Prognostic Assessment (ISO 13374 Block 5)

Prognostic Assessment (ISO 13374 Block 5)

Estimate degradation trends and remaining useful life from vibration signals. Orchestrate MCP tools in a precise prognostic sequence.

Prerequisite: The predictive-maintenance-mcp MCP server must be connected.

Workflow

Step 1 — Signal Discovery

Call list_stored_signals() to check cached signals, or list_signals() to browse the data/ directory. Identify the signal file to analyze.

The prognostics tools segment a single signal file internally, extracting a feature series over time from consecutive segments. This works best on run-to-failure recordings or long monitoring sessions where degradation evolves within the signal.

Step 2 — Trend Analysis

Call analyze_signal_trend(signal_file=..., feature_name="rms").

The tool automatically segments the signal, extracts the chosen feature per segment, and fits a linear trend.

Parameters:

• feature_name: degradation indicator — "rms" (default), "kurtosis", "crest_factor", "peak_to_peak", etc.
• sampling_rate: auto-detected from metadata if not provided
• segment_duration: segment length in seconds (default: 0.1s)
• overlap_ratio: overlap between segments (default: 0.5)

Good feature candidates:

Indicator	Use when	Notes
rms	General degradation	Most common, ISO 20816 aligned
kurtosis	Bearing degradation	Peaks early, then may drop
crest_factor	Impulsive faults	Sensitive to early damage
peak_to_peak	Looseness, imbalance	Good for mechanical looseness

Ask the user which indicator to track, or default to rms if unsure.

Interpretation:

R-squared	Trend direction	Meaning
> 0.7	Increasing	Strong degradation trend
0.3 - 0.7	Increasing	Moderate trend, monitor closely
< 0.3	Any	No clear trend (or non-linear)
Any	Decreasing	Improving or post-maintenance
Any	Stable	Stationary condition

Decision gate: If trend_direction is "stable" or "decreasing", report that no degradation trend is detected. Ask if the user wants to proceed with RUL estimation anyway.

Step 3 — Degradation Onset Detection

Call detect_signal_degradation_onset(signal_file=..., feature_name="rms", threshold_sigma=3.0).

• Default threshold: 3 sigma (99.7% confidence)
• Lower to 2.0 for more sensitive detection (may trigger false positives)

If onset is detected, report:

• At which segment the degradation began (onset_segment_index)
• How many total segments were analyzed

Step 4 — RUL Estimation

Call estimate_rul(signal_file=..., failure_threshold=..., method="linear").

Choosing the failure threshold:

• If ISO 20816 zone C/D boundary is known, use it (e.g., 4.5 mm/s RMS for Group 2 rigid machines)
• If the user has a maintenance policy threshold, use that
• If unknown, ask the user — do NOT guess

Choosing sampling_interval:

• Defines the time unit of the returned RUL
• Default 1.0 means RUL is expressed in "segments"
• Adjust based on what the user needs (e.g., set to actual time between segments for real time units)

Choosing method:

• "linear" — simpler, works for steady degradation (default)
• "exponential" — for accelerating degradation curves
• "weibull" — Weibull distribution fit (if available)
• "kalman" — Kalman filter estimation (if available)
• Compare linear and exponential if unsure — the one with higher confidence (R-squared) is typically more appropriate

If RUL is infinity or confidence is 0, the curve doesn't reach the threshold. Explain why and suggest:

• The threshold may be too high for the current trend
• Try the other method
• The signal may not show sufficient degradation

Step 5 — Interpretation and Recommendations

Combine all findings into a concise prognostic summary:

1. Trend: direction, strength (R-squared), slope, feature analyzed
2. Onset: which segment degradation started (if detected)
3. RUL: estimated time to failure with confidence
4. Urgency: based on RUL vs. next planned maintenance window

Recommended actions based on RUL:

RUL	Urgency	Recommendation
< 1 interval	Critical	Immediate inspection/replacement
1-3 intervals	High	Schedule maintenance soon
3-10 intervals	Moderate	Monitor more frequently
> 10 intervals	Low	Continue routine monitoring

Important Notes

• RUL estimates are extrapolations — actual failure time depends on operating conditions that may change
• Always use cautious language: "estimated RUL", "projected failure time"
• Confidence (R-squared) below 0.5 means the estimate is unreliable
• This tool augments expert decision-making; it does not replace engineering judgment for maintenance scheduling
• All processing happens locally — raw signal data never leaves the machine