Yang Xiang

In the research on navigation efficiency optimization in complex waters, accurate vessel traffic flow prediction has emerged as a critical challenge in the field of intelligent maritime navigation. Based on Resolution A.1158(32) of the International Maritime Organization (IMO) and expert knowledge of Vessel Traffic Service (VTS), this study proposes a novel framework for vessel port reporting and VTS decision-making, and constructs a feature-fused database integrated with meteorological data. A hybrid deep learning model based on the DeepAR framework is developed, which extracts the spatial dependencies of vessel traffic via a convolutional neural network (CNN), captures temporal features using a bidirectional long short-term memory network (BiLSTM), and acquires long-range dependencies with a self-attention mechanism (SAM) to achieve multi-dimensional feature fusion for vessel traffic flow prediction. A case study is conducted on the narrow section of the Lüsi inbound waterway in Tongzhou Bay, China, to verify the effectiveness of the proposed model. The results demonstrate that the proposed method can realize high-precision probabilistic prediction of vessel traffic flow. Compared with the benchmark methods, the proposed model reduces both the Mean Absolute Scaled Error (MASE) and the Mean Absolute Error (MAE) simultaneously for medium- and long-term forecasting ( > 24 hours) under specific working conditions (VTS six-shift mode). This study can provide probabilistic decision support for intelligent traffic management in narrow navigable waters, and serve as an important reference for VTS scheduling as well as port and shipping operation dispatching.

In recent years, deep learning-based fault diagnosis methods for diesel engines have been developed assuming that a comprehensive dataset is available for one-time learning. However, in reality, it is impractical to obtain a complete dataset all at once. Instead, new data is gradually acquired over time. To address this issue, this paper proposes an incremental fault diagnosis method with bias correction (IFD-BiC). When receiving incremental fault data, the IFD-BiC model incorporates new classifier nodes and updates network parameters using a combination of distillation loss and cross-entropy. During the update process, old samples saved in the exemplar set are replayed, and a bias correction layer is used to correct the prediction bias caused by the imbalance between new and old samples. This enables the model to learn new fault tasks without experiencing catastrophic forgetting. To ensure the applicability of the method to various operating conditions of diesel engines, cyclic angular vibration with consistent length at different speeds is used as the model input. Through experimental validation across a pair of distinct diesel engine models, the proposed method achieves highly accurate fault diagnosis on stream format training data. Moreover, an evaluation against leading incremental learning techniques affirms the proposed method’s advantage.

In rotating machinery, the condition of rolling bearings is paramount, directly influencing operational integrity. However, the literature on the fault evolution of rolling bearings in their nascent stages is notably limited. Addressing this gap, our study establishes an innovative nonlinear dynamic model for early fault evolution of rolling bearings based on collision impact. Firstly, considering the fault evolution characteristics, the influence of the rolling element and fault structure, the dynamic model of early fault evolution between the rolling element and the local fault is established. Secondly, according to the Hertzian contact deformation theory, a nonlinear dynamic model of rolling bearings expressed as mass-spring is established. Thirdly, the energy contribution method is used to integrate the fault evolution model and the nonlinear dynamic model of the rolling bearing. A nonlinear dynamic model of early fault evolution of the rolling bearing is proposed by using the Lagrangian equation. Comparing the simulation results of the nonlinear dynamic with the experimental results, it can be seen that the numerical model can effectively predict the evolution process and vibration characteristics of the fault evolution of rolling bearings in the early stage.

Self-supervised learning (SSL) aims to extract useful representations from unlabeled data by maximizing the agreement between positive pairs. However, traditional SSL relies on carefully designed data augmentation methods to generate positive pairs. When dealing with 1D vibration signals, data augmentation prone to potentially compromise the fault information in the original signals. Therefore, this paper proposes a data augmentation-free SSL framework for diesel engine fault diagnosis called Domain Adaptation Variance Invariance Covariance Regularization (DA-VICReg). The DA-VICReg uses cyclic angular vibrations collected within the same time period as positive pairs and extracts useful features from unlabeled data using a loss function composed of three terms: Variance, Invariance, and Covariance. We found that when positive pairs originate from different operating conditions, such as varying speeds and torques, the model can develop feature extraction capabilities that remain unaffected by changes in operating conditions. In addition, a spatial pyramid pooling layer and a trilinear attention module are used to extract vibration features at different scales and focus on critical spatial locations and channels. Finally, the proposed approach was validated through experiments on two types of diesel engines, and a comparison with prominent SSL methods confirms the superiority of the proposed approach. In engineering practice, this method can utilize a large amount of signals stored in different time periods for self-supervised training and learn useful features for downstream fault diagnosis tasks.

Based on the dimension invariance property of the data-driven bearing fault diagnosis method, unstable condition data can result in the loss of information and reduced diagnostic accuracy due to inconsistent data dimensions. Furthermore, the fixed parameters of the output layer restrict its ability to accurately diagnose faults beyond the training set, particularly compound faults with limited data. To address these challenges, this study proposes an ensemble deep learning approach for identifying untrained compound faults in bearings operating under non-stationary conditions. Firstly, a signal angular domain processing technique is employed to standardize the dimensionality of the bearing’s state information, effectively mitigating information loss. Secondly, a feature extraction model is established to dynamically capture local microscopic and multilevel features utilizing the adaptability of convolutional neural network (CNN), and it can mine the relevant features of compound faults through the single-fault features. In the verification process, the kmeans algorithm with scalable classification is used to optimize the classifier of CNN. Specifically, the number of cluster centers in kmeans is set to exceed the count of training fault categories. Identification of untrained compound faults is achieved by calculating the Euclidean distances between each feature and the cluster centers, based on the principle of minimum distance. It addresses the challenge of inadequate diagnostic rates for untrained compound faults. The diagnostic outcomes prove that the proposed method has a high diagnostic robustness and generalization ability, which can effectively solve the problem of insufficient fault data and provide a new way of diagnosing untrained compound faults.

The discrete model in the standard finite element method (FEM) always exhibits stiffer behavior than the actual continuous model. This leads to an unavoidable dispersion error that increases rapidly with the wave number. To overcome this issue in underwater acoustic scattering problems, a coupled hybrid smoothed radial point interpolation method (CHSRPIM) is proposed to reduce the dispersion error. In CHSRPIM, the modified Dirichlet-to-Neumann boundary condition is imposed on an artificial boundary to ensure that the acoustic wave correctly travels outward, and a hybrid acoustic stiffness is created using the local gradient smoothing operation to soften the stiffness. To obtain a very close-to-actual stiffness of the original model, a cell-based radial point interpolation method with three real and four virtual nodes is devised to create the acoustic shape function. The major benefit of the CHSRPIM is that, for a given mesh, the dispersion error is effectively reduced compared with that of the FEM without increasing the degrees of freedom. The performance of the proposed method is numerically evaluated. Numerical experiments are conducted to investigate the properties of the proposed method. The simulation results indicate that the CHSRPIM possesses superior computational performance and can significantly improve accuracy; therefore, it has good potential for resolving practical acoustic scattering problems in engineering.