Understand the three-sigma check intuitively

Compare each sensor with its normal distribution and distinguish a statistical alert from an explicit quality signal.

Question for this chapter

What does a three-sigma check find, and what effect does it have on these three alerts?

Why this matters now

Sigma sounds abstract, but its job is simple: express how many standard deviations a reading is from the average of the same sensor type. That creates a relative scale across different units.

Try it

Open the anomaly_detection pipeline under processed and select Run. It runs sensor_normalization and then anomaly_detection to populate anomaly_detections.

Read these two rules separately in the code.

Statistical rule: |value - mean for this sensor type| / standard deviation >= sigma
Status rule: if quality_flag starts with HIGH_, severity is HIGH regardless of z-score

Success looks like this

anomaly_detections contains three rows, all HIGH.

Machine	Sensor	z-score	Severity	Direct reason for HIGH
CNC-03	vibration	2.83	`HIGH`	`HIGH_VIBRATION`
CNC-05	pressure	1.15	`HIGH`	`HIGH_PRESSURE`
CNC-09	temperature	1.92	`HIGH`	`HIGH_TEMP`

Korean Portal anomaly results showing CNC-03, CNC-05, and CNC-09 all classified HIGH — The Portal capture is in Korean. All three alerts remain HIGH with their quality signal and z-score visible.

All three z-scores are below the default sigma of 3.0. The statistical rule added no rows in this sample; the source quality signals preserved all three.

Change the threshold once

Change sigma from 3.0 to 3.15 in the anomaly_detection code and run it again. The result is still the same three HIGH rows. Restore 3.0 after the experiment.

Next decision

All three alerts are valid. Next, compress them into machine-level health scores and choose a first stop among the tie.