# Notes on [[thesis-ideas-2025-07-30]]

What needs done:

- [X] 1 needs edited and reviewed
    - [X] Review outcomes. I really don't like outcome
    number 1.

- [X] Review and edit 2

- [X] Review and edit 3
    - [X] Write an impact section

- [X] Review and edit 4
    - [X] Needs more goal 

- [X] Review and edit 5

- [X] 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.