FailurePredictionEvents

Remaining useful life by linear extrapolation of the recent trend to a failure threshold, escalating warning/critical exceedances per window, and time to a threshold from the current rate of change.

ts_shape.events.maintenance.failure_prediction · Engineering guide · API reference

Example

import ts_shape
from ts_shape.events.maintenance.failure_prediction import FailurePredictionEvents

df = ts_shape.make_timeseries(["bearing_temp"], n_points=4320, freq="1min", baseline=60.0, noise=0.5, drift=25.0)

fp = FailurePredictionEvents(df, "bearing_temp")
rul = fp.remaining_useful_life(degradation_rate=0.0001, failure_threshold=95.0)
print(rul[["systime", "current_value", "rul_hours", "confidence"]].tail())
pattern = fp.detect_exceedance_pattern(80.0, 82.0, window="1D")
tte = fp.time_to_threshold(95.0, direction="increasing")

Methods

Method	Returns
remaining_useful_life(degradation_rate, failure_threshold)	per sample: current_value, rul_seconds, rul_hours, confidence
detect_exceedance_pattern(warning_threshold, critical_threshold, window='1h')	per window: warning_count, critical_count, escalation_detected
time_to_threshold(threshold, direction='increasing')	per sample: current_value, rate_of_change, estimated_time_seconds

Notes

• degradation_rate (units per second) is the fallback slope when no local slope can be computed; its sign gives the failure direction (positive: failure when the value rises to the threshold).
• rul_* is 0 once the threshold is reached and None when no failure is projected (flat trend or a trend moving away from the threshold).
• Exceedances compare the absolute value with the thresholds; escalation_detected marks increasing counts.
• On a signal without a real trend the RUL estimate is meaningless; confirm the trend first, e.g. with DegradationDetectionEvents.
• For a forecast with a fit quality and named thresholds see TimeToThresholdEvents.