Progress toward autonomous experimental systems for alloy development
- PDF / 2,630,241 Bytes
- 8 Pages / 585 x 783 pts Page_size
- 3 Downloads / 152 Views
Introduction Alloy development has traditionally been a lengthy, painstaking process, and insertion in safety-critical applications can take a decade or more from initial development to widespread adoption. This traditional path for alloy development typically involves incremental modifications to chemistry and process conditions based on a combination of expert knowledge, model-based guidance, and Edisonian experimentation. The resulting alloys are often marked improvements over previous solutions, yet they are likely not the optimal alloy design for a given application. For insight into the possibility of improved optimization, one can turn to mechanical engineering, where increased usage of topology optimization theory is changing the design paradigm from human experts using computer-aided design (CAD) tools to, instead, the human defining the requirements and boundary conditions, and the computer algorithms automatically iterating between CAD and finite element analysis to identify optimal solutions. The results are often complex organically shaped topologies, far more efficient and intricate than traditional designs.1 Translating these concepts to alloy development, one can envision a future where engineers define the requirements of the alloy (i.e., the functionality of the alloy in the operational environment), and artificial intelligence (AI) systems would identify optimal combinations of composition and processing
that produce microstructures with properties far beyond conventional alloys. Ultimately, such an expert system would not just serve as the metallurgist, but also the mechanical engineer—the material, processing, and topology would be holistically optimized for both performance and cost within the limits of the manufacturing capabilities.2 With mastery of the governing equations, all of this optimization can take place in silico without the need for experimental iteration. However, because process–structure–property relationships are so multidimensional and complex, in the foreseeable future, we must rely on iterative experimentation to inform material models on the pathway toward optimization. Autonomy requires both the ability to act independently (automation) and the ability to react based on evolving knowledge (intelligence)—an autonomous system is an “expert” that can incorporate new information to revise a governing hypothesis and guide improvements.3 Recently, researchers from the Air Force Research Laboratory demonstrated an Autonomous Research System, which was able to optimize the growth rate of single-walled carbon nanotubes through the combination of an automated growth reactor, rapid in situ characterization, and an AI system.4 The system utilized an AI planner to iteratively propose new experiments to optimize the process and converge on a maximum growth rate without a priori knowledge of the physical processes that control carbon nanotube growth.
Brad L. Boyce, Materials, Physical, and Chemical Sciences Center, Sandia National Laboratories, USA; [email protected] Michael D. Uchic,
Data Loading...
