Tingting Guo

The fault diagnosis of rolling bearings plays a critical role in ensuring the safe and reliable operation of mechanical systems. However, the generalization ability of traditional graph neural networks is limited in practical scenarios due to the domain shifts caused by various conditions, the presence of unlabeled data and the insufficient bearing fault samples. To further improve the fault diagnosis accuracy under different working conditions, an unsupervised domain adaptation (DA) method based on dual kernel attention perturbation enhanced multi-scale convolution and hypergraph convolution (DKAP-MSCHC) is proposed. First, DKAP module is dynamically designed by adjusting the size of convolution kernel to preserve the dimensional features and achieve dynamic data augmentation without losing global information. Second, the local and global features of vibration signals are extracted through multi-scale convolution, and the key features are identified by a temporal multi-head attention (T-MHA). In multi-head and multi-region adaptive hypergraph module, each node dynamically determines the range of its adjacent nodes to capture the structural details at different levels. Additionally, a DA method based on dynamic bandwidth maximum mean discrepancy is proposed to minimize the distribution discrepancy between the source domain and the target domain. Finally, experiments were conducted to verify the proposed method, and the results had achieved an average accuracy of 99.20% and 97.31% on two datasets. Compared with several state-of-art intelligent fault diagnosis methods, the proposed method has demonstrated its effectiveness and superiority in various transfer tasks and complex working environments.

Turbulence is of great significance for aerospace engineering, marine hydrodynamics, and optimizing fluid engineering systems. Deep learning techniques are increasingly being applied to turbulence prediction and modeling. However, there are still some issues that need to be addressed, such as limited ability to achieve high spectral accuracy, lack of physical consistency and degraded stability in long-term predictions. Therefore, a physics-guided latent diffusion model based on multi-scale Fourier autoencoder (PMF-LDM) is proposed. First, a multi-scale Fourier autoencoder (MFAE) is designed with a multi-scale Fourier convolution (FourierConv) structure. Each FourierConv block integrates spatial and Fourier paths in parallel, with the Fourier branch using different numbers of truncated modes to extract multi-scale features. Second, the diffusion probabilistic model solver (DPM-Solver) is integrated into the latent diffusion model (LDM), and the sampling in latent space is accelerated by defining a continuous-time evolution function. In addition, the physics constraint module based on partial differential equation (PDE) incorporates physics loss terms derived from the residuals of the Navier–Stokes (N-S) equations and combines the physical loss terms with the noise prediction through weighted combination to form the overall loss function. Finally, experiments were conducted on two turbulence datasets. In long-term predictions, both the correlation coefficient (CC) and the structural similarity index measure (SSIM) exceed 80% and maintain strong physical consistency and temporal stability. This capability enables more accurate and reliable turbulence prediction and provides a new insight for the spatiotemporal modeling of complex flow systems.

Conventional fire monitoring systems frequently exhibit high false alarm rates, delayed response times, and a lack of closed-loop control capabilities, which severely constrain their deployment in complex real-world environments. To address these issues, this paper proposes an embedded fire detection, tracking, and extinguishing system based on multimodal information fusion and a lightweight neural model. The system follows a “Perception–Decision–Execution–Feedback” closed-loop paradigm and is implemented on a heterogeneous cooperative computing architecture comprising OpenMV4 H7 Plus and STM32F103C8T6 microcontrollers. The perception layer implements a decision-level RGB-infrared fusion mechanism that incorporates a pruned, INT8-quantized lightweight FOMO model, enabling real-time fire detection with an inference latency of 210 ms and a model size of merely 1.8 MB under resource-constrained embedded conditions. The decision layer employs a Bayesian inference-based multimodal fusion framework that effectively suppresses spurious fire interference. The vision-only false detection rate is 15.3%. After infrared fusion verification, the system-level false alarm rate is reduced to 2.0% on the interference test set. In the execution layer, a sixth-degree polynomial jet trajectory model was established and combined with an improved PID–PI dual-loop controller to enable dynamic optimization of spray angle and flow rate in real time. Experimental results demonstrate that the proposed system achieves an average fire recognition accuracy of 95.6% with a false alarm rate as low as 1.4%. Furthermore, it realizes an extinguishing accuracy better than ±5 cm within an effective operating range of 10–60 cm and completes the entire perception-to-extinguishing cycle within 8.5 s under illumination conditions ranging from 50 to 100,000 lux. These results demonstrate the excellent real-time capability, robustness, and energy efficiency of the proposed system, providing a practical and scalable solution for autonomous embedded fire-fighting applications in household, industrial, and warehouse environments.

Navigation planning and execution for underwater gliders operating in thermocline-affected environments is challenging due to the coupled influence of energy constraints, spatially distributed environmental disturbances, and limited control authority. Spatially varying thermocline structures act as structured environmental disturbances that degrade motion efficiency and tracking accuracy, and therefore must be explicitly considered in both path planning and control design. This paper proposes a hierarchical control-oriented decision framework for underwater glider navigation in thermocline regions. At the planning layer, a thermocline-aware multi-objective optimization problem is formulated to regulate the trade-off between navigation efficiency and cumulative environmental disturbance, characterized by total path length and cumulative thermocline exposure, respectively. A multi-objective artificial bee colony (MOABC) algorithm is employed to generate a set of Pareto-optimal reference trajectories that explicitly reveal this trade-off. At the execution layer, pitch angle regulation is formulated as a stochastic tracking control problem under environmental uncertainty. A Markov Decision Process (MDP) is constructed to model the coupled effects of pitch control on energy consumption and trajectory deviation, and a deep deterministic policy gradient (DDPG) algorithm is adopted to synthesize a feedback control policy for adaptive pitch regulation during path execution. Simulation results demonstrate that the proposed framework effectively reduces cumulative thermocline exposure and overall energy consumption while maintaining improved trajectory consistency compared with representative benchmark methods. These results indicate that integrating multi-objective planning with learning-based control provides an effective control-oriented solution for constrained underwater glider navigation in thermally stratified environments.

