.. _auto_curve_failure_regression:

*DETECT FAILURE POINT ON A CURVE VIA SLIDING LINEAR REGRESSION*
===============================================================

Slides a linear-regression window across the curve and reports the index where R^2 first drops below the regression limit (= the curve stops being linear, signalling failure). Returns failure index/time/value and the pre-failure linear slope/intercept fit on the last good window.

When to use
-----------

Tagged: ``curve``, ``failure_detection``, ``linear_regression``, ``sliding_window``, ``r_squared``.

Inputs
------

.. list-table::
   :header-rows: 1
   :widths: 20 20 20 20 20 20

   * - Label
     - ID
     - Type
     - Default
     - Required
     - Description
   * - Curve
     - curve_in
     - vector
     - —
     - ✓
     - Curve to scan for the failure point; for FD-style calibration this is typically the displacement-vs-time signal.
   * - Percentage of Points (sliding window)
     - percentage_of_points
     - number
     - 0.01
     - 
     - Fraction of total curve points used as the sliding regression window (default 0.01 = 1%).
   * - Regression R^2 Limit
     - regression_limit
     - number
     - 0.9
     - 
     - R^2 threshold below which the regression window is considered non-linear and failure is declared (default 0.7).
   * - Extrapolate Past Failure
     - extrapolate_post_failure
     - select
     - no
     - 
     - When yes, the worker also emits an Extrapolated Curve output that replaces every post-failure point's y with a linear extrapolation from the failure anchor along the pre-failure regression slope. Pre-failure points pass through unchanged. Default no — extrapolated_curve mirrors the input curve.
   * - Pre-Failure Smoothing Window
     - pre_failure_smoothing_window
     - number
     - 1
     - 
     - Number of trailing pre-failure regression windows to average for the reported pre_failure_slope and pre_failure_intercept (and for the post-failure extrapolation when Extrapolate Past Failure = yes). Default 1 — use only the most recent regression (legacy behavior). 2+ averages the last N windows so a single noisy regression near the failure event doesn't skew the result.

Outputs
-------

.. list-table::
   :header-rows: 1
   :widths: 20 20 20 20

   * - Label
     - ID
     - Type
     - Description
   * - Failure Index
     - failure_index
     - scalar
     - Sample index in the input curve where the sliding regression first failed the R^2 limit.
   * - Failure Time
     - failure_time
     - scalar
     - X value (typically time) at the failure index.
   * - Failure Value
     - failure_value
     - scalar
     - Y value at the failure index.
   * - Pre-Failure Slope
     - pre_failure_slope
     - scalar
     - Slope of the last regression window that satisfied the R^2 limit.
   * - Pre-Failure Intercept
     - pre_failure_intercept
     - scalar
     - Intercept of the last regression window that satisfied the R^2 limit.
   * - Status
     - status
     - string
     - Human-readable summary including detected index, time, and parameters used.
   * - Extrapolated Curve
     - extrapolated_curve
     - vector
     - Curve where every post-failure point's y is replaced with the pre-failure regression line. Same x-grid as the input. Equals the raw input when Extrapolate Past Failure is no or no failure was detected.
   * - Input Curve (Passthrough)
     - input_curve
     - vector
     - Echoes the curve_in input back as an output. Lets downstream visualizers (e.g. curves_overlay) wire BOTH input + extrapolated curves from this single worker without a second edge from the start node.

Disciplines
-----------

- data.curve.transform
- engineering.material.characterization

.. raw:: html

   <hr style="margin-top:2em">
   <p style="font-size:11px;color:#888">
   Auto-generated from <code>platform</code> schema. Worker id: <code>curve_failure_regression</code>. Schema hash: <code>4270b24adf91</code>. Hand-curated docs in <code>workerexamples/</code> override this page when present.
   </p>