Help to Choose right ML Technique

Good evening all
I want guidance on my project which is related to condition based health monitoring of electrical motors.
Data from different IOT sensors are coming to database in different format. Which ML technique to apply to achieve following :-

  1. Detect If any parameter goes beyond permissible limit
  2. Predict when motor will fail in future based on present condition and due to which parameter.
  3. Suggest maintenance routine whenever due, basically motor itself tells when to do maintenance.

Please help, I’m very much confused.


As per your query, i’m putting my suggestion hope this will help you.
If you have previous data, like the all parameters at the time of some previous maintenance, Then you can opt for ML. And by doing some EDA on previous data-set you can easily find out the what should be the permissible limit.
otherwise, If you have not previous data,In this case you should finalize the permissible limit of parameters by your own intuition, and by writing a simple programme consisting of if-else you can do the first task easily.

and for the 2nd task you need some previous data including the failure time and after that you can use regression or tree-based algo to predict failure time.

Also the 3rd task required some previous data to decide maintenance schedule, you can do this by tree-based classification.

Good Evening Pankaj
Thanks for your prompt help, it surely provided me an insight about how to move forward. Few clarification required are:-

  1. This would be new project(im starting from the scratch), so far health of a motor was checked using test equipment and no digital database was maintained. Now, we are planning to move to IOT sensors so that we can further apply ML algos.
  2. My query is, if i run the machine from somedays with IOT sensors to gather data and train my model, would that be ok to apply your above mentioned technique.
  3. Also, the data from different sensors is in different format, how to standardise it.
  4. And as pere your opinion, what things i should keep in mind to work with IOT sensors.
  5. Also, can you help me with your mail id.

hello ankit,
here are the answer of your queries.
for query,
2. ya its fine, but it might take few months if motor is not in continuous use.
3.Not a big deal, If data is categorical it can be label easily in numeric form and if numeric
data on different scale you can standardize by converting to Z-score.
4.This time it is difficult for me to suggest anything without knowing what type of sensor you
are using for which purpose.One thing is that, must note down the repairing time this will act as label for training any model

© Copyright 2013-2019 Analytics Vidhya