High-Throughput Materials Discovery for Extreme Conditions (HTMDEC)
Within the Army science and technology enterprise, DEVCOM ARL is chartered to conduct disruptive foundational research, engage as the Army’s primary collaborative link to the scientific community, and interface to shape future fighting concepts. We crystalize these ideas and the impetus to perform these functions at the pace of innovation as ‘Operationalize Science for Transformational Overmatch’.
Simply put, we seek to accelerate discovery and transition breakthroughs to the Warfighter.
Rule-based artificial intelligence (AI) and machine learning (ML) tools present powerful avenues for exploring an information landscape in discovering novel materials for applications in extreme conditions (e.g. high-strain rate, high-g loading, high temperature). Such approaches present considerable opportunity in exploring new frontiers for materials used in protection and lethality applications, especially when coupled with new approaches that allow larger and richer datasets, computational tools, and data infrastructure for collaboration. Broadly, AI/ML can be used to augment individual steps in the synthesis-processing-characterization pipeline, be used for scale-bridging to draw greater information from more tractable experimental approaches, and be used to guide a broader research loop.
Advances in synthesis, modeling, and characterization will greatly advance our ability to exploit monolithic materials in extreme conditions. However, there is a need to contemplate how the capabilities of additive manufacturing and other processing techniques can be used to evaluate materials that exhibit spatial variations in composition, anisotropic characteristics, and contain interfaces between multiple materials. The parameter space expands exponentially as these variables compound the system inputs, but truly advanced materials performance will likely be dependent on an integrated systems-level approach to materials design.
ML toolsets coupled with advanced manufacturing and characterization is necessary to achieve accelerated discovery of new materials for application in extreme dynamic (impact, thermal, ablative) conditions. ML toolsets and software exist but may need to be adapted for the specific requirements of materials discovery and design. Full exploitation of the ML approach will certainly require extension and further development to focus on proof-of-concept for material classes of interest in Army applications. This could be achieved within a generalized and scalable framework that supports rapid, robust and trusted data exchange. New tools that consolidate/organize data and increase throughput throughout the workflow will require a specialized approach to be applied to ephemeral phenomena e.g. shocks, heating, localized deformation, and failure. ML models that incorporate these phenomena will critically rely on physics-based models that adequately capture the underlying driving mechanisms. Critical (targeted by ML approaches) physics models may require further development; ML offers opportunity to consolidate much of these physics into fast-running analytic frameworks compatible with the high-throughput approach and may be used to guide autonomous systems for high-throughput characterization of transient phenomena.
To accelerate improvements in Army armor and weapon system performance, DEVCOM ARL wants to leverage high-throughput methods in synthesis, processing, characterization, and modeling for materials used in these applications. Machine-learning techniques are in the nascent stage of integration with materials science but may present a path towards accelerated discovery, as these tools may uncover novel links between system performance and material science that have been previously underdeveloped or overlooked. DEVCOM ARL seeks collaboration with external investigators to leverage (and train experts on) machine-learning techniques in the discovery of materials that perform in extreme environments, but machine-learning techniques require large volumes of quantifiable data in order to best reveal links between the materials science and system performance. High throughput characterization and manufacturing techniques may present a viable approach to satisfy the data volume requirements to bring machine-learning to bear.
In summary, the US Army Modernization Priorities require materials that survive and perform in extreme environments; harsh military environments of high-acceleration (e.g. projectile launch and flight), high-temperature and rapid ablation (e.g. hypersonic flight), and impacts at very high velocity (terminal ballistics). The totality of these environments and accumulating requirements on future materials drives the imperative to consider an increasingly large number of constituent elements, structure and properties. Discovery must now parse through billions of candidate materials to achieve highly specialized and transformational functions. This drives a data-driven approach; one that fuses high-throughput materials synthesis and characterization with machine learning algorithms and close-loop discovery automation.
The overarching goal of this program is to couple automation and machine learning techniques to material manufacturing and characterization to withstand and perform under extreme conditions. The program will develop the necessary methodologies, models, algorithms, synthesis & processing techniques, and requisite characterization and testing to rapidly accelerate the discovery of novel materials through data-driven approaches. As such, it is expected the results of this program will be the above techniques as well as novel materials exhibiting unprecedented properties at the appropriate scales that have been developed utilizing all of the aforementioned tools which will be provided to DEVCOM ARL for further analysis and testing.
Proposals may draw from any number of these thrusts but must focus on research that employs high-throughput data-driven techniques to close material design loops connecting material selection, synthesis, and processing to system performance.
In order to achieve this paradigm shift in materials discovery, significant advances are needed in the following general thrust areas:
- Data-driven Material Design – meant to be a comprehensive term for all aspects of the material design phase of the material development cycle which are accelerated through the integration of data-driven methods.
- High-Throughput Synthesis & Processing – to include both modifying existing synthesis & processing methods to accommodate for high-throughput, as well as developing novel techniques.
- High-Throughput Characterization – to include implementation of automation for conventional techniques, and the development of surrogate tests to mimic techniques which are not amenable to automation, especially for experiments in extreme conditions (e.g. high strain rate, high temperature).
- ML-augmented Physics-Based Models – the use of ML tools to identify the most crucial parameters and parametrization experiments for physics-based models is poised to be a tipping point in materials science. To date, nearly all ML algorithms have been developed for big data (e.g. image recognition). It is critical that we discontinue ‘repurposing’ these types of algorithms and begin developing ML algorithms specifically designed for materials discovery, and informed by physics.