The rolling bearing is the key component of rotating machinery, and it is also a failure–prone component. The intelligent fault diagnosis method has been widely used to accurately diagnose bearing faults. However, in engineering practice, it is difficult to obtain sufficient sample data to train the intelligent diagnosis model. Therefore, in this article, a fusion diagnosis model CGAN-2-D-CNN that combines a conditional generative adversarial network (CGAN) and a two-dimensional convolutional neural network (2-D-CNN) is proposed for bearing fault diagnosis with small samples. Considering the problem of insufficient sample data, CGAN is used to learn the data distribution of real samples to generate new samples with similar data distribution by the confrontation training of the generator and discriminator. Then, 2-D preprocessing is conducted to convert the generated 1-D data into 2-D gray images. Finally, these images are input into the 2-D-CNN to extract the features and classify the bearing fault types. Two experimental cases are implemented to validate the effectiveness and feasibility of the proposed CGAN-2-D-CNN. The experimental results illustrate that the diagnosis accuracy of the proposed method used on the small sample data is close to that of the 2-D-CNN directly used on the enough original sample data whose size is equal to the expanded sample size. In addition, compared with the 1-D-CNN, support vector machine (SVM), and long short-term memory (LSTM) models, the 2-D-CNN model after 2-D preprocessing has the higher fault classification accuracy.

The key to intelligent fault diagnosis is to find relevant characteristics with the capability of representing different types of faults. However, the engineering problem is that a few simple empirical features (EFs) cannot obtain high classification accuracy, and complex feature engineering requires strong professional knowledge, which leads to limited applications on a general scale. In addition, intelligent feature extraction and classification methods without prior knowledge cannot guarantee that the model learned the general features used for classification, and its robustness and generalization are not strong when the objects with low-quality training data. Therefore, a fusion method of combining EFs and adaptive features extracted by a deep neural network is proposed. In this method, simple EFs that only need a few professional knowledge are adopted to realize general feature extraction and, hence, maintain the robustness of the model. A modified neural network structure (LiftingNet) is proposed to achieve the adaptive extraction of hidden features for specific objects, for realizing the high precision of bearing fault classification. In order to realize the fusing of EFs and adaptive features, XGBoost is utilized as the final classifier instead of common softmax. The feasibility and validity of the proposed method are verified by two data sets collected from motor bearings at stable work conditions. The experimental results show that the classification accuracy generated by the proposed method is improved. It also can maintain the robust performance on data sets even with various noises.

Class imbalance issue has been a major problem in mechanical fault detection, which exists when the number of instances presenting in a class is significantly fewer than that in another class. This article focuses on the problem of zero-shot fault detection of rolling bearing, which is the extreme case of class imbalance. Aiming at this problem, a two-stage zero-shot fault recognition method is proposed. First, inspired by the conditional generative adversarial network, a novel feature generating network which is composed of a feature extractor, a discriminator, and a generator is designed to capture the potential distribution of normal samples. Then, the generator will generate abundant pseudofault features by adding an additional sequence to the condition. Second, an improved deep neural network is trained with these synthetic pseudofault features as the classifier. Specially, a condition index is designed to represent different fault classes so that it can recognize the unseen fault samples. Finally, the effectiveness of the proposed method is verified by three datasets and a comparison method is also given to show the superiority. Results show that the feature generation network can effectively detect the typical faults even though the fault data are unavailable during training, which is practical for industrial application.

To reduce the economic losses caused by bearing failures and prevent safety accidents, it is necessary to develop an effective method to predict the remaining useful life (RUL) of the rolling bearing. However, the degradation inside the bearing is difficult to monitor in real-time. Meanwhile, external uncertainties significantly impact bearing degradation. Therefore, this paper proposes a new bearing RUL prediction method based on long-short term memory (LSTM) with uncertainty quantification. First, a fusion metric related to runtime (or degradation) is proposed to reflect the latent degradation process. Then, an improved dropout method based on nonparametric kernel density is developed to improve estimation accuracy of RUL. The PHM2012 dataset is adopted to verify the proposed method, and comparison results illustrate that the proposed prediction model can accurately obtain the point estimation and probability distribution of the bearing RUL.

The intelligent diagnosis model based on deep learning method can effectively and accurately diagnose the health state of bearings widely used in various mechanical equipment. However, in engineering practice, it is difficult to obtain sufficient labeled fault data, which would reduce the diagnosis performance. In order to solve the above problem, this paper proposes a self-attention metric learning based on multi-scale feature fusion to classify the bearing fault with few shot. The proposed few-shot intelligent diagnosis model mainly contains the feature extraction module and metric learning module. First, the vibration data need to be combined in pairs to form sample pairs, which are input into the feature extraction module based on multi-scale feature fusion. This input approach can increase the learning tasks so as to alleviate the problem of insufficient training samples. Then, the extracted features are concatenated and input into the metric learning module whose learning ability is improved by adding the self-attention network, and the input features are eventually converted into the similarity. Finally, the fault type can be identified according to the maximum similarity value between the test sample and samples with different labels. Three experiment cases are conducted to validate the performance of proposed intelligent diagnosis model. The experimental results show that the proposed model can accurately classify the bearing fault with few-shot case. Additionally, the comparison experiments with the traditional models have also validated the advantage of the proposed intelligent diagnosis model for few-shot case.