TECHNOLOGY AREA(S): Materials/Processes

ACQUISITION PROGRAM: Basic Research Challenge (BRC) Program

OBJECTIVE: Develop (1) algorithms to transform various raw data formats into information-rich features for machine learning (ML), and (2) software and modeling tools for ML that will automatically detect patterns in data; and learn and improve from experience the ability to predict new materials/optimal materials processes.

DESCRIPTION: Machine learning is a powerful subset of artificial intelligence for systems to learn from data, pattern identification, and decision making. Application of machine learning tools can enable accelerated new materials discovery by leveraging existing data for potential alloy composition and processing. A key challenge in applying ML algorithms to materials science data is that data comes in many formats. Determining how to featurize and utilize different materials data formats so that prior data can be used as training data for ML algorithms can be difficult. Feature engineering, including extraction, transformation, and selection, is critical for improved ML accuracy.

High entropy alloys (HEAs) or multi-principal element alloys (MPEAs) utilize a broad composition space through control of material entropy. This growing material set can allow for enhanced material properties compared to conventional materials. Basic research of HEAs requires development of a powerful computational infrastructure, which, when coupled with multiple database formats and ML approaches, can accelerate discovery of new viable HEAs. Additionally, additive manufacturing (AM) materials and processes would also be able to feature engineering tool sets to accelerate process development for new AM materials. Applying these tools to new materials and processes can be impactful to accelerate materials development.

To fully realize data analytics and machine learning tools for materials development, it is necessary to focus on developing algorithms to transform various raw data inputs into information-rich features suitable to model. The concept/approach/framework created by developing algorithms to classify and unify disparate data sets into a consistent format can be utilized and augmented by subsequent researchers. Feature transformation and selection would enable accuracy improvements to ML models applied to HEAs and AM.

PHASE I: Define and develop a concept/approach/framework for feature engineering tools to extract critical information from multiple formats. Key features may include material properties, chemistry, and processing variables. Include, in the concept/approach/framework, development of an alloy-related materials database with appropriate identification classifiers and interactions. Develop a Phase II plan. In a Phase I option, if exercised, the STTR team will demonstrate the feasibility of the proposed concept/approach to provide labeled data output for HEAs/AM.

PHASE II: Develop, demonstrate, and validate a materials database for supervised (e.g., support vector, neural networks) and unsupervised learning algorithms (e.g., cluster analysis) use for HEA/AM. Ensure that the database is able to identify prioritization of features whether it be structural, chemical, and physical properties or AM-related processing-microstructure-property phenomena.

PHASE III DUAL USE APPLICATIONS: Transition optimized computational/informatics handling engineering tools for commercialization in ML utilization through original equipment manufacturers (OEMs) or other partnering agreements. Commercialization of this technology may be through new material discovery or rapid process development. The STTR team will demonstrate the technology to DoD warfare centers/production facilities. Dual use applications could include aircraft, land vehicles, materials processing entities.

REFERENCES:

1. Miracle, D. B. and Senkov, O. N. �A critical review of high entropy alloys and related concepts.� Acta Materialia. Vol. 122, 1 January 2017, p. 448-511.� https://www.sciencedirect.com/science/article/pii/S1359645416306759

2. Witten, Ian and Frank, Eibe. �Data Mining: Practical Machine Learning Tools and Techniques.� ftp://ftp.ingv.it/pub/manuela.sbarra/Data%20Mining%20Practical%20Machine%20Learning%20Tools%20and%20Techniques%20-%20WEKA.pdf

3. Ling, Julia, et al. �Machine Learning for Alloy Composition and Process Optimization�. (Proceedings of ASME Turbo Expo 2018 Turbomachinery Technical Conference and Exposition.) https://arxiv.org/abs/1704.07423

KEYWORDS: Machine Learning; Feature Engineering; Additive Manufacturing; High Entropy Alloys; Data Analytics; Feature Extraction