Multimodal time-series contrastive generative network-based data augmentation algorithm
SHANG Rou
DONG Hong-li
WANG Chuang
ZHOU Guo-qiang
GUAN Chuang
YAN Tian-hong
Abstract：In this paper,a Markov chain-based multimodal time-series contrastive generative network(TCGN)is pro-posed to tackle the issues of small sample and class imbalance for industrial fault diagnosis.Firstly,a time-series trend consistency loss(TTC)is designed to enhance the similarity of the time-evolving properties between the real and synthetic data,which helps to improve the reality of the synthetic temporal structure.Subsequently,a class-aware contrastive loss(CAC)is proposed to align the class-conditional distributions between the real and synthetic datasets,which facilitates the formation of effective and proper decision boundaries.Furthermore,a Markov chain-based multimodal switching strategy is introduced in this paper,which enables the TCGN algorithm to perform adaptive switching optimization between the four modes of generation,depiction,exploration,and convergence,thus better maintaining the dynamic balance of differ-ent tasks.Finally,the proposed TCGN algorithm is applied to pipeline fault diagnosis.Experimental results show that the TCGN algorithm outperforms some state-of-the-art generation algorithms in terms of visual evaluation and quantitative metrics,and significantly improves fault diagnosis accuracy.
Keywords：pipeline fault diagnosisclass imbalancetime seriesdata augmentationMarkov chainmulti-task learning
Publication Date：2025-04-30
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：11( 805-815 )