capability_trending

capability_trending ¤

CapabilityTrendingEvents ¤

CapabilityTrendingEvents(
    dataframe: DataFrame,
    signal_uuid: str,
    *,
    upper_spec: float | None = None,
    lower_spec: float | None = None,
    upper_spec_uuid: str | None = None,
    lower_spec_uuid: str | None = None,
    value_column: str = "value_double",
    event_uuid: str = "quality:capability_trend",
    time_column: str = "systime",
    sigma_method: SigmaMethod = "moving_range"
)

Bases: Base

Quality: Process Capability Trending

Track process capability indices (Cp, Cpk, Pp, Ppk) over rolling time windows to detect capability degradation before quality escapes occur.

Specification limits can be provided as fixed floats or as UUIDs referencing signal rows in the DataFrame (matching the ToleranceDeviationEvents pattern).

Sigma (AIAG SPC manual; Montgomery, Introduction to Statistical Quality Control, section 8.3), per window:

Cp = (USL - LSL) / 6 sigma_within, Cpk = min(USL - mean, mean - LSL) / 3 sigma_within, with the within (short-term) sigma MR-bar / 1.128 of consecutive samples in the window (sigma_method="moving_range", default);
Pp / Ppk: the same formulas with the window's overall sample standard deviation s (ddof=1).

sigma_method="std" uses s for Cp/Cpk as well (Cp = Pp, the earlier Cp/Cpk). Before October 2026, Pp/Ppk combined the window mean with the standard deviation of the whole signal (later windows included); they now use only the window's own data.

Methods: - capability_over_time: Cp/Cpk/Pp/Ppk per time window. - detect_capability_drop: Alert when Cpk falls below threshold. - capability_forecast: Extrapolate Cpk trend to predict threshold breach. - yield_estimate: Estimated yield, DPMO, and sigma level per window.

Example

import ts_shape df = ts_shape.make_timeseries(["dia"], n_points=960, freq="1min", drift=3.0) cap = ts_shape.CapabilityTrendingEvents(df, "dia", upper_spec=105, lower_spec=95) out = cap.capability_over_time(window="8h") out[["start", "cpk", "ppk"]].round({"cpk": 2, "ppk": 2}) start cpk ppk 0 2025-01-01 00:00:00 1.56 1.33 1 2025-01-01 08:00:00 0.94 0.83

capability_over_time ¤

capability_over_time(window: str = '8h') -> pd.DataFrame

Compute Cp, Cpk, Pp, Ppk per time window.

Each window uses only its own samples (see the class docstring): Cp/Cpk from the within sigma (MR-bar / 1.128 of consecutive samples; the window's s with sigma_method="std"), Pp/Ppk from the window's sample standard deviation s (ddof=1).

Parameters:

Name	Type	Description	Default
`window`	`str`	Resample window size.	`'8h'`

Returns:

Type	Description
`DataFrame`	DataFrame with columns: start, end, duration_seconds, uuid,
`DataFrame`	source_uuid, cp, cpk, pp, ppk, mean, std (the window's `s`),
`DataFrame`	std_within (the sigma behind cp/cpk), n_samples.
`DataFrame`	`end` is `start` plus one `window`. Windows with fewer than
`DataFrame`	two samples are skipped. Without spread (constant values) the
`DataFrame`	indices are NaN; a signal without any spread also emits a
`DataFrame`	class:`~ts_shape.errors.DataQualityWarning`.
`DataFrame`	Canonical twin columns: `sample_count` (after `n_samples`); each holds
`DataFrame`	the same values as the deprecated column right before it, which is removed
`DataFrame`	in the next major release.

detect_capability_drop ¤

detect_capability_drop(
    window: str = "8h",
    min_cpk: float = 1.33,
    lookback: int = 5,
) -> pd.DataFrame

Detect windows where Cpk drops below threshold or degrades significantly.

Parameters:

Name	Type	Description	Default
`window`	`str`	Resample window size.	`'8h'`
`min_cpk`	`float`	Minimum acceptable Cpk value.	`1.33`
`lookback`	`int`	Number of previous windows for rolling average comparison.	`5`

Returns:

Type	Description
`DataFrame`	DataFrame with columns: start, end, duration_seconds, uuid,
`DataFrame`	source_uuid, cpk, prev_avg_cpk, drop_pct, alert.
`DataFrame`	`end` is `start` plus one `window`.

capability_forecast ¤

capability_forecast(
    window: str = "8h",
    horizon: int = 5,
    threshold: float = 1.33,
) -> pd.DataFrame

Extrapolate Cpk trend to predict when it will breach threshold.

Parameters:

Name	Type	Description	Default
`window`	`str`	Resample window size.	`'8h'`
`horizon`	`int`	Number of future windows to forecast.	`5`
`threshold`	`float`	Cpk threshold to predict breach for.	`1.33`

Returns:

Type	Description
`DataFrame`	DataFrame with columns: start, end, duration_seconds, uuid,
`DataFrame`	source_uuid, cpk, trend_slope, forecast_cpk,
`DataFrame`	windows_to_threshold. `end` is `start` plus one `window`.

yield_estimate ¤

yield_estimate(window: str = '8h') -> pd.DataFrame

Estimate yield, DPMO, and sigma level per window.

Assumes normal distribution within each window.

Parameters:

Name	Type	Description	Default
`window`	`str`	Resample window size.	`'8h'`

Returns:

Type	Description
`DataFrame`	DataFrame with columns: start, end, duration_seconds, uuid,
`DataFrame`	source_uuid, estimated_yield_pct, dpmo, sigma_level.
`DataFrame`	`end` is `start` plus one `window`.