Day3_​S2_​Feature_​Factory_​Lite

1 · Read and Order

Read the S1 output, parse timestamps, sort by pump then time.

Step 1 · Read and prepare

1. Read day03_s1.csv with CSV Reader
2. Parse the timestamp with String to Date&Time
3. Sort by pump then time with Sorter

2 · Smooth and Flag

Rolling mean per pump, then a quality rule on the result.

Step 2 · Smooth per pump

1. Loop per pump with Group Loop Start
2. Rolling mean with Moving Aggregator, collect with Loop End
3. Flag with Rule Engine

3 · RMS, Crest, Kurtosis

Unpivot the sensors, compute one formula per feature, pivot back.

Step 3 · Feature factory

1. Unpivot sensors with Unpivot, loop per column
2. RMS / crest / kurtosis with Math Formula, Moving Aggregator
3. Collect with Loop End x2, pivot back with Pivot
4. Write day03_s2.csv with CSV Writer

6 · AR-on-Self Limitation

Keep the old lag-1 model to show why it under-reacts.

Step 6 · The limitation, kept

1. Lag the signal 1 step with Lag Column
2. Fit on the reference period with Linear Regression Learner
3. Score the residual with Regression Predictor, Numeric Scorer

หมายเหตุ

AR ทำได้แค่นี้ เศษเหลือขาวก็จริง แต่การเสื่อมหายไปกับแบบจำลอง

Explore before you decide

Profile the data and justify the 24h window before it gets locked.

Explore · Justify the 24h window

1. Profile the aligned table with Statistics
2. Per-pump 24h kurtosis probe, label normal/degraded with Rule Engine
3. Decision picture: kurtosis by period with Box Plot
4. After the fact: do RMS/crest/kurtosis agree? Linear Correlation, Box Plot