Dear all,
I am trying to collect vibration data from two similar motors which one of them is healthy and new but another one has a bearing problem. Since I do not have any historical data and degradation path of the motors. what can I do with this data to predict the bearing failure? Can I implement RUL(remaining useful life) or should I set a threshold for anomaly detection? I will be thankful if someone could help me since I do not have much experience on machine learning algorithms.
We are currently carrying out a project in this area it would be interesting if we could collaborate and share information! Kindly contact me on [email protected].
It might be interested to start by analyzing the frequency response of the healthy and failing motor before proceeding to ML algorithms. Maybe a simple bandpass filter with threshold detection might do the trick!
Mehdi Rahmanian I have had some experience with both. It would be helpful if you could give some more context of the problem. Are you trying to predict failure in real-time? Or, is the objective only to find pattern that lead to failure and apply it all over. For lifetime analysis, you could try survival regression model if you have data across multiple motors, otherwise it just reduces to normal regression that you model with exponential distribution family. This would perhaps help you identify what is the average age of motor when it fails and could help you reset warranty - threshold (totally depends on your context) ? For real-time prediction of failure, you have multiple approaches like LSTM based autoencoder that you can predict anomaly in a short time period.
- Badges
- Science topic