Search

As an indispensable material in modern industry, steel requires real-time surface defect detection to ensure high-quality manufacturing. However, steel surface defects present significant challenges due to their tiny size, diverse morphology, and uneven feature distribution. To address these challenges and satisfy the balance between accuracy and detection speed, an efficient steel strip surface defect detection network, ACPP-Net, is proposed in this study. Firstly, Adaptive Ghost Convolution, the LM-block, is introduced to meet the need for rapid steel defect detection. By integrating Adaptive Ghost Convolution, this module increases efficiency by reducing redundant information acquisition and adaptively assigning weights to defect features. Secondly, a novel feature enhancement module, FEM-block, is proposed to address the complexity of steel defects and distinguish their subtle differences. This module excels at capturing complex defect textures, aiding in the accurate differentiation of various defects. Additionally, a channel spatial pyramid pooling (CSPP) module is incorporated into the final part of the backbone network. This module effectively helps the network understand the characteristics and distribution of steel defects. Extensive experiments on the NEU-DET and GC10-DET datasets demonstrate ACPP-Net’s superior performance, achieving 82.1% and 71.1% mAP respectively, while maintaining real-time detection capabilities. The detection performance of this model is superior to other methods. These results demonstrate the model’s high accuracy and real-time detection capabilities, thus contributing to efficient and high-quality steel detection processes.

The Classifier Chains (CC) method is an effective method for multi-label classification, with its performance significantly contingent on the label order. However, most existing algorithms focus on determining the label order by considering label correlations as symmetric, ignoring that label correlations prevail asymmetric in real-world applications. To overcome the problem, we propose a novel Relative Entropy and PageRank-based Classifier Chains for multi-label classification (REPCC). Firstly, to completely capture label correlation, REPCC exploits relative entropy to estimate the asymmetric correlation between pairs of labels, filter out those pairs of labels with lower correlation, and construct a label correlation matrix. Secondly, by treating each label as a node, a directed graph is generated based on the filtered label correlation matrix. Finally, drawing an analogy between labels and web pages, the idea of PageRank, commonly used to measure the importance of web pages, is employed to rank labels in the directed graph and establish a classifier chain. Experimental results on ten public datasets from various domains show that REPCC outperforms state-of-the-art methods in CC.

The detection capability of deep learning-based stereo matching in industrial applications is inherently limited due to challenges posed by weak texture and inconsistent reflectance, making it difficult to accurately recover complex surface details. To achieve accurate measurements, this paper presents an end-to-end speckle stereo matching network that incorporates fringe, Gray code, and speckle projection patterns. The model is trained using a high-precision dataset consisting of thousands of pairs generated through binocular Gray code-assisted phase shifting. After establishing local correspondences between the left and right images using speckle patterns, the images are used as inputs to the network. The proposed network consists of two siamese 2D feature extraction networks. One network is dedicated to cost volume computation, while the other focuses on weight refinement feature extraction. The former network incorporates a lightweight module for extracting high-dimensional fusion features. These features are obtained from different dilation scales and randomly concatenated along the channel dimension. Patch convolution is utilized to effectively adapt to pixel features at various levels, reducing redundancy within the cost volume and improving the network’s capacity to learn from ill-posed regions. Experimental results demonstrate that the proposed network achieves a significant improvement of approximately 10.7% in matching accuracy compared to state-of-the-art networks on public datasets. Furthermore, this method exhibits outstanding matching results when applied to diverse industrial scenarios. The reconstruction error for the radius of optical standard spheres is below 0.06-mm, which meets the demands of the majority of industrial applications.

Aiming at the problems of low efficiency and difficulty in constructing acceleration devices in traditional interval optimization algorithms (IOAs), this paper constructs a valid acceleration device based on a more concise point evolutionary strategy (ES), and then proposes a novel hybrid IOA (HIOA) with no requirement on the derivative of the objective function. The HIOA first divides the initial search area into $N$ equal parts, randomly selects multiple point individuals in each subinterval to represent their information, and performs the optimization with fewer iterations using ES for all point individuals to make them closer to the optima; then selects reliable subintervals containing more point individuals to split, and deletes unreliable subintervals without any point individuals; finally, provides a reliable upper bound to direct the pruning operation to further improve the search efficiency. Furthermore, the convergence property of the proposed algorithm is analyzed. Extensive numerical experiments on several typical test functions and the application to the bounded error parameter estimation demonstrate the superiority of HIOA by comparing it with the existing conventional algorithms, which confirms the effectiveness and applicability of the suggested algorithm.

