talk-icon
Description
Jennifer Carter1

1, Materials Science and Engineering, Case Western Reserve University, Cleveland, Ohio, United States

Data-driven materials design informed by legacy data-sets can enable the education of a new workforce, promote openness of the scientific process in the community, and advance our physical understanding of complex material systems. The performance of structural materials, which are controlled by competing factors of composition, grain size, particle size/distribution, residual strain, cannot be modeled with single-mechanism physics. The design of optimal processing route must account for the coupled nature of the creation of such factors, and requires students to learn machine learning and statistical modeling principles not taught in the conventional undergraduate or graduate level Materials Science and Engineering curricula. Therefore, modified curricula with opportunities for experiential learning are paramount for workforce development. Projects with real-world data provide an opportunity for students to establish fluency in the iterative steps needed to solve relevant scientific and engineering process design questions.

Exploratory data analysis (EDA) coupled with data-driven modeling allows new researchers to quickly orient in a field and gain insight into how and why decisions were historically made. The alloy development in 9-12wt% Cr martensitic steels has been of ongoing for thirty years. EDA quickly highlights the trade-off between short-term strength and long-term creep stability with increasing chromium concentration [1]. Though this knowledge can be gained from careful study of the literature, EDA allows the researcher to gain this institutional knowledge in short order without having to know what or where to look for first in the literature. This reduces the training cycle time and has implications to all R&D sectors facing knowledge transfer challenges as the “Boomer” generation retires.

Data-driven process|structure|performance (P|S|P) modeling provides insights into the oft-competing mechanisms that must be considered and optimized in the design of processing routes for new materials and design performance requirements. P|S|P modeling requires a foundation in statistical principles to assess the significance and quality of the findings. The constraints of different modeling techniques allow researchers to infer different physical qualities from the models [2].
Limitations of legacy data teach students the importance of statistical study protocol development. How to design the next study to mitigate issues of uncertainty and leverage prior knowledge to optimize inference gained from the modeling efforts? This goes hand-in-hand with the development of high-throughput experiments for microstructural characterization and mechanical behavior. For example, the study protocol to deconvolute the time-temperature effects on the kinetics of precipitate formation requires multiple processing steps to extract useful metrics. Next, robust data-science algorithms for analysis of microstructure [3], and mechanical performance [4] are needed to efficiently probe the design space. The approaches are useful for establishing protocols for the procurement of new databases of structure and performance for the process development of alloys from conventional and additive manufacturing.

1. Verma AK, Hawk JA, Bruckman LS, et al. (2019) Mapping Multivariate Influence of Alloying Elements on Creep Behavior for Design of New Martensitic Steels. Metallurgical and Materials Transactions A 50:3106–3120.
2. Verma AK, Huang W-H, Hawk JA, et al. Screening of Heritage Data for Improving Toughness of Creep-Resistant Martensitic Steels. In-prep
3. Smith TM, Senanayake NM, Sudbrack CK, et al. (2019) Characterization of nanoscale precipitates in superalloy 718 using high-resolution SEM imaging. Materials Characterization 148:178–187.
4. Senanayake NM, Yang Y, Verma AK, et al. (2019) An Automated Technique to Analyze Micro Indentation Load-Displacement Curve. Reno, NV

Tags