Reliability Analysis — AITL under Plant Aging (1000 days)

This section provides a reliability-oriented analysis of the
AITL Controller A-Type under long-term plant degradation,
modeled as friction aging equivalent to 1000 days.

Unlike conventional performance-driven evaluations,
the focus here is on temporal reliability:

• ⏱️ timing consistency (Δt)
• 🎛️ motion authority preservation
• 🚫 explicit decisions on when adaptation should be stopped

🔗 Links

Language	GitHub Pages 🌐	GitHub 💻
🇺🇸 English

🎯 Purpose and Scope

The goal of this analysis is not to optimize controller performance.

Instead, it aims to answer the following design questions:

• How does plant aging affect response timing?
• What does adaptive control actually preserve?
• When does adaptation begin to harm reliability?
• Can this degradation be quantified and detected by design?

The analysis compares:

• fixed-gain PID control
• AITL control with FSM-based adaptive gain retuning

under identical friction aging conditions.

🧩 Analysis Structure (Demo Mapping)

This reliability study is organized as a design-to-evidence sequence, linking architectural intent to reproducible results.

Rather than independent experiments, each demo represents a progressive refinement of reliability reasoning.

Interpretation

• Demo 12 answers what happens under aging and disturbance
• Demo 13 quantifies how reliability degrades using explicit metrics
• Demo 15 explains why supervisory decisions occur, enabling auditability

Together, these demos establish that:

• performance recovery alone is insufficient to guarantee reliability
• temporal consistency (Δt) must be monitored explicitly
• adaptive actions require design-time permission and stopping logic

This sequence defines the design boundary of the A-Type controller and provides the evidence base motivating a reliability-oriented B-Type architecture.

Each demo builds on the previous one and represents a
design-phase progression, not isolated experiments.

🔬 Demonstration Results (Reproducible Evidence)

This section presents direct, reproducible results generated from the Python demos in demos/ and stored in data/.

All figures below are auto-generated and traceable to code.

1️⃣ V–I Current Control under Aging & Disturbance

Waveform comparison of Fixed PID / PID×FSM / AITL under:

• resistance (friction-equivalent) aging
• external current disturbance

Observations

• Fixed PID: recovery depends on nominal tuning
• PID×FSM: supervised gain switching improves robustness
• AITL: adaptive recovery under degradation

⚠️ Waveforms alone are not sufficient to judge reliability.
Quantitative metrics follow.

2️⃣ Reliability & Safety Metrics vs Aging (Δt, max|e|)

Aging sweep over resistance step ratio (ΔR/R0), evaluating:

• Δt: recovery/settling time (temporal reliability)
• max|e|: worst deviation (safety envelope)

Interpretation

• Δt increases with aging for all controllers
• AITL maintains comparable safety (max|e|)
• Temporal reliability degradation becomes explicit and measurable

3️⃣ FSM Explainability — Why Adaptation Switched

Explicit visualization and logs explaining why FSM mode switches occurred, based on thresholded error magnitude.

What this proves

• Supervisory decisions are deterministic
• Mode switches are auditable and timestamped
• FSM-based adaptation is not a black box

♻️ Reproducibility

All results shown above are generated by the following scripts:

• demos/12_vi_current_control_sales_demo.py
• demos/13_aging_sweep_delta_t.py
• demos/15_fsm_explainability_demo.py

Generated figures are stored in:

• data/12_vi_current_control_sales_demo.png
• data/13_aging_sweep_delta_t.png
• data/15_fsm_explainability_demo.png

No manual post-processing is applied.

🧠 Reliability Design Conclusion (Evidence-Based)

Based on the results above:

• adaptive control can compensate for aging-induced delay
• adaptation may degrade temporal reliability
• this degradation is detectable by design-time metrics (Δt)
• FSM-based logic can decide when adaptation should be stopped

These results define the design boundary of the A-Type controller and motivate a B-Type architecture explicitly designed for reliability control.

🧩 Key Design Message

Adaptive control is not automatically reliable.

In this study, AITL successfully compensates for delay,
but at the cost of:

• excessive lead (over-compensation)
• reduced motion authority
• and degraded overall reliability

Crucially, these effects are:

• not obvious from waveforms alone
• but become explicit through Δt, amplitude, and FSM logic

This distinction marks the boundary between
performance optimization and reliability-oriented design.

🧭 Navigation

▶ Detailed Demo Analyses (Reproducible Results)

• Demo 12 — V–I Current Control under Aging & Disturbance
  Phenomenological waveform comparison (Fixed PID / PID×FSM / AITL)
  → Results: data/12_vi_current_control_sales_demo.png
  → Code: demos/12_vi_current_control_sales_demo.py

• Demo 13 — Reliability Metrics vs Aging (Δt, max|e|)
  Quantitative evaluation of temporal reliability and safety degradation
  → Results: data/13_aging_sweep_delta_t.png
  → Code: demos/13_aging_sweep_delta_t.py

• Demo 15 — FSM Explainability (Why Adaptation Switched)
  Audit-ready visualization of supervisory decisions and thresholds
  → Results: data/15_fsm_explainability_demo.png
  → Code: demos/15_fsm_explainability_demo.py

These demos constitute the evidence layer supporting the reliability conclusions of the A-Type controller.

🖼 Reference Figure

• Timing degradation under friction aging
  pid_vs_aitl_friction_aging.png

This figure provides a phenomenological overview only.
All reliability conclusions are derived from quantitative metrics
and FSM-based decisions described in the demos above.

🔗 Relation to Other Documentation

• This section corresponds to the Reliability Investigation
  introduced briefly in the main documentation index.
• The index page presents only representative conclusions.
• Detailed reasoning, metrics, and design implications are contained here.

✅ Summary

This reliability chapter demonstrates that:

• timing degradation can be quantified (Δt)
• authority loss can be detected (amplitude ratio)
• adaptive behavior can and should be stopped by design

The result is not a failure of AITL,
but a clear design boundary between:

Adaptive control
and
Reliable adaptive control

➡️ Design Implication and Next Step

The A-Type controller successfully demonstrated adaptive control capability under plant aging conditions.

While the feasibility of reliability-oriented control was investigated, the current A-Type architecture was not designed to guarantee reliability, as adaptive actions may degrade timing consistency and motion authority.

This result clarifies the design boundary of the A-Type controller and motivates the investigation of a B-Type architecture explicitly designed for reliability control, in which adaptive actions are evaluated and accepted only if overall reliability is improved.