Report
Unit 1303 is flagged with a high failure risk of , at or above the false-alarm rate threshold. Sensor S060 contributes most to the model's risk score, with a SHAP value of log-odds, and its sensor-health check reports status ok. Sensors S200, S336 and S472 are on watch status with out-of-reference rates of , and , and they are the largest contributors to the SPE and T² statistics for this unit. The EWMA alarm is active across the units in the FDC window, while the T² and SPE values stay below their control limits and the FDC alarm level is none.
Findings
- 1.
Sensor S060 contributes most to the model's risk score for Unit 1303, with a SHAP value of log-odds.
Strong evidenceVerifier: not checked - 2.
Sensors S200, S336 and S472 are on watch status, with , and of readings after the reference outside the reference range respectively.
Moderate evidenceVerifier: not checked - 3.
Sensors S200, S336 and S472 are the largest contributors to the SPE and T² statistics for Unit 1303, which coincides with their watch status.
Moderate evidenceVerifier: not checked - 4.
The EWMA alarm is active for Unit 1303 and the units in the FDC window, while the T² and SPE values remain below their respective control limits and the FDC alarm level is none.
Moderate evidenceVerifier: not checked
Proposed actions
- 1.
Check the calibration and physical condition of sensor S060, even though its health status is ok.
S060 has the largest SHAP contribution of log-odds to the model's risk score for Unit 1303, and its health status is ok with an out-of-reference rate near .
- 2.
Inspect the chambers and process conditions associated with sensors S200, S336 and S472, comparing their readings with the reference medians.
These sensors are on watch status, with , and of readings after the reference outside the reference range, and they are the largest contributors to the SPE and T² statistics for Unit 1303.
- 3.
Review the recent process history for the units in the FDC window where the EWMA alarm is active.
The EWMA alarm is active across the whole window of units while the T² and SPE values stay below their control limits, so an engineering check of the recent history is warranted.
This recorded example is the same for every visitor, so nobody decides it. To decide a report of your own, run a sample on the line and investigate one of its alarms.
Open the lineData & Process Limitations
5 notesStatistical boundaries and observational constraints established for this run:
The evidence shows statistical associations only; it cannot establish that any sensor or process condition causes the failure risk.
SHAP values describe the model's risk score in log-odds and the T² and SPE contributions describe the control chart, not the physical process.
16 sensor readings are missing for Unit 1303, so the risk score and contributions rest on incomplete input.
The PCA reference period used for FDC limits covers 106 units and is flagged as using the shortest candidate stretch according to the reference period evidence.
No significant sensor groups were identified (0 of 468 sensors), so no group-level association is supported.