Automotive Predictive Maintenance modeling

sgenzer

Hello.  I would like to share with this SIG a sample process in RapidMiner that will model automotive predictive maintenance.  It may be useful for anyone working in the manufacturing space.  Thank you @bhupendra_patil and colleagues for sharing.

 

Scott

 

 

lionelderkrikor

Hi,

 

I'm interested in this subject of predictive maintenance but

I can't import the .rmp process in RM : When I try to import this file, there is the following message in the log view : 

"PM WARNING: While updating process editors: java.lang.reflect.InvocationTargetException"

and nothing happens.

 

What can I do ?

 

Thanks,

 

Regards,

 

Lionel

sgenzer

hi @lionelderkrikor- that's odd.  Did you unzip the file and place that folder in your .RapidMiner repository?

lionelderkrikor

Hi,

 

That's I made and it worked. So I can study this process.

(I was writing this reply when I received your message......)

 

Thanks you for your feedback,

 

Regards,

 

Lionel

 

 

spacios

Hello, 

 

I tried the model. It works very well but I would like to have an equation predicting the failure of the part. Like for example if the oxygen sensor 3 is 0.15 then the part automatically fails. 

How can I do that ?

Thank you for your help

Selina

sgenzer

hello @spacios - so the way these models work is not necessarily with a "formula" but rather with an algorithm.  In this model it's nearest-neighbor (k-NN).  You put new, unlabeled data in the Apply Model operator to get a prediction.

 

 

If you want to see a traditional formula, you can always do regression or something similar.

 

Scott

spacios

Hello,

 

Thank you @sgenzer, it worked with apply model but I dont see the algorithm or the correlations between the failure and the part and like one or multiple sensors. I am very new on this software, but  I tried polynomial regression and i dont have an equation for the failure of the part. Here are what I did: 

 

 

Did I do something wrong ? thank you for your help!

 

Regards, 

 

Selina

 

sgenzer

Hello @spacios - oh that's an easy one.  Just connect the model output for the polynomial regression to a results nub:

 

 

Scott

spacios

Hello,

Thank you @sgenzer, I did what you said, it works fine.

 

I have another question where do I visualize the graph and the equation ? 

I only have this right now :

Thank you for your help, 

 

Selina

sgenzer

hello @spacios - so you cannot "graph" equations in RapidMiner like you would in Mathematica or similar because it's not that kind of software.  To the best of my knowledge, if you wanted to graph a function you would create x and y attributes using those functions and then graph those attributes as a scatter plot etc...

 

Scott

 

 

spacios

Hello,

 

Thank you @sgenzer. 

I have additional questions : 

So I ploted some points shown in the picture. 

What does the numbers in y axis refer to ? What is the unit of PID 05 ?

What is exactly a blue dot ? 

How can I get the name of the part when I click on the blue dot ?

Does this graph mean that when the PID 05 is a precise number there is either  a prediction of failure or not ?

 

Thank you so much !! 

 

Selina

sgenzer

Hello @spacios - so those are all good questions that I did not know the answers to.  However if you just use Google and type "automotive sensor PID" you arrive at a nice Wikipedia page that explains what PIDs are.    

 

So the "blue dots" are the value of PID5 for each row in your data set vs whether or not that row was classified as no or yes.  That's all.

 

Scott

 

spacios

hello,

 

Thank you @sgenzer. Do you know if there is a way to show the PID 05 data set linked to prediction (failure) yes or no ?

 

Selina

sgenzer

So to me “linked” means “correlated” or some predictive relationship. So that’s the whole purpose of the predictive model, right?