Obsidian/Presentations/20251215-Emerson-Pres/actual-presentation-outline.md

Presentation Outline: ERLM Presentation

Audience:

• Engineering PhD students
• Dr. Cole,
• Potentially other faculty members.

Presentation Style: Proposal, Assertion Evidence

SLIDE 1: HOOK

Message

THE UNITED STATES STANDS ON THE PRECIPICE OF A SEVERE ENERGY CRISIS

1. We're looking down the barrel of a severe energy shortage with the introduction of data centers for AI buildout
2. The cheapest way to build new power right now is natural gas combined cycle power plants
3. We also have a climate crisis, which nat gas definitely will not help with
4. The only baseload power solution we have to meet this demand is nuclear power

BUT NUCLEAR POWER IS VERY EXPENSIVE TO OPERATE

1. Nuclear power is actually really cheap when it comes to fuel
2. What makes nuclear power expensive is capital and operating costs.
3. Capital costs are being solved by new modular reactors
4. Labor costs today actually get worse figuring modular reactors. Reason being same staff required for different MWh
5. This is the challenge we're going to take on. By making autonomous systems that are safe, we can eliminate reliance on human operators

Presentation Strategy

1. First, present a graph on energy consumption estimates in US in a graph.
2. Then, show the LCOE of different forms of energy production. Highlight the relative cost of labor and operating costs.
3. LCOE costs expect large nuclear reactors. Modular reactors on usually a third of the power, so labor costs are a big deal.
4. Bullet ending, we need to reduce labor costs of advanced nuclear power

SLIDE 2: STATE OF THE ART

Message

Modern nuclear reactor operation is highly prescriptive and labor intensive

1. We've been talking about labor, who's in the reactor room? Usually a senior reactor operator, and 2-3 reactor operators. Usually there's a chemist floating around too.
2. It's this staffing, 24/7/365
3. Reactor operators are extensively trained individuals. They have to train for multiple years and pass extensive and recurrent examinations.
4. Reactor operator jobs aren't always super attractive jobs. The work is somewhat monotonous, and requires individuals to usually live in very rural locations.
5. What is the work? Well, it's extremely prescriptive operations manuals. Nuclear reactors are so highly regulated and capital intensive that procedures and their creation happen in the design stage, before reactors are built. Safety is ensured at the design stage.
6. Thus, we're using humans basically as controllers for highly prescriptive tasks.
7. What are humans really good at? Well for the most part, general intelligence. Humans can use judgement and adapt to situations.

But, human operators in nuclear reactor operating rooms are trained and instructed to follow strict procedural guidelines.

8. There's also evidence that humans are actually not very good at being controllers.
9. Humans have very limited baud rates. We can only perceive so much information at a time, and the probability of human error increases dramatically when we perceive an emergency and are overwhelmed.
10. Enter, the whole damn field of human factors engineering. We bend over backwards to design control rooms and operating procedures to minimize the possibility of human error. This takes a lot of time and is expensive to implement.

Presentation Strategy

1. First, show a picture of a reactor operating room, and explain who the people are inside.
2. Explain how these operators are trained, the qualifications necesssary
3. What are they actually doing in here?
4. Split screen with reactor design photo. Show that there's a wall between them.
5. introduce point and details afterwards
6. Talk about how designing these procedures, building these control rooms, and training these people is extremely expensive

SLIDE 3: LIMITATIONS

Message

These are going to just be a summary of the limits.

Presentation strategy

Basically just a bulleted list of the limits.

