The materials development is often slow and guided by trial and error. The typical process development includes synthesizing a new material, characterizing its structure, and then examining its physical properties. The process is repeated iteratively by changing processing parameters and measuring how parameter changes influence the properties of interest. This research will bring to bear new advances in experimental hardware, artificial intelligence (AI) and machine learning (ML) for materials processing in order to increase the rate of materials development while decreasing cost and removing human biases. By employing an autonomously-operating system, that employs AI and ML tools, carbon nanotube (CNT) forest processing will be accelerated. CNT forests could be an important source material for reinforcement materials in many key technologies, such as plastic-based composites and automobile tires. With limited or no human intervention, the autonomous system will launch experiments, observe and characterize the results in real time, and then learn from the results to refine the process. The tools and procedures developed will be used to enhance the national manufacturing base and global competitiveness while enabling rapid development of new materials and applications to advance public prosperity.
The autonomous system incorporates a) in-situ scanning electron microscope (SEM) synthesis of CNT forests, b) computer vision to quantify the governing mechanisms of CNT assembly, c) a complementary finite element simulation, d) deep learning networks to predict CNT forest properties, and e) a knowledge-based driven distributed control algorithm for autonomous decision making. The system will first operate in manual mode to establish baseline protocols. Process automation will then be demonstrated by allowing the system to execute user-defined tasks. Autonomy will occur when the system plans experiments based on previous results in a knowledge-base and then executes experiments to observe the outcomes. As proof of concept, the system will demonstrate the ability to deterministically synthesize CNT forests having an effective modulus between 100 kPa and 1 GPa without human input. The open-source software, data and the simulations generated in this work will be provided freely to researchers and educators to benefit efforts for effective design and control of materials discovery processes. Broadening participation in computation among minorities and women will be pursued through engaging undergraduates into the research environment and project activities.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.