Predective maintenance modeling?



  • I’m working on Predictive maintenance exercise. the question I have to answer is "When the asset (machine) will fail?

  • Available data: History of failures, history of services (maintenance activities) done before, history of set of sensors measures.

  • I’m trying to use the folloeing as the independent variable: cycle (day/date) which is everyday in the asset life-cycle, the measures of the sensors.

  • the dependent vars are: Running (Yes/No), in-service (Yes/No), broken (Yes/No).

  • I work using ML algorithm for predicting multi-class classification, i.e. a point in the future, what should be the status of the machine (running, in-service, broken).

  • Well, my issue is, how to predict a future point of time, I know the cycle (could be in 20 days time) but I don’t have the sensor measures by this time, so I’m missing the most critical independent vars values?

My question is:

  • Is the way I modelled the problem right/wrong? what is the best way to model this problem?
  • I explored some modelling templates, but they depend on a variable call Remaining Useful Time - RUL). I really cannot integrate this parameter in the modelling, plus lacking the measures values in the future will remain as an issue.

My apology if I’ve fundamental misconception as I really think so?

Please advice on the problem modelling?