Calibrated Adaptive Teacher for Domain Adaptive Intelligent Fault Diagnosis

Florent Forest1    Olga Fink1
1 Intelligent Maintenance and Operations Systems (IMOS), EPFL, Lausanne, Switzerland

Sensors, 2024

   

Source code for the implementation of the paper Calibrated Adaptive Teacher for Domain Adaptive Intelligent Fault Diagnosis.

In this paper, we propose a novel unsupervised domain adaptation method called Calibrated Adaptive Teacher (CAT), where we propose to calibrate the predictions of the teacher network on target samples throughout the self-training process, leveraging post hoc-calibration techniques.

Our proposed Calibrated Adaptive Teacher (CAT). The main novelty involves a post-hoc calibration of the teacher predictions in the target domain throughout the self-training process, improving the quality of pseudo-labels.

Requirements

• numpy
• tqdm
• sklearn
• scipy
• pytorch
• torchvision
• tensorboard

Basic usage

To train source-only models:

python train_base.py --help

To train single-model DA baselines:

python train_advanced.py --help

To train teacher-student models, including CAT:

python train_advanced_ts.py --help

In particular, to train the best-performaing CAT-TempScaling model on the Paderborn bearing dataset, the command is:

python train_advanced_ts.py \
  --data_name PU \
  --transfer_task "[[0],[1]]" \
  --data_dir /path/to/dataset \
  --domain_adversarial True \
  --adversarial_loss DA \
  --self_training True \
  --self_training_criterion confidence \
  --adaptive_confidence_threshold \
  --calibration TempScaling \
  --mcc_loss --sdat \
  --checkpoint_dir output/cat_tempscaling_output

Project structure

CAT/
├── datasets/: dataset loading classes
├── loss/: loss functions
├── models/: models
├── optim/: SAM optimizer
├── transcal/: confidence calibration techniques
├── utils/
    ├── calibration.py: ECE metric calculations
    ├── entropy_CDA.py: entropy-weighted CDAN loss
    ├── logger.py: base logger
    ├── mixup.py: mix-up techniques (not used in paper)
    ├── self_training.py: self-training strategies (pseudo-labeling)
    ├── train_utils_base.py: training loop for source-only models
    ├── train_utils_combines.py: training loop for single-model DA baselines
    └── train_utils_combines_ts.py: training loop for teacher-student DA models
├── train_advanced_ts.py: main script for training teacher-student models
├── train_advanced.py: main script for training single-model DA baselines
├── train_base.py: main script for training source-only models
└── train_net.py: main script for training and evaluation

Datasets

The following datasets are supported:

• CWRU Bearing Dataset
• PU Bearing Dataset
• PHM 2009
• SEU Gearbox Dataset
• JNU Bearing Dataset

Citation

If this work was useful to you, please cite our paper:

@article{forest_calibrated_2024,
    title = {Calibrated Adaptive Teacher for Domain Adaptive Intelligent Fault Diagnosis},
    author = {Forest, Florent and Fink, Olga},
    journal = {Sensors},
    year = {2024},
    volume = {24},
    number = {23},
    url = {https://www.mdpi.com/1424-8220/24/23/7539},
    doi = {10.3390/s24237539}
}

Acknowledgements

The code for this work is based on following repositories:

• UDTL   
• TransCal   
• SAM   

We would like to thank their original authors!