Accidental failures of rotating machinery components such as rolling bearings may trigger the sudden breakdown of the whole manufacturing system, thus, fault diagnosis is vital in industry to avoid these massive economical costs and casualties. Since convolutional neural networks (CNN) are poor in extracting reliable features from original signal data, the time-frequency analysis method is usually called for to transform 1D signal into a 2D time-frequency coefficient matrix in which richer information could be exposed more easily. However, realistic fault diagnosis applications face a dilemma in that signal time-frequency analysis and fault classification cannot be implemented together, which means manual signal conversion work is also needed, which reduces the integrity and robustness of the fault diagnosis method. In this paper, a novel network named WPT-CNN is proposed for end-to-end intelligent fault diagnosis of rolling bearings. WPT-CNN creatively uses the standard deep neural network structure to realize the wavelet packet transform (WPT) time-frequency analysis function, which seamlessly integrates fault diagnosis domain knowledge into deep learning algorithms. The overall network architecture can be trained with gradient descent backpropagation algorithms, indicating that the time-frequency analysis module of WPT-CNN is also able to learn the dataset characteristics, adaptively representing signal information in the most suitable way. Two experimental rolling bearing fault datasets were used to validate the proposed method. Testing results showed that WPT-CNN obtained the testing accuracies of 99.73% and 99.89%, respectively, in two datasets, which exhibited a better and more reliable diagnosis performance than any other existing deep learning and machine learning methods.

Machine health monitoring has become increasingly important in modern manufacturers because of its ability to reduce downtime of the machine and cut down the production cost. Enormous signals acquired from machinery are capable of reflecting current working conditions by in-depth analysis with various data-driven methods. Hand-crafted feature extraction and representation from the traditional methods are essential but daunting tasks, and these methods may not be suitable for these massive data. Compared with traditional methods, deep learning ones are able to extract the best feature combination during model training without any artificial intervention, which makes it easier, more efficient, and more effective to monitor machine health, but the training cost and training time hamper its application. The short-time Fourier transform is adopted as the data preprocessing method to cut down the training cost and boost the training procedure. Inspired by the great achievements of ResNet, the new optimized model based on ResNet has been proposed with layer-by-layer dimension reduction of the feature maps. The proposed model is also able to avoid information loss in the conventional pooling layer. All the potential candidate model blocks are introduced and compared, and the best one is selected as the final one. Repeated model block layers are adapted for the best feature combinations, followed by a two-layer full connection layer for the final targets. The proposed method is validated by conducting experiments on bearing fault diagnosis and tool wear prediction dataset. The final results show that the proposed model achieves the best accuracy rate in the classification task and the lowest root mean squared error in the prediction task.

Fastener screws are critical components of rail fasteners. For the fastener screw maintenance robot, an image-based fast fastener screw detection method is urgently needed. In this paper, we propose a light-weight model named FSS-YOLO based on YOLOv5n for rail fastener screw detection. The C3Fast module is presented to replace the C3 module in the backbone and neck to reduce Params and FLOPs. Then, the SIoU loss is introduced to enhance the convergence speed and recognition accuracy. Finally, for the enhancement of the screw detail feature fusion, the shuffle attention (SA) is incorporated into the bottom-up process in the neck part. Experiment results concerning CIoU and DIoU for loss, MobileNetv3 and GhostNet for light-weight improvement, simple attention mechanism (SimAM), and squeeze-and-excitation (SE) attention for the attention module, and YOLO series methods for performance comparison are listed, demonstrating that the proposed FSS-YOLO significantly improves the performance, with higher accuracy and lower computation cost. It is demonstrated that the FSS-YOLO is 7.3% faster than the baseline model in FPS, 17.4% and 19.5% lower in Params and FLOPs, respectively, and the P, mAP@50, Recall, and F1 scores are increased by 10.6% and 6.4, 13.4%, and 12.2%, respectively.

Transfer learning has been widely applied in bearing defect diagnosis across various working conditions. However, some existing methods overlook the local and discriminative manifold structures hidden in samples with different distributions, which may degrade model diagnostic performance. In this study, a local manifold discriminant domain adaptation (LMDDA) transfer learning method is proposed for bearing fault diagnosis. The source and target domains are firstly converted in manifold feature subspace to obtain robust feature representations. Then, an intermediate domain with similar distribution to target domain is generated from the source domain in the potential common subspace to reduce the distribution difference between source and target domains. Meanwhile, the local generated discrepancy metric is used to evaluate the discriminability of data, preserve the local manifold structure of the generated data, and align the local data. The intermediate and target domains are aligned by the maximum mean discrepancy metric to enhance the consistency of the local and global structure of data. On this basis, the low-rank constraint is introduced to guarantee the correlation between source and target data in virtue of the block diagonal property, which can improve the generalization ability and performance stability of the model. Finally, the least squares method is utilized to achieve bearing fault diagnosis. A series of cross-domain fault diagnosis experiments were carried out to verify the effectiveness and superiority of the proposed method, and the results illustrate that LMDDA can accurately detect diverse fault types and handle intricate cross-domain adaptive scenarios.

Rolling bearing is the core component of industrial machines, but it is difficult for common single signal source-based fault diagnosis methods to ensure reliable results since sensor signals are vulnerable to the pollution of background noises and the attenuation of transmitted information. Recently, multi-source information-based fault diagnosis methods have become popular, but the information redundancy between multiple signals is a tough problem that will negatively impact the representational capacity of deep learning algorithms and the precision of fault diagnosis methods. Besides that, the characteristics of various signals are actually different, but this problem was usually omitted by researchers, and it has potential to further improve the diagnosing performance by adaptively adjusting the feature extraction process for every input signal source. Aimed at solving the above problems, a novel model for bearing fault diagnosis called multi-branch selective fusion deep residual network is proposed in this paper. The model adopts a multi-branch structure design to enable every input signal source to have a unique feature processing channel, avoiding the information of multiple signal sources blindly coupled by convolution kernels. And in each branch, different convolution kernel sizes are assigned according to the characteristics of every input signal, fully digging the precious fault components on respective information sources. Lastly, the dropout technique is used to randomly throw out some activated neurons, alleviating the redundancy and enhancing the quality of the multiscale features extracted from different signals. The proposed method was experimentally compared with other intelligent methods on two authoritative public bearing datasets, and the experimental results prove the feasibility and superiority of the proposed model.