Day3_​S1_​ok

1 · A value that cannot be true

A dead transmitter writes a number, not a blank.

Step 1 · Read the minute file and flag the impossible values

1. CSV Reader reads day3_pump02_minute_v4.csv and String to Date&Time parses timestamp as yyyy-MM-dd HH:mm:ss: 173,549 rows, 0 failures
2. One Expression node, four formulas evaluated in order. The first APPENDS flag_range = if(($bearing_temp_c ?? 50) < 20 or ($bearing_temp_c ?? 50) > 120 or ($flow_m3h ?? 1) < 0 or ($vib_cm_mm_s ?? 1) < 0, 1, 0)
3. The other three REPLACE_EXISTING the offending cell only, e.g. if(($flow_m3h ?? 1) < 0, MISSING, $flow_m3h) - the flag is raised while the value is still readable
4. Output of the stage: 173,549 rows, 12 of them carrying flag_range = 1 and one blank cell each. No row is deleted, so the two healthy tags in those rows stay usable
5. The lower lane is a dead end that hangs off String to Date&Time, before any repair: Statistics shows the table exactly as the TS DB delivered it - 173,549 rows, 2024-12-01 00:00 to 2025-03-31 23:59, blank cells 45 / 45 / 120 on the three tags, and a minimum bearing_temp_c of -999.0 that is still sitting in the data at that point

2 · The same minute written twice

Only the later write is the corrected value.

Step 2 · Keep the corrected write, and remember which minute was rewritten

1. Sorter orders by timestamp ascending, so the two writes of one minute end up adjacent
2. Expression APPENDS flag_dup = if($["timestamp", -1] = $timestamp, 1, 0). The flag must be raised BEFORE the row goes: measured on this machine, Duplicate Row Filter writes no status column when it removes
3. Duplicate Row Filter on timestamp, keep LAST: 173,549 rows become 173,449
4. Output: one row per written minute, 100 of them carrying flag_dup = 1 all the way into the DQ report

3 · One row per minute, and what is behind each hole

Rows are not one minute apart, and the TS DB never writes an empty row.

Step 3 · Put every minute on the axis, then tell a slow logger from a stopped plant

1. The lower lane measures the holes: Expression carries $["timestamp", -1] as ts_prev, Date&Time Difference gives gap_min in MINUTE, Row Filter keeps gap_min >= 2 and the Histogram shows 301 gaps: 2 min x172, 3 x64, 4 x32, 5 x30, then nothing at all until 26, 71 and 181
2. Two cleanly separated populations - that measurement is what puts the sparse/outage cut at 5 minutes
3. Date&Time Range Creator builds the 174,240 minute calendar (2024-12-01 00:00 to 2025-03-31 23:59, PT1M), Joiner LEFT with the calendar on port 1 leaves 791 minutes unmatched, Sorter puts the grid back in time order because the Joiner returns ARBITRARY
4. Expression then builds cov = ($p_temp+$p_flow+$p_vib) / 3 and a 0/1 presence column, and two Moving Aggregator windows of 6 rows, one BACKWARD and one FORWARD, ask: is there a real reading within 5 minutes on this side? Output: Sum(pres_b) and Sum(pres_f) on all 174,240 rows

4 · A sensor that stopped moving

The value still arrives every minute. It just never changes.

Step 4 · Catch the stuck tag on the minute axis

1. Expression APPENDS the minute-to-minute change of all three tags, e.g. d_flow = abs($flow_m3h - $["flow_m3h", -1])
2. Moving Aggregator with windowLength 10 and windowType BACKWARD takes Sum and Count of d_flow, d_temp and d_vib. On the calendar 10 rows is exactly 10 minutes, and a window lying across a hole returns Count under 10
3. Expression APPENDS flag_frozen = 1 when ($["Sum(d_flow)"] ?? 1) = 0 and ($["Count*(d_flow)"] ?? 0) = 10, or the same pair for d_temp or d_vib. Output: 190 minutes flagged, every one of them flow
4. The lower lane proves it: Date&Time-based Row Filter cuts 14 Jan 02:00-07:00, 300 minutes, and the Line Plot shows flow flat at 161.45 m3/h for 200 of them, with normal signal on either side

5 · One label per minute

Resolve every flag into one verdict, then count it.

Step 5 · Turn the flags into one quality label per minute and count them

1. Expression APPENDS quality = if($cov = 0, if($["Sum(pres_b)"] >= 1 and $["Sum(pres_f)"] >= 1, "sparse", "outage"), if($flag_frozen = 1, "frozen", if($cov < 0.34, "low_cov", "ok")))
2. An empty minute with a real reading on BOTH sides within 5 minutes is sparse; with none on a side it is outage
3. The same node APPENDS is_ok = if($flag_frozen = 0 and $cov >= 0.34, 1, 0), the one gate the hourly roll-up uses
4. Output: ok 173,214 - sparse 516 - outage 275 - frozen 190 - low_cov 45 = 174,240. GroupBy on quality, flag_dup AND flag_range gives 7 rows, written by the CSV Writer to day3_s1_dq_report.csv

6 · Hand over to Session 2

Only the hours you can defend leave here - and you look at them before you trust them.

Step 6 · Roll up to hours, write the file, and read the three series

1. Time Rounder APPENDS timestamp (hour) at HOURS_1 and Row Filter keeps is_ok = 1: 173,214 minutes. Nothing is imputed anywhere in this workflow
2. GroupBy on timestamp (hour) gives 2,898 hours - Mean for temperature and flow, Maximum for vibration, a Count per tag - and Column Renamer writes back ts_hour, bearing_temp_c, flow_m3h, vib_cm_mm_s, n_temp, n_flow, n_vib, n_min
3. Row Filter keeps n_temp, n_flow AND n_vib >= 30, per tag because n_min never sees an hour where one single tag is empty: 3 hours go and 2,895 rows reach day3_s1_hourly.csv. At 30 real minutes the hourly figure is off by at most 0.042 C, 0.29 m3/h and 0.033 mm/s
4. Three Line Plot nodes on one lane, one tag per axis: 50.0-71.1 C, 103-184 m3/h and 2.16-5.38 mm/s cannot share a y axis. Temperature reads the renamed table (2,898 h, no blank cell), flow and vibration read the filtered 2,895 h; the three dropped hours are invisible at this scale