search

Frakert / Predictive-Maintenance-Membrane-pumpm

In this project i will attempt to make a Predictive maintenance model for a membrane pump at CPP.
0 stars 0 forks source link

False positives on new data #15

Open Frakert opened 1 year ago

Frakert commented 1 year ago

Post training both models have been fitted on all trainingsdata and new data (10th of April 2022 until 5th of june 2023). This prediction gives an alarming amount of positives. Part of this presumably is caused by the mechanical seal troubles, but still. Investigate if possible.

Predictions_2
Predictions_2
DateTime BatchName Failure within 20 Batches
30-5-2023 17:17:00 YD112315004 1
26-5-2023 13:41:00 KD122314605 1
26-5-2023 07:09:00 KD112314602 1
25-5-2023 20:24:00 YD112314509 1
25-5-2023 10:25:00 YD112314505 1
25-5-2023 05:36:00 YD112314503 1
25-5-2023 03:28:00 YD112314502 1
24-5-2023 11:27:00 MD112314401 1
17-5-2023 12:14:00 MD112313702 1
16-5-2023 16:13:00 MD112313601 1
4-5-2023 04:43:00 MD112312403 1
4-5-2023 03:09:00 MD112312402 1
3-5-2023 22:35:00 MD112312304 1
24-4-2023 23:09:00 KD112311405 1
20-4-2023 08:16:00 KD112311003 1
20-4-2023 04:01:00 KD112311001 1
4-4-2023 22:29:00 YP7012309406 1
23-3-2023 12:27:00 KD112308202 1
9-3-2023 00:51:00 KD112306704 1
8-3-2023 04:31:00 YD112306701 1
7-3-2023 23:51:00 MD112306607 1
7-3-2023 21:29:00 MD112306606 1
7-3-2023 16:23:00 MD112306604 1
Frakert commented 1 year ago

Seal problems started 19-4 according to the SAP notification. Even if that would be so, that would mean 7 False positives before then. (Seal problems could be causing false postives by for example creating unusually short batches creating weird TSfresh features.)

Frakert commented 1 year ago

I currently see a few possible causes:

Will need to investigate these further if the membrane model is to be deployed. NOTE: the ball model only had 1 false positive in this period, and thus performs significantly better.

Frakert commented 1 year ago

Or, the membrane is already ruptured.