Maintenances and Failures

[agiannoul]

Hi,

Thank you for the Dataset, a valuable contribution, since real industrial data are rare.

I am confused regarding the consecutive rows with "m_id" and "m_subid" having the same value (for example: in period 2018-12-09 09:12:12 - 2018-12-09 23:00:00 "m_id"=M9 and "m_subid"=MS6). In the description states that, in order for a maintenance to take place the machine has to stop.

So the questions here is:

1. These data (or similar ones) refer to a maintenance that last for all this period?

2. In case we try to predict the failure that lead to this maintenance, can we assume that the failure took place on the beginning of that period, or in the end, and thus this period just depict a "deviating" period?

Finally,
In the work of [1], states that we should use a time horizon of 6 hours to produce succesful warnings (this time is correct to be measured from the first instance of each maintenance period or the last ?

[1] Ignacio Aguilera-Martos, Marta García-Barzana, Diego García-Gil, Jacinto Carrasco, David López, Julián Luengo, and Francisco Herrera. 2023. Multi-step histogram based outlier scores for unsupervised anomaly detection: ArcelorMittal engineering dataset case of study. Neurocomput. 544, C (Aug 2023). https://doi.org/10.1016/j.neucom.2023.126228