Community Tip - Need to share some code when posting a question or reply? Make sure to use the "Insert code sample" menu option. Learn more! X
I want to create an analytics model which can detect two different defects and the model should be able to distinguish between two different defects. The examples show datasets with low-grease fault, I believe the vibration data was collected for different bands for good condition and low-grease condition. If I want to identify two different faults, would I need two columns with defect x and y and run the physical asset in normal, defect x alone, defect y alone. Do I need to also data when defect x and y are induced at the same time? To use the prediction score, do I need separate models for each faults since the goals are different?
Also, are there any more examples of predictive maintenance and is there a theory manual to understand the different algorithms that are available in Thingworx
Solved! Go to Solution.
You first train a model with historic data that have rows with error set to 0 (no issue) and some with error set to 1 and other with error set to 2.
Then once the model is created, you receive new row that you score and it will predict if error should be 0 1 or 2.
Note this is in the case where you have one goal "error" that can take several values.
If you take the other approach to have different models, one for error1 and one for errro2, then you will need to score twice. the first score will tell if error1 is likely to occur or not and the second will predict if error2 is likely to occur or not.
Hi @cnuga
Here are some input that hopefully will help.
The 2 defects scenario can be approached in different ways depending on the situation. For example:
- if the 2 different defects can be resolved as different error code of sort, then you could build the dataset with only one feature errorcode, taking the value 0 if no defect or 1 if defect 1 or 2 if defect 2.
Then only one model can be used to score the categorical goal.
- if the 2 defects are very different in nature, then you will have to create 2 different models with 2 different goals.
In term of data collection you will need to have historic data that shows normal behaviour and instances of both defect - this could be the same dataset or 2 different ones.
When training, simply chose the right defect as goal, and possibly exclude the other one depending on the case.
For scoring, if you can use option 1, you only score your entry on that single categorical model. If you need to use option 2, then you will have to execute 2 scoring job one against each model.
Hope this helps
Kind regards
Christophe
Thanks for the information, very helpful. Once I collect my data with 0, 1, 2 codes and train the model or models, when defect 1 or 2 occurs does the scoring predict the right defect? i.e., how do we get the model to distinguish between the two defects
You first train a model with historic data that have rows with error set to 0 (no issue) and some with error set to 1 and other with error set to 2.
Then once the model is created, you receive new row that you score and it will predict if error should be 0 1 or 2.
Note this is in the case where you have one goal "error" that can take several values.
If you take the other approach to have different models, one for error1 and one for errro2, then you will need to score twice. the first score will tell if error1 is likely to occur or not and the second will predict if error2 is likely to occur or not.