Project Description:

Addressing safety risk during the design of complex systems is difficult and time consuming. Prevailing practices centered around component failures rapidly run into scale problems, even as they inadequately address partial failures, common-cause failures, software, and humans. During operation, updating risk assessments to ensure that risk is managed is also difficult and time consuming, because the underlying risk analyses themselves are also large and complex. So, risk assessments quickly become “stale”.

Building on RETH Institute work, we propose to create a “digital risk twin” of a physical system. The twin is a network representation of the system’s possible states and transitions, which may be undesired disruptions, intended changes between operating modes, or controls activated to return the system from unsafe to nominal states. The twin is updated using inputs from sensor and fault detection and diagnosis (FDD) systems integrated into the physical system. We use the FDD information to generate an overall network risk status, to identify specific issues leading to the risk status (e.g., a particular state transition chain to an accident state is currently likely), and to inform safety-critical human decision-making. We will develop methodologies to automatically grow the state network representation when encountering unanticipated faults.

Start Date:

January 2023

Postdoc Qualifications:

PhD in computer science/statistics/machine learning. Engineering PhD with emphasis on machine learning.
Ideally well-versed in causal inference and deep learning

Co-Advisors:

Ilias Bilionis, ibilion@purdue.edu, ME, https://www.predictivesciencelab.org
Karen Marais, kmarais@purdue.edu, AAE, https://engineering.purdue.edu/VRSS

Bibliography:

https://www.purdue.edu/rethi/

Richens, J.G., Lee, C.M. & Johri, S. Improving the accuracy of medical diagnosis with causal machine learning. Nat Commun 11, 3923 (2020). https://doi.org/10.1038/s41467-020-17419-7

Prasanna Tamilselvan, Pingfeng Wang, Failure diagnosis using deep belief learning based health state classification, Reliability Engineering & System Safety, Volume 115,
2013, Pages 124-135, ISSN 0951-8320, https://doi.org/10.1016/j.ress.2013.02.022.