SLIDE 4: RESEARCH APPROACH

Message

We will create high assurance autonomous control systems by breaking down the problem into smaller steps

1. One does not go from zero to hero easily with these systems.
2. Instead, we're going to create a chain of proof that our system is high assurance 2.1. We're ACTUALLY going to start by explaining we'll use hybrid control systems What is a hybrid system? well it's a system with both continuous dynamics and discrete dynamics. This is a system that both flows and jumps!.
3. We'll start with the procedures. We'll take the natural language and turn them into FRETish requirements
4. We can do realizability checks at this point
5. We take the requirements from FRET as temporal logical statements, and move to the next step
6. We take our temporal logic statements, and use reactive synthesis tools to break them down into discrete automata
7. These are our switching behvaior between continuous modes
8. Then, once we have this automata, we have two things:
   1. We have the switching behavior with the boundary conditions
   2. We have a map of how one mode goes to another mode.
9. At this point, we will build individual controllers for each of the discrete modes
10. To ensure the continuous dynamics actually satisfy boundaries between states, we will use a couple of techniques from formal methods.
    1. Reachability. Reachability will ensure that our input and output conditions only satisfy the discrete transition boundaries that we define.
    2. Barrier Certificates. These will ensure that on the interfaces, we won't develop zeno behavior

Each of these links together is what will allow us to prove that the whole system satisfies requirements.

Presentation Strategy

This isn't really going to be one slide. Instead, I'll present an arrow from left to right about what the steps are, and dive into each subpiece for a slide, then jump back out to the original slide.

The arrow should be from current operational procedure to autonomous hybrid control system. The steps should be

• requirement synthesis in FRET
• Reactive synthesis in STRIX or similar
• building individual control modes
• badda bing we're there.

SLIDE 5: METRICS OF SUCCESS

Message

In order to evaluate the progress of this research, we need to have a way to measure progress

1. This work is trying to make a real impact on building autonomous control systems in nuclear power. Because of this, the relevancy to industry partners is what's most critical.

To measure success, we're going to use technology readiness levels

1. We're shooting for TRL 5.
2. TRL 3 is critical function and proof of concept. This is individual components working in isolation. If we can bumble through each of these steps in a hacky way, I'll call that TRL 3. This isn't necesarily a flushed out control system.
3. TRL 4 is Laboratory Testing of Integrated Components. This is a bench top simulation of a complete hybrid autonomous control system. This includes a start up and shutdown procedure, and load following with checks for xenon poisoning and an ability to handle component failures.
4. TRL 5 is Laboratory testing in Relevant Environment. This is TRL 4, plus putting it on the Ovation control system instead of a purely code (MATLAB / PYTHON) simulation.

Presentation Strategy

Show a TRL timeline, With TRL 3, 4, 5 arrows. Insert Pictures along each step talking about what is what

SLIDE 6: RISKS AND CONTINGENCIES

Message

Possible challenges will be identified early and have planned mitigations

1. Computational tractability

   • It might be really hard to generate these automata and do reacability
   • Exponential scaling with specification complexity
   • Early indicators are synthesis times >24 hours, very large automata, etc.
   • Contingency is we can reduce scope to just a startup sequence
   • We can exploit time / scale separation of reactor dynamics too, and also use the high performance compute at CRC

2. Boolean guard conditions may not map cleanly to continuous guard conditions

   • early indicator: Continuous modes can't be built ot reach transition boundaries, and safety regions can't be expressed as polytopes.
   • contingency: Restrict to polytopic invariants where certain states are conservatively ignored. Sucks and requires manipulation but could get the job done.

3. Procedure Formalization is not within reach yet.

   • early indicator is we have a really hard time forming complete specifications in FRET from written procedures or synthesizing automata
   • contingency is we document the taxonomy and figure out what's missing to get us there. What is missing from the written procedures? This becomes a research contribution.

Presentation Strategy

Basically just top and bottom comparison of risk, and what the contingencies are

SLIDE 7: BROADER IMPACTS

Message

Automating nuclear reactor control is a billion-dollar-a-year problem

1. We need a lot of energy
2. The only clean baseload option is nuclear power
3. If we build advanced nuclear to meet this need, operating costs are expensive

But automating control can reduce operator burden, and significantly reduce operating costs.

Presentation Strategy

Basically copy over the one slider I made from earlier for the emerson CEO visit.

SLIDE 8: MONEY SLIDE

Message

Presentation Strategy