Design space visualization for guiding investments in biodegradable and sustainably sourced materials
- PDF / 543,903 Bytes
- 7 Pages / 612 x 792 pts (letter) Page_size
- 8 Downloads / 155 Views
Research Letter
Design space visualization for guiding investments in biodegradable and sustainably sourced materials James S. Peerless, Emre Sevgen, Stephen D. Edkins, Jason Koeller, Edward Kim, and Yoolhee Kim, Citrine Informatics, Redwood City, CA 94063, USA Astha Garg†, A*STAR, Singapore Erin Antono and Julia Ling, Citrine Informatics, Redwood City, CA 94063, USA Address all correspondence to Julia Ling at [email protected] (Received 24 November 2019; accepted 2 January 2020)
Abstract In many materials development projects, scientists and research heads make decisions to guide the project direction. For example, scientists may decide which processing steps to use, what elements to include in their material selection, or from what suppliers to source their materials. Research heads may decide whether to invest development effort in reducing the environmental impact or production cost of a material. When making these decisions, it would be helpful to know how those decisions affect the achievable performance of the materials under consideration. Often, these decisions are complicated by trade-offs in performance between competing properties. This paper presents an approach for visualizing and evaluating design spaces, where a design space is defined as the set of possible materials under consideration given specified constraints. This design space visualization approach is applied to two case studies with environmental impact motivations: one in biodegradability for solvents, and the other in sustainable materials sourcing for Li-ion batteries. The results demonstrate how this visualization approach can enable data-driven, quantitative decisions for project direction.
Introduction Data-driven methods for materials development have become increasingly prevalent over the past decade.[1–5] One widespread machine learning approach for materials development is screening.[2,6–8] In materials screening, a machine learning model is trained to predict materials properties given the chemical formula and processing information and then is applied to a set of candidate materials to predict their properties. The materials predicted to have the best performance are then selected for experimental testing. Meredig et al.[2] applied this screening approach to sift through millions of potential ternary compounds to surface thermodynamically stable combinations. Ward et al. [9] showed how a similar approach could be used to find bulk metallic glasses. A related data-driven approach is sequential learning, also known as active learning.[10,11] This workflow involves pairing the machine learning model with a sampling or optimization routine to select new experiments to perform, then iteratively retraining the model using the new data so that it can provide successively more informed suggestions. Ling et al.[12,13] illustrated how this approach could be applied to a variety of application cases, including the development of hightemperature superconductors, resilient superalloys, and novel † This author was employed by Citrine Informatics
Data Loading...
