Process Control Parameter Optimisation · Wisdom notebook

Ask the follow-up.

Agentic conversational BI · one thread over the quality lab, the process historian and what the two lines actually produced.

Which lines are running defect rate above their learned band this week, and how far above it is the worst one?

Pulled defect rate (DPM) by line from the quality lab for the week, scored each line against its own learned band rather than against the other line, and ranked what came out.

One line, and it is well clear of the band. Line 2 is comfortably inside it; Line 1 is the one running hot at 320K against a band that tops out near 271K:

Defect rate by line · this week (DPM)

Line 1above the learned band320K

Line 2inside the band197K

The learned band tops out near 271K — only Line 1 is above it this week. The band is learned per line and per product, so the two figures are each scored against their own history rather than against each other.

Which of the 19 PLC parameters on Line 1 and Line 2 are outside their learned band right now, and do they sit in one stage or across the line?

Checked the latest hourly median for all 19 PLC parameters on both lines against each one's learned band, then grouped whatever came back out of band by process stage to see whether it clusters.

Four are out of band, and all four are in the wash stage on Line 1 — the curing and cooling stages are clean, and Line 2 is clean end to end:

Parameters outside their learned band · latest hourly median

Wash tank 3 temperature · Line 156 °C vs 39 °C band

Wash tank 5 temperature · Line 147 °C vs 39 °C band

Wash tank 4 temperature · Line 145 °C vs 39 °C band

Wash tank 2 temperature · Line 143 °C vs 39 °C band

All 19 parameters · Line 2in band

Four out of nineteen, and the four are adjacent tanks on one circuit of one line. That is a stage answer rather than a parameter answer — you check the wash circuit's heat supply, not four independent controllers.

Does Yield % agree with that, and how much of the estate's quality output is each line actually carrying?

Split this week's yield — quality output over total output — across the two lines, and weighted each line by the share of quality output it contributed so the estate figure is the real average rather than the mean of two ratios.

Yield is healthy on both lines, which is why nothing here reads as a crisis — but Line 1 is the one pulling the estate average down, and it carries nearly half the output while doing it:

Share of this week's quality output by line

92.4%estate yield

Line 2 — 94.2% yieldin band on all 19 parameters53%

Line 1 — 90.8% yieldfour wash-stage parameters out of band47%

The two lines split output almost evenly, but Line 2 converts about 3.4 points more of it into quality product. On a near-even split that spread is the whole gap between the estate's 92.4% and what it would run at if both lines behaved like Line 2.