Anomaly detection on NASA Bearing Dataset

1. Aim was to detect the anomaly in the incipient stage of damage and also report the kind of defect if needed.
   
2. Due to the absence of labelled data and presence of data from essentially one class, predictive modelling techique, like LSTM was used.
   
3. LSTM models the normal behaviour of the bearings, and the prediction errors are subsequently used to identify abnormal behaviour.
   
4. A level 0 warning is triggered when the errors cross the calculated upper/lower bounds. Continuous level 0 warnings lead to a level 1 warning (which means that the bearing is broken, and the machinery should be stopped to prevent further damage).
   

Detecting fault in a motor with broken rotor bars, with data recorded at different loads

1. Healthy and unhealthy data were both recorded at 4 load conditions: no load, minimum load, medium load, full load.
   
2. The vibration data was loaded and used in further steps for model building.
   
3. Features were extracted and selected and LSTM and One-Class SVM models were built. Both of them gave comparable results.