PHM prediction by residual conventional neural network

This is a program for predicting RUL estimation for a high-speed CNC milling machine cutters, you can download data from https://www.phmsociety.org/competition/phm/10 and run it to predict RUL.

Dataset download

The original PHM10 contest data file seems depreciated, we provide a Google drive mirror for researchers. You can find all files at https://arch-blog.kidozh.com/projects/keras_detect_tool_wear/download_database.html. If you find them out of date, please contact me by email and I will send you the updated ones if possible.

Please note that the copyright is reserved by PHM society.

Notice

This project's page is HERE, you may find something interesting about our work.

RUL estimation is prediction for engine by the same method, the accuracy is ideal.

This program are under development.

recent achivement

By using wavelet transformation, loss (MSE) has been down to 9.0.

According to PHM Error method, our CNN model's error is 393, way more better than the best score(5500) at 2010 leaderboard.

visualization

Topology Data Analysis is implemented for visualizing model's weights, t-SNE method for hierarchy feature.

dependency

• Tensorflow / Theano / CNTK
• Keras

Introduction

A beamer in demonstration_of_work is to demonstrate what we have done in summer holidays. It is writen in Simplified Chinese.

Please feel free to fork it.

Licence

MIT LICENSE