The traditional physical layer key generation technology uses channel temporal variation, reciprocity, and spatial decorrelation to realize key distribution and fast update. However, due to the complex, time-varying and time-delayed features, the underwater acoustic channel environment often changes during channel detection, resulting in impaired channel reciprocity. To solve this problem, this paper proposes an orthogonal frequency division multiplexing pilot structure based on the features of underwater acoustic time-varying multipath channel. The structure can complete key detection when channel reciprocity is impaired, promotes the consistency of observation sequences, and reduces the key disagreement rate. The theoretical deduction proves the validity of the proposed method under an underwater acoustic multipath time-varying channel. Also, the simulation results verify the feasibility of the proposed method.

Restricted Boltzmann machines (RBMs) are commonly used as pre-training methods for deep learning models. Contrastive divergence (CD) and parallel tempering (PT) are traditional training algorithms of RBMs. However, these two algorithms have shortcomings in processing high-dimensional and complex data. In particular, the number of temperature chains in PT has a significant impact on the training effect, and the PT algorithm cannot fully utilize parallel sampling from multiple temperature chains for the divergence of the algorithm. The training can quickly converge with fewer temperature chains, but this impacts the accuracy. More temperature chains can help PT achieve higher accuracy in theory, but severe divergence at the beginning of the training may ruin the training result. To exploit fully the advantages of PT and improve the ability of RBMs to process high-dimensional and complex models, this article proposes dynamic tempering chains (DTC). By dynamically changing the number of temperature chains during the training process, DTC starts training with fewer temperature chains and gradually increase the number of temperature chains with training going on, and finally get an accurate RBM. And one-step reconstruction error is proposed to measure the convergence, which can decrease the influence of the dynamic training strategy on reconstruction error. Experiments on MNIST, MNORB, Cifar 10, and Cifar 100 indicate that, compared with PT, the classification accuracy of DTC algorithm improved by up to 8%. DTC quickly converges in the early stage of training because of few exchanges among temperature chains and produces higher accuracy at the end for the global optimum model learned by more temperature chains, especially when learning high-dimensional and complex data. This proves that the DTC algorithm effectively utilizes parallel sampling of multiple temperature chains, overcomes divergence challenges, and further improves the training effect of the RBM.

Deep learning (DL) has been widely applied in fault diagnosis field. The ensemble of DL can significantly improve the generalization ability of DL. Snapshot ensemble learning, which uses the cyclic learning rate scheduler (CLR) and then combines the local minima in once training to form an ensemble method, has shown powerful performance. However, the learning rate range of CLR should be pre-defined by experiences, which may limit its prediction accuracy. In this paper, a new snapshot ensemble convolutional neural network (SECNN) is proposed, which can find the proper range of learning rate for SECNN automatically when facing a new dataset. First, a max-min cosine cyclic learning rate scheduler (MMCCLR) is designed to avoid learning rate range being affected by other parameters. Then, a new learning rate testing (logLR Test) is applied to estimate the proper learning rate range for MMCCLR. Finally, the SECNN with MMCCLR is developed. The SECNN is tested on three famous datasets, including bearing dataset of Case Western Reserve University, self-priming centrifugal pump dataset, and bearing dataset provided by the Machinery Failure Prevention Technology. The results have been improved a lot by the proposed SECNN methods. The SECNN has also achieved the start-of-the-art prediction accuracies by comparing with other DL and traditional methods.

