Prognostic and Remaining Useful Life Prediction of Bearing

This is the project of hybrid prediction based on deep learning technology.

Key technology

• Sparse condition encoding method
• Shared features modeling with fault prognostic and RUL
• Data reinforcement

TODO

• Increase generality
• Classification and regression on XJTU bearing dataset
• Transfer learning appliance
• Digital twins by Deep Conditional Generative Adversarial Neural Network(DCGAN)

Achievements

Type	Value on Train set	Value on Test set
Classification Accuracy	~99.86%	~65%

Classification result

You can find them in SVG file:

• classification_on_train_data.svg
• classification_on_test_data.svg

Since they are not intuitive, we didn't publish them on README.md file.

Confusion matrix

• On train data
  
• On test data
  

Train online

AI studio seems to support only PaddlePaddle framework which is very poor and unfriendly in supporting 1D signal. It will waste a lot of time, so we don't suggest you to use that even they are free.

You may use Baidu AI Studio to train this model by yourself. [HERE] is the project hosted on AI studio.

Current difficulty

• Follow Fast-RCNN architecture to design brand-new model

LICENCE

Project is licensed MIT, and XJTU Bearing data is copyrighted by Biao Wang of Xi'an Jiaotong University