Machine Learning and Data Analytics for Design and Manufacturing of High-Entropy Materials Exhibiting Mechanical or Fatigue Properties of Interest (2012.07583v1)

Abstract: This chapter presents an innovative framework for the application of machine learning and data analytics for the identification of alloys or composites exhibiting certain desired properties of interest. The main focus is on alloys and composites with large composition spaces for structural materials. Such alloys or composites are referred to as high-entropy materials (HEMs) and are here presented primarily in context of structural applications. For each output property of interest, the corresponding driving (input) factors are identified. These input factors may include the material composition, heat treatment, manufacturing process, microstructure, temperature, strain rate, environment, or testing mode. The framework assumes the selection of an optimization technique suitable for the application at hand and the data available. Physics-based models are presented, such as for predicting the ultimate tensile strength (UTS) or fatigue resistance. We devise models capable of accounting for physics-based dependencies. We factor such dependencies into the models as a priori information. In case that an artificial neural network (ANN) is deemed suitable for the applications at hand, it is suggested to employ custom kernel functions consistent with the underlying physics, for the purpose of attaining tighter coupling, better prediction, and for extracting the most out of the - usually limited - input data available.

Overview of Machine Learning and Data Analytics for High-Entropy Materials Design and Manufacturing

This paper introduces a comprehensive framework for utilizing ML and data analytics in the design and manufacturing process of high-entropy materials (HEMs), particularly focusing on alloys and composites with vast compositional spaces. These HEMs are considered for structural applications due to their exceptional mechanical properties, such as ultimate tensile strength (UTS) and fatigue resistance.

Framework and Models

The authors propose a method that leverages ML for the rapid identification of alloy compositions exhibiting desirable mechanical and fatigue properties. The ML models are designed to incorporate physics-based constraints, thereby improving prediction accuracy and making the most out of typically limited datasets. The framework prioritizes the selection of suitable optimization techniques that fit the given application and available data.

The paper emphasizes the importance of forward and backward prediction models. Forward prediction models involve estimating material properties from given compositions and processing parameters, whereas backward prediction models focus on identifying potential compositions that meet target property specifications.

Physics-Based Incorporation

A critical aspect of this framework is the integration of physics-based models. These models account for dependencies such as those informed by thermodynamics (e.g., CALPHAD), dislocation dynamics, and empirical rules. By embedding a priori physical constraints into ML models, the framework aims to enhance the robustness of predictions. For instance, custom kernel functions consistent with underlying physics are recommended for artificial neural networks when appropriate.

Statistical and Computational Models

The paper contrasts traditional statistical models with physics-based approaches. For example, the paper presents an augmented Statistical Fatigue Life model to depict S/N curves for AM components. This model can accommodate a wide variety of input parameters affecting component fatigue life, offering potential insights into the dominant determinants of fatigue.

In terms of numerical results, the model's capacity was demonstrated through insights gained into defect levels and process impacts on fatigue resistance, showing a significant correlation with ultimate tensile strength. Predictions yielded consistency with empirical stability rules, showing promise for wider application in identifying strong compositions within the expansive composition space.

Machine Learning in Additive Manufacturing (AM)

A particular focus is on using ML to optimize parameters and quality control for additive manufacturing (AM) processes, especially powder-bed fusion AM. The integration of ML can address the inherent variability and complex parameter interdependencies in AM processes, potentially improving the quality and consistency of manufactured components.

Implications and Future Work

The implications of this work suggest that ML can effectively navigate the enormous compositional design space of HEMs, providing a pathway toward designing materials with tailored properties for advanced applications. The paper suggests potential improvements in manufacturing processes and the optimization of mechanical properties through intelligent data analytics paired with physics-informed constraints.

Future work will likely involve more sophisticated models that further integrate process conditions and defect characteristics into predictions of tensile strength and endurance limits. Additionally, advanced joint optimization techniques considering multiple objectives such as strength, ductility, and oxidation resistance, remain an intriguing area for further exploration.

In conclusion, this paper underscores the transformative potential of machine learning in materials science and engineering, particularly for complex systems like high-entropy materials. By bridging data analytics and the physical principles governing material behavior, researchers can accelerate the discovery and optimization of next-generation materials with multifunctional capabilities.