How to create a degradation feature profile from available sensor data?

I have gathered sensor data. Now I want to derive degradation feature profile for my data to perform RUL estimation using exponential degradation model. The main reason behind using this model is that I have threshold values, but no run-to-failure data.

[RAGHUVEER RAJESH]

https://in.mathworks.com/matlabcentral/answers/2105611-how-to-generate-synthetic-run-to-failure-data-in-simulink

Answer 1

[Aman]

Hi Raghuveer,

As per my understanding, you want to create the feature vector/degradation feature profile using the sensor data that you have collected.

The sensor data that you have collected is a raw form of data that you need to process before giving to any model for RuL prediction. There are four major steps that you need to perform:

• Remove any null value that you have in the data.
• Resample the data to an appropriate sampling rate, which should be in sync with your controller.
• Identify and remove the outliers from the data.
• Select only appropriate sensor data, i.e., to perform feature selection, as every feature is not necessary for model creation.

Apart from these four steps, you can optionally transform your data before feeding it to the model for training.

Once you have performed all these steps, your feature vector will be ready to use for model training.

Hope it helps to proceed ahead with your workflow!

[RAGHUVEER RAJESH]

Hello,

Thank you very much for your answer. Although it doesnt answer my question, it was helpful. My main question was how to dereive a degradation profile without knowing the fialure dynamics of the component. Just by the knowledge of mathematical equation or behaviour of the component(kind of grey box understanding), can we derieve degradation profile. Also, the sensor data is available, but there is no failure happening in actual. So the sensor data doesnt capture the failure states.