Metal and process

A pot runs cold for two days and the first anyone knows is the month end

A PI historian, ABB 800xA operator stations, a Siemens S7 on utilities, a spreadsheet a process engineer keeps, and the lab system. Five naming schemes, two languages, five spellings of degrees Celsius.

ANODE BAKE4 furnacesPOTLINES22 pots, 1 line coldCASTHOUSE3 linesFUME TREAT3 fansUTILITIESair, cooling water, tanksPI historian · ABB 800xA · Siemens S7 · a spreadsheet · the lab

The works as five systems see it. Green dots are instruments reading, red ones are alive in the tag list and dead in the field. The amber blocks are where something is happening.

What actually goes wrong

A potline is a row of cells that each want to sit in a band a few degrees wide. One of them drifting out of that band is not dramatic. It is a slow loss of current efficiency that shows up as metal you did not make, and the first place it becomes undeniable is a month end report.

The reason it runs that long is not that nobody is watching. It is that watching means reading 31TI0121.PV on one screen, AL1.ELE.POT003.BATH_TEMP on another, and knowing that the second one is the same sensor. Somebody at the works knows. That knowledge is in a person, on shift, and it does not scale to twenty-two cells and five systems.

What Auge does here is unglamorous. It reads the tags from all five systems into one model, settles which of them are the same physical point, decodes the ones still arriving as unscaled current loops, and lists every instrument that died and kept publishing a flat line. Then it scores each point against the band it held in its own first two days, so a cell drifting out of its own normal is visible on the day it starts rather than at the month end.

What it looks like in the data

Both of these are drawn from the compiler's own decoded values, not from an illustration. The line draws itself as you arrive.

scrubber-001 · vibrations7:DB10.DBD12day 1day 142.209 to 7.169 mm/s

fan vibration rising steadily over the second week

scrubber-001 · currents7:DB10.DBD16day 1day 14192 to 235.5 %

motor current follows the vibration rise

How far we have got

the label on this story

modelled

Our aluminium works is one we built, so the numbers are honest about the compiler and say nothing about yours. Over fifty plants it had never seen: 0.876 (0.869 to 0.883) of tags resolved, 0.985 (0.982 to 0.988) precision where it claims confidence.

read next

The numbers, including the ones that missed

The aluminium works is scored in the table on the front page, including the two criteria that missed.