Yongze Jin

With the widespread application of unmanned aerial vehicles (UAVs), intrusion detection based on wireless communication traffic has become increasingly critical. While deep learning (DL) models have significantly enhanced anomaly detection accuracy, recent research indicates that they remain vulnerable to adversarial sample attacks, which pose severe threats to the open communication networks used by UAV systems. To address this challenge, an enhanced UAV intrusion detection framework is proposed for defending against adversarial samples, integrating generating adversarial samples with adversarial networks (AdvGAN) and decision boundary reconstruction. Specifically, a convolutional neural network (CNN) was first developed as the baseline intrusion detection model to detect anomalous traffic in UAV wireless communication. Subsequently, to overcome the gradient-dependent perturbation limitations of traditional adversarial attack methods, AdvGAN is employed to analyze real monitored traffic data and generate semantically consistent adversarial samples that retain the intrinsic characteristics of authentic network flow. Finally, a hybrid adversarial-real data training strategy is proposed to dynamically reconstruct the model’s intrusion decision boundaries. This strategy augments the training dataset with AdvGAN-generated adversarial samples, enabling the model to adaptively adjust decision thresholds through iterative backpropagation. On a real-world dataset, the superior performance of the proposed method is verified: it maintains high detection accuracy for legitimate traffic while substantially reducing the success rate of adversarial attacks. Experimental results confirm the stealthiness of generated adversarial samples and the robustness of the enhanced detection model, validating the effectiveness of the proposed framework.

With the continuous increase of train speed, track irregularity poses a serious threat to the safety and comfort of train operation. In this paper, a track irregularity detection method based on improved masked sparse Transformer (MST) is proposed to alleviate the contradiction between detection efficiency and accuracy. By analyzing the coupling mapping relationship between dynamic offset of train body and track irregularity, a lightweight sparse Transformer model is constructed. The sparse attention mechanism based on double exponential attenuation is adopted in this model, and the algorithm complexity is significantly reduced, while the Decoder structure is simplified to further reduce the number of model parameters and improve the detection efficiency. The experimental results show that the proposed method is superior to the traditional recurrent neural network and the standard Transformer, the track irregularity can be detected in real-time with high precision, and the reliability of the model is verified by an interpretable analysis. The research in this paper provides an efficient and low-cost solution for track condition monitoring, which has important engineering application value.

In recent years, domain adaptation methods have been widely applied in cross-domain fault diagnosis. However, these methods are based on the assumption that the source domain and target domain share the same fault types. In practical operation, unknown fault types may occur in mechanical equipment, leading to a mismatch between the fault types of the source domain and the target domain. To address the diagnostic challenges in open-set cross-domain tasks, an improved parallel adversarial complementary domain adaptation network (PACDAN) is established. Firstly, a parallel channel complementary attention mechanism (PCCAM) module is constructed, in which the dual parallel branching structure and the multi-scale feature interaction mechanism are designed, so that the model’s ability to pay attention to key information is improved. Second, a parallel adversarial complementary (PAC) module is designed, a macroscopic classification perspective is added, and the shortcoming of a single view of fault classification is overcome. Finally, experiments are conducted on three public datasets, systematically comparing the performance of the proposed method with existing methods. Through visualization techniques and ablation experiments, the results are intuitively presented and analyzed, validating the effectiveness of the proposed method.

In this article, a novel prediction index is constructed, a hybrid filtering is proposed, and a remaining useful life (RUL) prediction framework is developed. In the proposed framework, different models are built for different operation states of rolling bearings. In the normal state, a linear model is built, and a Kalman filter (KF) is implemented to determine the failure start time (FST). In the degradation state, a dimensionless prediction index CRRMS is constructed, based on the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and wavelet threshold. Then, a double exponential model is established, and the hybrid filtering is proposed to estimate the future trend of CRRMS, which is combined by a particle filter (PF) and an unscented Kalman filter (UKF). At the same time, dynamic failure threshold technology is adaptively used to determine the failure thresholds of different bearings. Furthermore, the RUL is extrapolated at the moment the prediction index exceeds the failure threshold. Finally, the effectiveness and practicability of the proposed method is verified on the bearing dataset given by the PRONOSTIA platform.

