloy that outperforms existing commercial alternatives in terms of yield stress and oxidation resistance, among other properties. Specifically, experimental results corroborate the computational predictions, with the proposed alloy exhibiting superior oxidation resistance and yield stress compared to alloys like RR1000 and Udimet720. Figure results within the paper illustrate that, across various tested properties, such as density and $\gamma'$ solvus temperature, the new alloy maintains a competitive edge.

Practical and Theoretical Implications

The primary implication of this research is its potential to revolutionize the process of materials design, enabling faster and more efficient discovery of alloys tailored to specific engineering needs. Commercially, this could shorten the development cycle of aerospace and industrial components, leading to more efficient and environmentally friendly products. Theoretically, this paradigm shift in materials design framework enriches the field of computational materials science, paving the way for future studies that might explore the integration of more complex networks or alternative optimization strategies.

The research also sets the stage for applying this methodology beyond nickel-base superalloys to other material systems where multi-criteria optimization is critical. Future advancements could focus on increasing the scalability of this approach, integrating real-time data input for on-the-fly adjustments to predictions, and exploring more complex neural network architectures that may yield even higher fidelity in predictions.

In conclusion, the paper contributes to both the practical application and theoretical foundation of materials science and engineering by introducing a systematic, computationally-driven method aimed at identifying optimal material combinations with high predictive accuracy. Its success in developing a commercially promising nickel-base superalloy suggests significant promise for similar applications across various material domains.