The focus of this Faculty Early Career Development (CAREER) project is on predicting malfunction in a physical system and maintaining functionality in spite of severe faults, a common challenge in many technical fields, and of great ethical and economic importance. While traditional methods for fault-tolerance rely on passive measures such as redundancy, safety margins, and preventative maintenance, this project will develop a new active approach. This approach continuously reconstructs models of the current system from sensory data, and uses these models to assess failure and uncertainty, and to generate compensatory actions using design automation in-situ. This approach will be applied and evaluated on both quasi-static systems and dynamical systems across various scales and complexities.
If successful, the broader impacts of this research will lead to improvements in the ability of systems to autonomously identify and warn about potential failure, and even sustain operation and recover from some forms of damage. As systems are becoming increasingly complex, there is a need to transition from passive fault tolerance to active fault tolerance, where systems that can diagnose and repair faults autonomously. The method proposed here is domain-independent, and as such may be applied across many complex systems beyond those explicitly addressed in the proposed work, such as modeling of biological systems and medical diagnosis.
Educational activities will focus both on expansion of undergraduate curriculum in design for fault tolerance, and on a broader outreach program whose goal is to enhance hands-on engineering and science education for children and students through a library of 3D-printable educational models across several scientific fields