Considering the competitive failure that exists in the operation of industrial systems, including degradation failure and sudden failure, this paper presents a reliability assessment method based on the three-parameter Weibull distribution and the Wiener process. Nonlinear functions are proposed to establish the relationship between the different failure processes, and the reliability model is derived. The Metropolis-Hastings (MH) sampling algorithm of the Monte Carlo Markov Chain (MCMC) method is employed to estimate the parameters in this study. The reliability assessment results are obtained by the real degradation samples. The results show that the reliability model incorporating the three-parameter Weibull distribution produces more comprehensive and dependable results. Furthermore, MH sampling can solve the issues of complex likelihood functions that cannot directly obtain parameter evaluation results. Additionally, this paper analyzes the sensitivity of the model parameters, thereby offering theoretical support for enhancing the safe operation of the system.

Classification of coronary artery stenosis is essential in assisting physicians in diagnosing cardiovascular diseases. However, due to the complexity of medical diagnosis and the confidentiality of medical images, it is difficult to obtain many coronary artery stenosis image samples for scientific research in general. In addition, the degree, location, and morphology of coronary artery stenosis in different patients, as well as the noise in computed tomography angiography (CTA) images, make it challenging to extract typing features effectively. To address the above problems, first, a joint segmentation method is proposed based on maximum between-class variance and region growing to extract key regions from CTA images to facilitate further feature extraction. Then, a coronary artery stenosis classification model based on convolutional block attention module (CBAM) and transfer learning is constructed, which can effectively improve the model training effect under the insufficient samples. Finally, a dataset of CTA images from actual patients is applied for experimental verification. The experimental results show that the accuracy (Acc) of the proposed method is up to 98.99%, which is greatly improved compared with several machine learning algorithms and neural network methods. It can be concluded that the accuracy of coronary stenosis classification can be considerably improved, and a reasonable scientific basis is provided for clinical medical diagnosis.

Robot visual place recognition is a research hotspot, and robot place recognition systems based on 3D point clouds are even more focused. In recent years, scan-context descriptor imagery has become a standard method for 3D point cloud positioning. Based on its system characteristics, we integrated it with the Shuffle Attention (SA) mechanism module to improve its system performance. And we tested it on the public database NCLT dataset (North Campus Long-Term Vision dataset), and our method has good place recognition results.

In this article, a condition monitoring approach is proposed based on vibration signal, aiming at improving the adaptability of feature extraction and the accuracy of classification. First, the original vibration signal acquired under certain working condition is preprocessed by dividing it into multiple segments, followed by the signal decomposition. Then, the features of each decomposed signal are extracted based on the theory of diversity entropy (DE). Two parameters in the DE are optimized considering the fact that these parameters are crucial for the classification result. The optimization objective is to make the different segments of the signal collected in the same working condition have approximate feature characterization. By this means, the feature of the signal is captured adaptively and accurately using the optimized entropy value. Finally, the support vector machine is used to identify the extracted feature vectors to realize condition classification. The experiments on three representative platforms, including a crystal lifting-rotation system in our laboratory, are conducted to verify the effectiveness of the proposed method.

In this paper, we propose an output feedback flight tracking control scheme for helicopter attitude and altitude systems with unmeasured states under full state constraints. Firstly, a state observer is constructed based on the measured output signals, which is proven to be rigorous since all states are constrained within the desired and assigned scopes. Secondly, the flight tracking controller is built using the state estimations with the full state constraints control method. Then, the Barrier Lyapunov function method is adopted to guarantee the stability of the composite closed-loop nonlinear error systems. Meanwhile, the linear matrix inequality technology is applied to calculate the gains of the state observer. Finally, a numerical simulation example is provided to confirm the reasonableness of the full state constraint output feedback flight tracking control method.