【Control】🛡️ 13. FSM Design

Making Adaptive Control Explainable — What a Reliability Guard Really Is

topics: [“FSM”, “control design”, “reliability”, “adaptive control”]

---

❌ “The AI Just Stopped It” Is the Worst Answer

There is one phrase that adaptive control must never produce.

“Why did it stop?”
“…the AI decided.”

That answer is unacceptable in real systems.

If the reason cannot be explained,
the control system cannot be trusted.

---

🧭 The Role of the FSM Reliability Guard

The FSM Reliability Guard is a decision-only layer.

It does not:

• Optimize
• Learn
• Compute control values

It only does this:

Observe → Compare → Permit / Block

Nothing more.

---

📏 What the FSM Actually Monitors

Typical monitored metrics include:

• $R_{\Delta t} = \Delta t / \Delta t_0$
• $R_{K_P} = K_P / K_{P0}$

These metrics are chosen because they are:

• Dimensionless
• Ratio-based
• Directly explainable from logs

---

🔎 FSM Decisions Are Deterministic

FSM guard logic is explicitly defined.

if R_Δt > threshold:
    ADAPT_BLOCKED

Therefore, the system can always say:

“At this time, this value exceeded the threshold,
so adaptation was stopped.”

This is not interpretation.
It is execution of a rule.

---

🧯 Stopping Is the Correct Outcome

This is the core principle of the FSM Guard.

Stopping = not broken

Only a control system that can stop adaptation
can be used over long operational periods.

---

🔗 A-Type → B-Type → Robust Control

• A-Type: measure the limits
• B-Type: protect the limits
• Robust Control: operate within the limits

The FSM Reliability Guard
is the bridge between all three.

---

✅ Summary

• Adaptive control is not universal
• Therefore, a stopping design is required
• The stopping decision must be explainable via FSM

Adaptive control becomes engineering
only when it can explain itself.

---

End of Article