Digital twin-assisted self-supervised contrastive learning: A novel framework for electromechanical equipment fault...

In modern industry, electromechanical equipment plays a crucial role in ensuring the efficient, stable, and safe operation of production systems. However, fault diagnosis of electromechanical equipment in real industrial environments faces challenges such as a scarcity of labeled data, high cost of fault reproduction, and insufficient generalization across varying operating conditions. In response to these challenges, this paper proposes a novel framework using digital twin-assisted self-supervised contrastive learning for electromechanical equipment fault diagnosis. First, a five-dimensional model for digital twin fault diagnosis of electromechanical equipment (DTEE) is constructed. This model generates simulated data via dynamic simulations of localized faults combined with ten data augmentation methods, alleviating the problem of insufficient fault data and class imbalance. Second, a self-supervised time-frequency contrastive learning model based on digital twin (DT-SSTFCL) is designed to achieve complementary learning of time domain and frequency domain features by using a dual-stream transformer encoder. By learning cross-modal latent relationships through a cross-correlation loss matrix (CCLM), the model better captures intrinsic time-frequency consistency. Furthermore, the smoothed wavelet kernel (SWK) optimization method is proposed, where Laplace wavelet parameters are used as initialization weights of the convolution kernel, and combined with the translation, scaling and smoothing factors to enhance the model's understanding of the time-frequency features. The experiments are conducted on rolling bearings as an example. The results of three experimental cases show that our method significantly outperforms existing self-supervised and supervised learning methods. This provides a solution for electromechanical equipment fault diagnosis with limited labeled data and across different scenarios.