Day3_​S3_​Health_​Indicator

1 · Readings and Reference

Read the monitoring readings and the reference-period file.

Step 1 · Read both inputs

1. Read Day 2 table, drop z-columns: CSV Reader
2. Parse, sort by pump then time: String to Date&Time, Sorter
3. Fill sensor gaps: Missing Value
4. Same for the reference-period file (raised row)

2 · Combine and Model

Stack both periods, fit a driver-based baseline, score the residual.

Step 2 · Fit the driver baseline

1. Stack the 2 periods with Concatenate
2. 8h mean flow, hour-of-day, sin(hour) with Moving Aggregator, Math Formula
3. Learn temp from drivers on 2024H1 with Linear Regression Learner
4. Apply the frozen model, compute the residual with Regression Predictor

3 · Flag the Residual

Fleet-wide baseline, then per-pump-per-regime.

Step 3 · Flag against both baselines

1. Flag fleet-wide 3SD with Rule Engine
2. Join the per-pump-regime baseline with Joiner
3. Flag per-pump-regime 3SD with a second Rule Engine

4 · Trend Per Pump

One line per pump shows a rising machine before it fails.

Step 4 · Plot the trend

1. Pull Day of year out of the timestamp with Extract Date&Time Fields
2. Keep the worst indicator per pump per day with Pivot
3. Draw one line per pump with Line Plot

5 · Compare and Write

Compare the two flags, then write the state file for Day 4.

Step 5 · Compare and export

1. Compare fleet vs per-pump-regime flag with Table View
2. Format the timestamp with String Manipulation
3. Keep 5 columns, write day03_state.csv

หมายเหตุ

เพราะฉะนั้นวันนี้เปลี่ยนไปหาเหตุจากตัวขับแทน

Explore before you decide

Profile the input, then sweep the lag/window that justifies MA_H=8.

Explore · Profile and correlation sweep

1. Profile the readings table with Statistics
2. Lag bearing_temp_c 1-12h with Lag Column
3. 8h mean flow with Moving Aggregator, per pump
4. Correlate with Linear Correlation, plot with Line Plot