Counterfactual Faithful Data Generation Based on Disentangled Representation for Compound Fault Diagnosis of Rolling Bearings
ID:65 View Protection:PUBLIC Updated Time:2022-12-22 00:54:43 Hits:316 Poster Presentation

Start Time:Pending (Asia/Shanghai)


Session:[No Session] » [No Session Block]

Recently, deep learning and human-out-of-the-loop methods enjoy their prosperous applications in mechanical fault diagnosis. Nonetheless, the None-IID(independent and identically distributed) issue radicated in acquired data severely limits the stability and accuracy of compound fault diagnosis of rolling bearings. This paper proposes a sample augmentation method for generating simulated signals based on the concept of counterfactuals. Firstly, disentangled representations and counterfactual faithful theory are applied to classify the original signal into two categories of properties. One is the fault semantics encoded from the original vibration signal. And the other is the sample attribute encoded by the encoder of Variational Autoencoders (VAEs). Secondly, the counterfactual faithful pseudo-samples are conjured through the Generative Adversarial Network(GAN) using the seeds of the “factual” sample attributes and “counterfactual” fault semantics to compensate for the drawback of distribution shift. Finally, the original samples and pseudo-samples are used as the CNN classifier dataset to realize bearing fault diagnosis. Experiments show that this method can generate counterfactual signals that are highly consistent with the original data distribution and can achieve better classification accuracy after balancing the dataset.
rolling bearing;fault diagnosis;counterfactual faithfulness;structural causal model;VAEGAN
Qiang Zhu
student Hefei University of Technology

2016.9-2020.6 Bachelor Degree, School of Mechanical Engineering, Anhui University of Technology.
2020.9-23.6 Master Degree, School of Mechanical Engineering, Hefei University of Technology.

Comment submit
Verification code Change another
All comments

Important Dates

15th August 2022 25th September 2022 - Manuscript Submission
15th October 2022 - Acceptance Notification
1st November 2022 - Camera Ready Submission
1st November 2022 10th November 2022Early Bird Registration

Contact Us


Scan the QR  code and join the

WeChat Group