Vibration signals collected in real industrial environments are usually limited and unlabeled. In this case, fault diagnosis methods based on deep learning tend to perform poorly. Previous work mainly used the unlabeled data of the same diagnostic object to improve the diagnostic accuracy, but it did not make full use of the easily available unlabeled signals from different sources. In this study, a signal momentum contrast for unsupervised representation learning (SMoCo) based on the contrastive learning algorithm—momentum contrast for unsupervised visual representation Learning (MoCo)—is proposed. It can learn how to automatically extract fault features from unlabeled data collected from different diagnostic objects and then transfer this ability to target diagnostic tasks. On the structure, SMoCo increases the stability by adding batch normalization to the multilayer perceptron (MLP) layer of MoCo and increases the flexibility by adding a predictor to the query network. Using the data augmentation method, SMoCo performs feature extraction on vibration signals from both time and frequency domains, which is called signal multimodal learning (SML). It has been proved by experiments that after pre-training with artificially injected fault bearing data, SMoCo can learn a powerful and robust feature extractor, which can greatly improve the accuracy no matter the target diagnostic data with different working conditions, different failure modes, or even different types of equipment from the pre-training dataset. When faced with the target diagnosis task, SMoCo can achieve accuracy far better than other representative methods in only a very short time, and its excellent robustness regarding the amount of data in both the unlabeled pre-training dataset and the target diagnosis dataset as well as the strong noise demonstrates its great potential and superiority in fault diagnosis.
To address the limited data problem in real-world fault diagnosis, previous studies have primarily focused on semi-supervised learning and transfer learning methods. However, these approaches often struggle to obtain the necessary data, failing to fully leverage the potential of easily obtainable unlabeled data from other devices. In light of this, this paper proposes a novel network architecture, named Signal Bootstrap Your Own Latent (SBYOL), which utilizes unlabeled vibration signals to address the challenging issues of variable working conditions, strong noise, and limited data in rotating machinery fault diagnosis. The architecture consists of a self-supervised pre-training-based fault feature recognition network and a diagnosis network based on knowledge transfer. The fault feature recognition network uses ResNet-18 as the backbone network for self-supervised pre-training and transfers the trained fault feature extractor to the target diagnostic object. Additionally, a unique vibration signal data augmentation technique, time–frequency signal transformation (TFST), is proposed specifically for rotating machinery fault diagnosis, which addresses the key task of contrastive learning and achieves high-precision fault diagnosis with very few labeled samples. Experimental results demonstrate that the proposed diagnostic model outperforms other methods in both extremely limited sample and strong noise scenarios and can transfer unlabeled data utilization between similar and even different device types.