Deep learning provides a feasible fault diagnosis method for intelligent mechanical systems. However, this method requires a large amount of marking data, which greatly limits its application in the actual industry. Therefore, this paper proposes a multi-layer adaptive convolutional neural network unsupervised domain adaptive bearing fault diagnosis method (MACNN), which is especially suitable for bearing fault classification under variable working conditions. First, a new method to improve domain alignment is proposed (LD-CORAL). This method uses Log-Euclidean distance to measure deep coral loss, which solves the problem that the covariance matrix cannot be aligned correctly in the manifold structure. Then, it proposes multi-layer adaptation of LD-CORAL loss in the fully connected layer, and combines Center-Based discriminative loss to improve the feature learning ability of the model, which can improve the classification accuracy and domain adaptation performance of the model. Finally, in order to verify the effectiveness and feasibility of the proposed method, the method is applied to the multi-fault diagnosis of gearbox bearings under variable working conditions. Comparing the classification results of different methods, the conclusion shows that this method is more effective for bearing fault classification under variable working conditions.

Nowadays, the methods of remaining useful life (RUL) prediction based on deep learning only use single model, or a simple superposition of two models, which makes it difficult for to maintain good generalization performance in various prediction scenarios, and ignores the dynamic sensitivity of features in the prediction, limiting the accuracy. This paper proposes a method of RUL prediction of bearing using fusion network through two-feature cross weighting (FNT-F). First, a fusion network with two subnets is proposed in this paper to adapt to the prediction problem in different scenarios. Meanwhile, a method of cross weighted joint analysis of the two features is proposed to make up for the shortcomings of feature analysis and achieve complementarity between time-domain and time-frequency features.

Aiming at the problem that the time-frequency image of bearing fault characteristics is relatively weak and difficult to identify. This paper presents a time-frequency analysis method of local maximum synchrosqueezing transform based on image enhancement. Firstly, the instantaneous frequency of the collected vibration signal is obtained through local maximum synchrosqueezing transformation. Secondly, a local histogram cropping equalization image enhancement algorithm is proposed, which is used to obtain time-frequency images with clearer textures. Then, in order to extract fault features from the enhanced instantaneous frequency (IF) image, A new neural network is proposed. The network consists of Multi-size convolution kernel module, Dual-channel pooling layer and Cross Stage Partial Network (MDCNet). Finally, the fault signal was collected on the bearing fault test bench for prediction, and the accuracy rate reached 99.7%. And compared with AlexNet, VGG-16, Resnet and other methods. The results show that the method can meet the needs of actual engineering.INDEX TERMS Fault diagnosis, image enhancement, instantaneous frequency.

In recent years, residual network has been widely used in the field of intelligent diagnosis because of its powerful function. This paper proposes a novel dense residual network (DRNet), which combines the advantages of dense connections and residual learning to prevent gradient disappearance and network degradation caused by network deepening for efficient fault diagnosis of rolling bearings. First, each sub-block in the dense network is deeply processed so that it has better nonlinear expressive ability to extract deep fault features. Then, the residual learning is embedded in each sub-block of the dense network, so that each sub-block processed by deepening will not show the phenomenon of network degradation. Finally, an Adam-Subtracted momentum (Adam-S) optimization algorithm is proposed, which adds the first-order momentum and the second-order momentum of the previous gradient into the expression of the second-order momentum of the current gradient, which enhances the connection between the parameters in the two adjacent gradients in the Adam algorithm. It makes the algorithm more reliable and the gradient prediction more accurate. Without adding additional parameters, the training stability of the algorithm in complex environments is further improved. Experiments on two kinds of data sets under different working conditions are carried out for many times, and comparison with Random Forest, Support Vector Machine, Dense Network, Residual Network, AlexNet and DRNet-Adam proves the effectiveness and feasibility of the proposed model and optimization algorithm.

In recent years, data-driven intelligent diagnosis methods have been widely applied in the field of bearing fault diagnosis. However, these methods involve some expert experience and knowledge, and cannot accurately mine bearing fault characteristics under different loads. To solve this problem, this paper proposes a First-order differential filtering spectrum division method (FDFSD) and an information fusion multi-scale network (IFMSNet) to realize bearing fault diagnosis under different working conditions. First of all, the proposed spectrum division method based on the first-order differential filtering, the first-order differential processing of time domain signals, and the introduction of triangular filter, reclassify the spectrum features, highlight feature information, can accurately extract bearing fault features. Secondly, a new multi-scale network model of information fusion is constructed in this paper. Convolution kernels of different sizes are used to extract fault features of bearings of different scales, and information fusion is carried out to identify bearing working conditions and realize intelligent diagnosis of bearings under different loads. Finally, in order to verify the effectiveness and accuracy of the proposed method, it is verified on a variety of bearing experimental data sets. The results show that the average prediction accuracy of the proposed method is 99.11% and 97.74%, respectively. Compared with the proposed three single-scale network, K-Nearest Neighbor (KNN), Naive Bayes (NB), Support Vector Machine (SVM) and Random Forest (RF) methods, the proposed method has more advantages in bearing intelligent diagnosis under different loads.