The output voltage of a dual LCL inductive power transfer (IPT) system varies with the load and mutual inductance, which can affect the system performance, e.g., the charging speed of electrical equipment. This paper proposes a robust controller to maintain constant output voltage against the load and mutual inductance variations. A μ-synthesis method is proposed based on structured singular value to design the dual LCL IPT closed-loop system. The frequency-domain nominal model of the system is established by using the generalized state-space averaging (GSSA) method. A standard M-Δ configuration is generalized through separating the nominal and uncertain block by upper linear fractional transformation (LFT). The μ-controller is finally obtained by D-K iteration. The robust performance analysis and time-domain performance analysis are also carried out. The simulation and experimental results verified the effectiveness of the proposed modeling and controller design, and demonstrated the robust stability and robust performance of the dual LCL IPT system.

Point cloud is a collection of many unordered points. Deep learning network encounters difficulties in utilizing the local information of point cloud because of its irregular format. This is not conducive to the network to identify the details of the object. Some strategies that design new network structures are used to capture the local information, but they make the networks become complicated. This paper proposes an effective method by encoding points into feature vectors. The local information is represented using neighborhood points that are searched by the k-nearest neighbor method. The neighborhood points are converted into feature elements based on the distance between these points and the encoded point. The feature vector of each point consists of its coordinates and the corresponding feature elements. A simple deep learning network is used to process these feature vectors. The proposed method is applied to ModelNet40 shape classification benchmark. The experimental results show that the classification accuracy of the simple deep learning network is improved by the proposed method.

Fiber-optic distributed acoustic sensing (FDAS) with phase-sensitive optical time-domain reflectometry (Φ-OTDR) is a promising technique for high-sensitivity measurement. In this paper, an improved Φ-OTDR system with a weak fiber Bragg grating (wFBG) array for partial discharge (PD) detection in cross-linked polyethylene (XLPE) power cables is demonstrated; and an event recognition method based on a convolutional neural network (CNN) model is proposed to identify and classify different types of events, including internal PD, corona PD, surface PD, and noise. A multiscale wavelet decomposition and reconstruction method is used to extract PD signals and a two-dimensional spectral frame representation of the PD signals is obtained by the mel-frequency cepstrum coefficients (MFCC). The experimental results based on 1280 training samples and 832 test samples have demonstrated high values of precision, sensitivity, and specificity for each event (up to 96.3%, 96.4%, and 98.7%, respectively), which means that the combination of multiscale wavelet decomposition and reconstruction, the MFCC and CNN may be a promising event recognition method for the FDAS systems.

Recently, dynamic business-IT alignment (BITA) has attracted increasing attentions. However, the description and the governance of dynamic BITA are insufficient in the literature. To address this gap, a conceptual BITA evolvement model is proposed combining enterprise architecture with the BITA maturity framework, and a sustainable BITA factor on dynamic resource allocation is introduced in this evolvement model. To discuss the sustainable factor in a comprehensive way, we explored a series of possible changes and corresponding resource influences within the alignment of business functions and applications. A dynamic resource allocation algorithm is proposed, and an illustrative case study is presented to verify the algorithm's efficiency.

To improve the prediction accuracy of monocular depth estimation networks and address issues such as edge blurring and excessive artifacts in the generated depth maps, this paper proposes a deep network architecture based on a global-local feature fusion module and a residual pooling module. The encoder utilizes a Hierarchical Transformer, while the decoder incorporates a U-Net structure model that combines multi-dimensional attention features aggregation and residual pooling. The residual pooling module facilitates better extraction of background contextual information from the feature maps to obtain more precise scene depth information. The global-local feature fusion module enables the network to learn features that encompass both global and local information. Experimental evaluations conducted on the NYU Depth V2 and KITTI datasets demonstrate that the proposed method achieves a $\delta 1$ of 0.916 on the NYU Depth V2 dataset, along with enhanced generalization ability and robustness. Furthermore, the effectiveness of each module is validated through ablation studies on the NYU Depth V2 dataset.