In recent years, the CNN-Transformer model, which combines convolutional neural networks with transformers, has gained significant attention due to its ability to effectively capture both local and global features. However, these CNN-Transformer models usually have high computational complexity and their performance tends to decline in noisy environments. This makes it a challenge to balance diagnostic accuracy and computational efficiency, and limits their widespread application in industrial environments. To address these issues, a novel lightweight anti-noise CNN-Transformer model based on feature decoupling knowledge distillation (TIDAEformer-FDKD) is proposed in this paper. This model first employs a high-capacity CNN-Transformer (IDAEformer) as the teacher network, which combines the improved denoising autoencoder (IDAE) with multiple consecutive multi-scale enhanced transformer blocks (METBs) to extract more discriminative fault features. Each METB block consists of a multi-scale depthwise separable convolution module (MDSM), an interactive spatial pyramid masked attention (ISPMA) module, and an efficient transformer (ETransformer) module, and these modules work together to capture both local and global fault features. The student network is a tiny IDAEformer (TIDAEformer), and the TIDAEformer-FDKD utilizes FDKD strategy to effectively transfer the comprehensive knowledge from the teacher network to the student network. Experimental results show that the diagnostic accuracy of the proposed model on the Paderborn University (PU) dataset and Jiangnan University (JNU) dataset reached 99.28% and 99.87% respectively. Even in noisy conditions, the proposed model consistently outperforms the state-of-the-art methods and maintains a low model complexity. The proposed TIDAEformer-FDKD effectively achieves a balance between high diagnostic accuracy and computational efficiency, and provides a practical solution for real-time bearing fault diagnosis under resource-constrained conditions.

Accurate trajectory estimation for underwater gliders remains a significant challenge due to the limited observability of horizontal displacement during submerged operations. This work presents a novel hybrid framework that enhances horizontal trajectory prediction by integrating reinforcement learning (RL)-based inference with the dynamic response of a motion model via fuzzy logic fusion. A proximal policy optimization (PPO) agent is trained to generate an angle of attack (AoA) sequence that minimizes terminal displacement error, providing task-optimal guidance based on a kinematics driven transition model. To address the absence of physical constraints and the lack of detailed dynamic modeling in the RL process, a fuzzy fusion strategy is introduced. This strategy adaptively blends the RL inferred and dynamics derived AoA according to discrepancies in motion state predictions. Experimental results from indoor trials demonstrate that the proposed method achieves endpoint localization accuracy within 0.3m. Meanwhile, the fused AoA maintains strong consistency with the measured AoA sequence, with a Kullback–Leibler (KL) divergence below 0.06. The proposed framework provides an adaptive and computationally efficient solution for glider localization under partial observability, highlighting the potential of fuzzy system integration in advancing data-driven autonomy for marine platforms.

This paper investigates the joint state and fault estimation problem for a class of nonlinear systems subject to both measurement censoring (MC) and random missing measurements (MMs). Recognizing that state estimation for nonlinear systems in complex environments is frequently compromised by MMs, MC phenomena, and actuator faults, a novel joint estimation framework that integrates improved Tobit Kalman filtering and federated fusion is proposed, enabling simultaneous robust estimation of system states and fault signals. Among them, the Tobit measurement model is introduced to characterize the phenomenon of MC, a set of Bernoulli random variables is used to describe the MM phenomenon and common actuator faults (abrupt and ramp faults) are considered. In the fusion estimation stage, each sensor transmits observations to the local estimator for preliminary estimation, then sends the local estimated values to the fusion center for generating fusion estimates. The local filtering error covariance is ensured and the upper bound is minimized by reasonably determining the filter gain, while the fusion center performs fusion estimation based on the federated fusion criterion. In addition, this paper proves the boundedness of the filtering error of the designed estimator under certain conditions. Finally, the effectiveness of the estimation framework is demonstrated through two engineering experiments.

Reconstructing high-resolution flow fields remains a significant challenge in fluid mechanics due to the limitations of measurement techniques. In recent years, physics-informed neural networks (PINNs) have been proposed to reconstruct flow fields by embedding physical constraints. However, accurate solutions are difficult to obtain when faced with complex fluid properties or high noise levels. Therefore, a physics-informed convolutional network based on feature fusion (FFPICN) is proposed to reconstruct high-resolution flow fields from sparse and noisy velocity data. The framework mainly consists of a feature extraction module and a physical information module. Unlike traditional PINNs that mainly rely on fully connected networks to extract features, the new feature extraction module comprises a parallel convolutional layer, a feature fusion layer, and a convolutional fusion module to extract and fuse shallow and deep fluid features. The spatial feature extraction ability of convolutional neural networks is used to enhance the ability to capture small-scale fluid structures. The physical information module embeds the constraints of the transformed Navier–Stokes equations into deep features to improve the accuracy of flow field reconstruction. Furthermore, a preprocessing technique is designed to linearly transform the data and the Navier–Stokes equations concurrently. Finally, wake and turbulence data are investigated at different levels of sparsity and noise. Experimental results show that FFPICN can reconstruct high-resolution flow fields even at 1% sparsity or 20% noise level. The relative L2 error of the reconstructed flow field by FFPICN is reduced by about 30% compared to traditional PINNs and their variants.