I have this situation. We are tracking the power consumption of an industrial machine and by looking at the power consumption (in watt) we're trying to predict whenever something will break resulting in the machine needing maintenance. The machine has 1 specific manufacturing proces where the powerusage will spike up for certain periods and come down again.
I personally am very new to machine learning and don't quite know how I should structure the dataset. I have seen companies split it per component in the machine or per fase in the production process. Currently I have this structure in mind.
I'm not sure if UNIX would be the right time unit or if I should just start counting up from 1 whenever the machine starts running. I'm also unsure about the productionfase part of the algorithm, since the new input data will only include time and kilowatt. Do you guys think I should switch to having it per divided per component, or just leave both options out and only look at time and kilowatt?
Other than that the labels I have in mind are LOW|NORMAL|HIGH for the powerconsumption of the machine.
Could someone verify some of my ideas or give me some tips to go in the right direction? Thanks:)