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.

Damage identification is a key problem in the field of structural health monitoring, which is of great significance to improve the reliability and safety of engineering structures. In the past, the structural strain damage identification method based on specific damage index needs the designer to have rich experience and background knowledge, and the designed damage index is hard to apply to different structures. In this paper, a U-shaped efficient structural strain damage identification network SDFormer (structural damage transformer) based on self-attention feature is proposed. SDFormer regards the problem of structural strain damage identification as an image segmentation problem, and introduces advanced image segmentation technology for structural damage identification. This network takes the strain field map of the structure as the input, and then outputs the predicted damage location and level. In the SDFormer, the low-level and high-level features are smoothly fused by skip connection, and the self-attention module is used to obtain damage feature information, to effectively improve the performance of the model. SDFormer can directly construct the mapping between strain field map and damage distribution without complex damage index design. While ensuring the accuracy, it improves the identification efficiency. The effectiveness and accuracy of the model are verified by numerical experiments, and the performance of an advanced convolutional neural network is compared. The results show that SDFormer has better performance than the advanced convolutional neural network. Further, an anti-noise experiment is designed to verify the anti-noise and robustness of the model. The anti-noise performance of SDFormer is better than that of the comparison model in the anti-noise experimental results, which proves that the model has good anti-noise and robustness.

Chaos-based secure communication technology can provide secure protection for the information transmission at the physical layer. In this paper, a neural network-based chaos shift keying communication scheme is proposed. This scheme carries digital information by altering the underlying time-delayed dynamical mapping relationships of the system. It utilizes the nonlinear modeling capability of neural networks to extract and judge the underlying time-delayed dynamical mapping relationships, thus achieving demodulation without the need for physical synchronization. In numerical simulations, the scheme exhibits high noise resistance and can effectively resist return mapping attacks and time delay signature extraction, providing possibilities for secure communication under poor conditions.