SensorDriftEvents

Calibration drift of an inline sensor per window: zero-point offset, span (gain) changes, the trend of the window mean or spread, and a 0-100 calibration health score, against a reference value, a reference signal or the first window.

ts_shape.events.quality.sensor_drift · Quality guide · API reference

Example

import ts_shape
from ts_shape.events.quality.sensor_drift import SensorDriftEvents

df = ts_shape.make_timeseries(["ph_probe_01"], n_points=4320, freq="1min", baseline=7.0, noise=0.02, drift=0.15)

drift = SensorDriftEvents(df, signal_uuid="ph_probe_01", reference_value=7.0)  # pH 7 buffer
zero = drift.detect_zero_drift(window="8h", threshold=0.05)
span = drift.detect_span_drift(window="8h")
health = drift.calibration_health(window="8h", tolerance=0.1)
print(zero[["start", "mean_offset", "drift_rate", "severity"]])

Methods

All arguments are keyword-only.

Method	Returns (per window)
detect_zero_drift(window='8h', threshold=None)	mean_offset, drift_rate, severity
detect_span_drift(window='8h')	sensitivity (window mean / reference), sensitivity_change_pct (vs the first window)
drift_trend(window='1D', metric='mean')	trend of the window mean or std: value, slope, r_squared, direction
calibration_health(window='8h', tolerance=None)	bias, precision (window std), health_score

Notes

• The reference is reference_value (a constant) or reference_uuid (a signal in the same frame, averaged per window). Without one, detect_zero_drift measures the offset from the first window's mean, and detect_span_drift needs a reference.
• drift_rate is the offset change from the previous window. severity is critical above threshold, high above 66 %, medium above 33 %, else low; without threshold it is 3 × the first window's standard deviation.
• With tolerance, health_score = 100 − up to 50 for bias (relative to tolerance) − up to 50 for precision (relative to tolerance / 3). Without it, the score is relative to the window's own spread.
• Windows with fewer than 2 samples are skipped.