3.7 KiB
Notes on thesis-ideas-2025-07-30
What needs done:
-
1 needs edited and reviewed
- Review outcomes. I really don't like outcome number 1.
-
Review and edit 2
-
Review and edit 3
- Write an impact section
-
Review and edit 4
- Needs more goal
-
Review and edit 5
-
Review and edit 6
Discussion Cheat Sheet
Chat helped with this
Temporal Logic Specifications for Autonomous Controller
Synthesis
- Feasibility: ★★★★★
- Impact: ★★★★☆
- Merit: ★★★★★
Scope Boundaries: Focus on one subsystem (e.g., rod supervisory control), one specification language, and existing synthesis tools (TLA+, FRET, Strix).
Key Risk: State space explosion during synthesis could make controller generation intractable.
Mitigation Strategy: Use bounded abstractions, compositional synthesis, and validate the synthesized controller on a high-fidelity simulation before scaling up.
Formally Verified Runtime Monitoring and Fallback
- Feasibility: ★★★★★
- Impact: ★★★★☆
- Merit: ★★★★☆
Scope Boundaries: Single primary controller with one fallback controller, one LTL specification set, and integration with ARCADE.
Key Risk: Limited novelty if scoped too narrowly or perceived as a straightforward engineering integration.
Mitigation Strategy: Emphasize automation of specification-to-monitor translation, nuclear-specific verification, and proof artifact generation to show novelty.
Verified Adaptive Control
- Feasibility: ★★★★☆
- Impact: ★★★★☆
- Merit: ★★★★☆
Scope Boundaries: One subsystem (rod control), one adaptation method, runtime contract monitoring only.
Key Risk: Over-scoping to multiple adaptation targets or attempting plant-wide adaptive control.
Mitigation Strategy: Pick representative degradation types (e.g., HX fouling, pump efficiency drop); limit adaptation to parameter tuning inside pre-verified safe envelopes.
Integrating Shielding into Nuclear Power Control
- Feasibility: ★★★★☆
- Impact: ★★★★☆
- Merit: ★★★★☆
Scope Boundaries: One ML control task (e.g., startup or load-following), one shield synthesis approach from temporal logic.
Key Risk: Regulatory and industry reluctance toward ML in safety-critical nuclear applications.
Mitigation Strategy: Demonstrate shielding benefits for both ML and conventional controllers to broaden acceptance.
Improved: Data-Driven Fault Detection Using
High-Assurance Digital Twins
- Feasibility: ★★★★☆
- Impact: ★★★★☆
- Merit: ★★★★☆
Scope Boundaries: Limit to 3–4 high-impact fault types (e.g., secondary coolant loss, HX fouling, sensor drift), residual-based detection with physics-informed models.
Key Risk: Scope creep into too many fault scenarios or overly complex ML methods.
Mitigation Strategy: Focus on explainable, physics-informed detection; tie mitigation responses directly to NRC-aligned safety procedures.
Formally Verified Neural Network Control of Control Rod
System
- Feasibility: ★★★☆☆
- Impact: ★★★★☆
- Merit: ★★★☆☆
Scope Boundaries: Small, well-structured NN architecture; bounded state space; one primary safety property (shutdown margin).
Key Risk: Scalability issues in SMT/MILP verification for larger or more complex networks.
Mitigation Strategy: Constrain network size and complexity; limit verification domain to tractable operating regions; focus on proof-of-concept that shows nuclear-specific applicability.