predictive maintenance

Thiru

hi all,


1) referring the process for "predictive maintenance" in rapidminer & the video in https://rapidminer.com/resource/data-science-predictive-maintenance/
https://docs.rapidminer.com/9.4/server/use/web-services/predictive-maintenance.html

 in the given data set,   the machine is characterised by the label : failure - yes/no.
Does it mean the machines of respective id is under failure - yes or no  condition,  for the given sensor data?  how does that label value is decided?

2.  In my case,   If I can build a machine learning classification model for a 'single' machine with columns  as sensor data ( eg. no of columns =5 , means no. of sensor = 5) with many rows of data.  This classification model can classify the fault as Yes or No.  
can I extend this 'classification'  model into ' prediction' of failure?  
thanks

regds
thiru

hbajpai

Hi @Thiru ,

On,

a. Essentially if you have a classification model, it can be used to predict when the label is missing. For example, in this case we might not know the machine will fail up till a certain time x after the current time. Assuming we built an accurate classification model, we can use it to predict failure in real-time and take actions, thus avoiding line shutdowns and reducing downtimes.

b. Determine influence factors operator was used for feature importance amongst the 25 sensors and optimize parameters grid was used to find out the value of k let's say in 1 - 100 range that would result in best accuracy and least classification error. 

Let me know if you need further clarification.

hbajpai

Hi @Thiru,

1. I believe the data is mocked up and the file I have on my end consists of 136 sample examples wherein the column "Failure" indicates whether the machine has failed or not. The data has 25 sensor values that act as attributes that help us to understand whether the machine will fail or not. In this use case, a classification model was used to learn and predict for a new set of 25 sensor values whether the machine will fail or not.

2. Can you elaborate more on the second question? I am not sure about the data description to answer whether machine ID is should be used as an attribute or we can generalize the trend to develop a model that can be used for rescoring on all machines. Maybe @JeffChowaniec can elaborate more on this. 

Thiru

hi @hbajpai ,

thanks for your reply.  

a.  My 2nd question is about:

I build a "classification" model for a machine to "classify"  whether it is failure - Yes or No,  eg. Lets say using KNN. 
 This was done using inputs of 5 sensors.  these 5 sensors were columns while building the classification model. Im able to  classify the failure (it is just classification based on given data , not prediction) with these 5 columns. My question is:     is there any way to convert an  already  build  classification model into a model suitable for predictive maintenance?

b.  Referring your answer for my 1st qn: 
   as per the example set,  25 sensor values act as attributes and labelled column says whether the machine has failed or not.  My understanding is : it is only the  normal data set typically used for any classification problem.  Looks like the operator - " determine influence factors"  & " optimize parameters (grid)" contributes in using the sample set for predictive maintenance.  ( correct me If Im wrong).


thanks & regards
thiru

Thiru

hi @hbajpai ,

thanks for your reply.  

a.  My 2nd question is about:

I build a "classification" model for a machine to "classify"  whether it is failure - Yes or No,  eg. Lets say using KNN. 
 This was done using inputs of 5 sensors.  these 5 sensors were columns while building the classification model. Im able to  classify the failure (it is just classification based on given data , not prediction) with these 5 columns. My question is:     is there any way to convert an  already build  classification model into a model suitable  for predictive maintenance?

b.  Referring your answer for my 1st qn: 
   as per the example set,  25 sensor values act as attributes and labelled column says whether the machine has failed or not.  My understanding is : it is only the  normal data set typically used for any classification problem.  Looks like the operator - " determine influence factors"  & " optimize parameters (grid)" contributes in using the sample set for predictive maintenance.  ( correct me If Im wrong).


thanks & regards
thiru