You do not have to think too much about difficult things. Regarding the failure prediction method of the machine, you suffer from difficulties too much.

The Artificial intelligence and a deep learning are useless in your case. Those techniques require huge amounts of bad data and good data. Bad data? There is not a lot of bad data! 

Please do not be deceived by the trend for money making which  is called Mr.A I.


The most important thing is quality control.
 It is so-called the QC.The QC uses statistics abundantly. 

I invented tools to measure. I observe the data and discover "the beginning of machine failure".

 Let 's make use of the wisdom of our predecessors. 

Let 's cherish the basic!