Multi-Domain Time-Frequency Fusion Feature Contrastive Learning for Machinery Fault Diagnosis-- Wide Kernel Time-Frequency Fusion （WTFF）

The source code is for the following paper :

Wei Y, Wang K. Multi-Domain Time-Frequency Fusion Feature Contrastive Learning for Machinery Fault Diagnosis[J]. IEEE Signal Processing Letters, 2025 (99): 1-5.（https://ieeexplore.ieee.org/document/10910177）

If you find this code is useful and helpful to your work, please cite our paper in your research work. Thanks.

If there are any questions about source code, please do not hesitate to contact Yang Wei([email protected]) and me ([email protected]). Special thanks to Rui Luo for helping to arrange and re-implement the code.

How to use the code

Running environment:

The proposed methods are implemented in Python 3.9.7 with PyTorch 1.12.0 framework on a desktop computer equipped with an Intel i9 12900k CPU and an NVIDIA RTX 3090 GPU.

Dataset used in this paper:

1. PU
2. JNU

How to reproduce the experimental results of Machinery Fault Diagnosis.

1. Download the required datasets: According to the provided dataset addresses, download the required datasets for the experiment and store them in the respective subfolders under the dataset folder.

2. Configure Environment： Find the requirements.txt file in WTFF folder and configure Environment：pip install -r requirements.txt

3. Configurable arguments: Find the run_main.py file in WTFF folder，According to the experiment you are going to conduct, set the parameters in the run_main.py file.

   For example：If you wish to use a different dataset for training and testing the model, you can achieve this by modifying the file path specified under "data" .You can select the model's training mode by modifying the value of "train-mode" in the parameters, which includes 'finetune' representing fine-tuning the model, 'time-frequency' indicating both pre-training and fine-tuning the model, and 'evaluate' signifying testing the model.

   In brief, you can modify various aspects of the model including its training mode ，number of data loading threads, maximum training epochs, batch size, optimizer, learning rate, and other parameters by adjusting the values of the respective parameters in the configuration.

4. Train your backbone,run: python run_main.py

5. Evaluating results will automatically show after training.

Tips:

The experimental results included in the above "runs" files are re-calculated when submitting code, which may be a slight deviation from the results reported in our paper due to the effects of randomness of choosing training samples.You can conduct multiple experiments to obtain results that are the same as or similar to those reported in the paper.