【Control】🛑 06. What Reliability Control Needed Was Not “Going Faster,” but Knowing When to Stop

topics: [“control engineering”, “PID control”, “reliability”, “FSM”, “simulation”]

Introduction (Continuation from the Previous Article)

In the previous article, we demonstrated that:

Under friction aging, control strategies that preserve only timing (Δt) inevitably fail.

Through a comparison between:

we reached a clear conclusion:

In this article, we go one step further and ask:

What was actually missing for “Reliability Control”?

❌ The Problem Was Not “Retuning”

First, to avoid misunderstanding:

The AITL implementation in this study did succeed in:

Yet it still failed as reliability control.

The reason turned out to be very simple.

🧠 What Was Missing Was the Ability to “Stop”

The FSM logic in AITL was essentially:

However, one critical question was never asked:

Did this retuning actually improve reliability?

That judgment was completely absent.

📊 What Was Actually Happening (Numerical View)

Let us revisit the results under friction aging equivalent to 1000 days.

Controller | Δt mean [s] | |Δt| [s] | Amp ratio
----------------------------------------------
PID_only  |   -0.4730   |  0.4730 |  0.902
AITL      |   -1.3807   |  1.3807 |  0.888

In other words:

A well-intended retuning action was actively degrading reliability.

🧩 The Concept of a Reliability FSM

This led to the introduction of a new idea:

An FSM that judges whether retuning is allowed at all.

The principle is extremely simple.

🔁 Three States Are Sufficient

The FSM only needs three states:

The key insight is:

LEAD (too fast) must also be treated as degradation.

⚠️ “Too Fast” Is Also an Abnormal Condition

From a traditional control engineer’s intuition:

But from a reliability perspective, this is incorrect.

An overly advanced response:

Thus:

LEAD is a form of hidden reliability degradation.

🛡 Guard Conditions as a Design Decision

The following guard condition was added to the FSM:

The resulting behavior was:

AITL | State = LEAD
     | Amp ratio = 0.888
     | Action = BLOCK

This enabled the system to decide:

“Any further adjustment will break reliability.”

💡 The Most Important Insight

The core lesson from this study is:

Reliability control is not about continuously improving performance,
but about stopping before things get worse.

❓ Did AITL Fail?

No.

The AITL implementation (A-Type):

This is not a failure, but a clear design milestone:

It clarified where adaptive control ends and reliability control begins.

▶ What Comes Next

At this point, the following have become clear:

The next step is therefore:

Designing a new AITL (B-Type)
with Reliability as the explicit objective function

Summary

🔗 References

👉 Next Preview
“PID vs AITL Through Reliability Cost — Why ‘Ordinary PID’ Still Wins”