About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
AI-Enabled Materials Processing: Integrating Accelerated Experimental Workflows and Processing-Aware Machine Learning
|
| Presentation Title |
An Autonomous Agentic System Enabled by Harness Engineering for Active Learning in a Self-Driving Laboratory for Materials Spallation |
| Author(s) |
Jing Luo, Fay Long, Chunyi Liu, Evan Sponseller, Christopher Stiles, Jaafar El-Awady |
| On-Site Speaker (Planned) |
Jing Luo |
| Abstract Scope |
The development of self-driving laboratories can significantly accelerate materials discovery, but their success depends on agentic systems capable of reliable decision-making based on both data and domain knowledge. Here, we present an autonomous agentic system enabled by harness engineering to perform machine learning and active learning in AIMD-L for materials spallation. Through a harness integrating domain-specific skills, the system coordinates specialized agents for dataset diagnosis, model selection and training, uncertainty quantification, candidate generation, and result interpretation. Using spall-strength prediction as the primary case study, we demonstrate that the system can automatically develop surrogate models, evaluate predictive uncertainty, and select candidates for subsequent active-learning cycles. We further benchmark the framework against manual modeling and prompt-only LLM guidance, and conduct ablation studies to quantify the contributions of individual harness components. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Modeling and Simulation, Computational Materials Science & Engineering |