NRC RFI: Risk-Informed Approach to Evaluate Engineering Applications of Machine Learning

Notice ID: ADM-25-0005

The NRC requires technical assistance support for the development of a TLR providing insights and guidance on the development of a risk-informed framework for the regulatory acceptance of data-driven and ML-enabled applications for condition and performance monitoring of mechanical systems and components. This effort will examine ML-enabled applications performing regression-like analyses to focus the scope, with applications of interest including: 1) anomaly detection, 2) condition monitoring, and 3) predictive health monitoring. The TLR will explore the influence of risk on the acceptance of ML-enabled applications. Additionally, the TLR will provide insights and guidance on ML specific safety considerations, such as explainability, robustness, model performance monitoring; and how those considerations may influence the risk and approval of such a system. The TLR will also examine performance metrics which could be used to provide insights on the behavior, and guardrails to bound the performance, of an underlying data-driven or ML model to provide reasonable assurance of the adequate performance of an ML-enabled application.

The anticipated scope of the research includes the following:

• Influence of model section on the risk of ML-enabled applications
• Data-driven or ML specific safety considerations and how those consideration influence risk
• Performance metrics characterizing data-driven or ML-enabled applications
• Performance guardrails bounding performance and reduce risk of data-driven and ML-enabled applications
• Impact of deployment considerations on the risk of data-driven and ML-enabled applications
• Development of data-driven or ML-enabled performance or condition monitoring case-studies to exercise a risk-informed acceptance framework

Read more here.

🔊 Listen to Story