topics: [“PID”, “FSM”, “control engineering”, “reliability design”, “system design”]
— Final Consolidation of AITL B-Type Control
Throughout this series, we have examined PID control, FSMs, reliability control,
and long-term degradation to clarify one central question:
What should be automated, and what must remain a human decision?
The conclusion is surprisingly simple:
PID performs control,
FSM performs judgment,
gains are decided by humans.
With this, reliability control is complete.
PID control is classical and not academically novel.
Yet in real-world systems, it remains unmatched due to:
The most critical factor is explainability.
In real products, any control behavior that cannot be explained
will not be accepted.
PID remains the optimal solution in this regard.
FSMs are often misunderstood as:
However, as demonstrated in this series,
systems become fragile when FSMs actively control behavior.
The correct role of an FSM is strictly limited:
To judge whether the system is still in a normal state
or has deviated from its expected behavior.
An FSM must not:
It is a supervisory judgment layer only.
The “10%” threshold appears repeatedly in this series.
But the number itself is not the essence.
What truly matters is:
The threshold could be:
That value is a design parameter,
chosen by engineers based on system, application, and safety requirements.
This design does not advocate “10% control.”
It advocates explicitly designing thresholds.
When degradation progresses, it may seem elegant to
“compute optimal gains on the fly.”
In real systems, this is dangerous:
Therefore, B-Type control enforces a strict rule:
Gains are designed, verified, and approved offline.
They are never computed during operation.
The FSM only decides
whether to apply a pre-approved gain set or not.
Large Language Models are powerful,
but they do not need to be embedded into control loops.
In this design, LLMs are limited to:
Final decisions are always made by human engineers.
LLMs are advisors, not decision-makers.
This boundary is what makes the design usable in practice.
This design is not academically cutting-edge.
Yet this is precisely why it works in the field:
This is not a weakness.
Designs strong in papers and designs strong in the field
are not the same.
This series deliberately chose the latter.
Beyond this point:
So we stop here.
PID controls,
FSM judges,
humans design.
With this three-layer structure,
reliability control is fully realized.
This article does not claim to present the only correct answer.
However, it can confidently state this:
This is a design endpoint that can be used to the end in real systems.
Nothing more needs to be added.
Nothing needs to be removed.
Design is also the act of
deciding where to stop.
This is that stopping point.
End of Final Article