PHMGNNBenchmark

The emerging graph neural networks for intelligent fault diagnostics and prognostics: A guideline and a benchmark study

Implementation of the paper:

Paper:

@article{PHMGNNBenchmark,
  title={The emerging graph neural networks for intelligent fault diagnostics and prognostics: A guideline and a benchmark study},
  author = {Tianfu Li and Zheng Zhou and Sinan Li and Chuang Sun and Ruqiang Yan and Xuefeng Chen},
  journal={Mechanical Systems and Signal Processing},
  volume = {168},
  pages = {108653},
  year = {2022},
  issn = {0888-3270},
  doi = {https://doi.org/10.1016/j.ymssp.2021.108653},
  url = {https://www.sciencedirect.com/science/article/pii/S0888327021009791},
}

Requirements

• Python 3.8 or newer
• torch-geometric 1.6.1
• pytorch 1.6.0
• pandas 1.0.5
• numpy 1.18.5

Guide

We provide a novel intelligent fault diagnostics and prognostics framework based on GNNs. The framework consists of two branches, that is, the node-level fault diagnostics architecture and graph-level fault diagnostics or regression architecture. In node-level fault diagnosis, each node of a graph is considered as a sample, while the entire graph is considered as a sample in graph-level fault diagnosis.
In this code library, we provide three graph constrcution methods (KnnGraph, RadiusGraph, and PathGraph), and two different input types (Frequency domain and time domain). Besides, seven GNNs and four graph pooling methods are implemented.

Pakages

• datasets contians the data load method for different dataset
• model contians the implemented model for nodel-level task
• model2 contians the implemented model for graph-level rask

Run the code

For fault diagnostic

• Node level fault daignostic
  python ./train_graph_diagnosis.py --model_name GCN --data_name XJTUGearboxRadius --data_dir ./data/XJTUGearbox/XJTUGearboxRadius.pkl --Input_type TD --task Node --checkpoint_dir ./checkpoint
• Graph level fault daignostic
  python ./train_graph_diagnosis.py --model_name GCN --data_name XJTUGearboxRadius --data_dir ./data/XJTUGearbox --Input_type TD --task Graph --pooltype EdgePool --checkpoint_dir ./checkpoint

  For prognostic

  python ./train_graph_prognosis.py --model_name GCN --pooltype EdgePool --data_name CMAPSS_graph --data_file FD001 --data_dir ./data/CMAPSS/ --checkpoint_dir ./checkpoint/FD001

  The data for runing the demo

  In order to facilitate your implementation, we give some processed data here for node level-fault diagnosis and graph-level prognosis Data for demo.

Datasets

Fault diagnostic datasets

Self-collected datasets

• XJTUSpurgear Dataset
• XJTUGearbox Dataset

  Open source datasets

• CWRU Bearing Dataset
• MFPT Bearing Dataset
• PU Bearing Dataset
• SEU Gearbox Dataset

  Prognostic datasets

• CMAPSS dataset
• PHM2010 dataset

Note

This code library is run under the windows operating system. If you run under the linux operating system, you need to delete the ‘/tmp’ before the path in the dataset to avoid path errors.

Related works

• Li et al. Multireceptive Field Graph Convolutional Networks for Machine Fault Diagnosis, IEEE Transactions on Industrial Electronics
• Li et al. Domain Adversarial Graph Convolutional Network for Fault Diagnosis Under Variable Working Conditions, IEEE Transactions on Instrumentation and Measurement
• Li et al. Hierarchical attention graph convolutional network to fuse multi-sensor signals for remaining useful life prediction, Reliability Engineering & System Safety