Patterning is a crucial fabrication step for successfully applying two-dimensional materials in electronic and optoelectronic devices. It can realize miniaturization and help explore new physical phenomena of 2D materials. However, the manufacturing process inevitably introduces defects, which require harsh conditions to recover. Here, we propose a sputtering-lithography-annealing (SLA) strategy for patterning graphene nanofilm with pattern sizes ranging from microns to 100 nm scale without lattice damage. The sputtered masking agents can introduce easily repairable defects into graphene films. Especially, defects introduced by aluminum can be removed entirely. To confirm the validity of the SLA strategy, we prepared macro-assembled graphene nanofilms (nMAG)/Ge and nMAG/Si heterojunction arrays for infrared detection. The patterned detectors present a responsivity of 0.09 A/W at $2 \mu \text{m}$ (nMAG/Ge) and 26.4 mA/W at 1550 nm (nMAG/Si) with a high array homogeneity, similar to the devices without patterning. This strategy lays the foundation for further exploration of new superstructures of nMAG and can be extended to other 2D materials.

The air pollution problem has been a serious issue for public health and city development in recent years, which rises an urgent demand for accurate air pollution prediction models. Traditional time series prediction scheme has been challenged in such case because the air pollution signals can be highly dynamic and uncertain. Distinct latent physical processes and complex environment changes over time make it hard for one fixed model to give consistently good performance. Also probabilistic prediction and uncertainty-aware estimation are critical when the model is used for public decisions in practice. However, few previous works can simultaneously provide adaptive and probabilistic air pollution prediction abilities, which are important practical issues for the real-world implementation of data-driven air pollution prediction models. In this paper we propose an adversarial meta learning approach to address these concerns. After reformulating the target signals as a collection of independent but similar mini-tasks along time, we adversarially learn a meta model that includes an implicit generator to adaptively give more informative probabilistic predictions across tasks. We further provide a bayesian meta learning interpretation that recasts the proposed model as an approximate minimizer for the Wasserstein distance between a generative latent model and the true data distribution. Our model provides an algorithm that explicitly models how the prediction distribution is conditioned on underlying data patterns and simultaneously gives adaptive uncertainty estimation. Experiments on both synthetic and real-world air pollution datasets show that the proposed model can simultaneously provide better probabilistic and adaptive prediction while keeping stratifying point prediction error.

Fine-grained air quality can provide essential urban environmental information for administrators and residents. With advances in communication and sensing technologies, low-cost portable sensors installed on vehicles enable high-coverage air quality monitoring. However, data collected by low-cost mobile sensors may be inaccurate and inconsistent in complex operation environments, which brings the issue of data uncertainty. Moreover, due to uncontrolled vehicles and human activities, the coverage of mobile nodes is dynamic over time, leading to uneven or sparse spatial distribution. To address these challenges, we propose AQI-M3, a novel framework for fine-grained air quality inference via multi-view learning with mobile sensing memory. Specifically, an encoder-decoder structure is applied in the region view for modeling the spatial dependencies in pollution maps. More importantly, sensing gradients are extracted in the trajectory view to enable the utilization of uncertain mobile sensing data. In addition, a memory network is designed to capture the spatial patterns from the historical sensing data and provide the global patterns as a complemental guide to overcome dynamic coverage sampling. Extensive experiments are conducted on three real-world deployments of hybrid sensing systems with both static and mobile sensors. Experimental results show that our proposed approach outperforms competitive baselines with 17% $\sim 29$ % reduction in mean absolute error. Furthermore, detailed evaluations demonstrate the effectiveness and robustness of the proposed framework under dynamic coverage.

In view of the difficulties in optimal dispatch caused by diverse energy and various types of equipment in the regional integrated energy system(RIES), this paper established a generalized energy storage model by extracting the flexibility characteristics of cooling, heat, and electrical load based on abstract equivalence. Instead of reducing the accuracy of regulation, this modeling method can greatly simplify the information interaction between the energy dispatching center and energy consumers, thus cutting down the time of collaborative analysis calculations. With the scheduling characteristics of multiple types of flexible loads in day-ahead and intra-day time scales considered comprehensively, a multi-time-scale optimization scheduling model of RIES is established, which takes into account the economy of system operation and the demand response. The simulation results show that the proposed optimization model can fully tap the load-side scheduling potential, along with effectively reducing the computational complexity. Through the coordination of resources with different characteristics, the pressure of balancing energy supply and demand can be relieved, and the operating cost of the system can be minimized.

Aiming at the low frequency oscillation and voltage drop caused by the wind-thermal bundled system transmitted to the load center across AC and DC lines, an adaptive sliding mode gain control method based on the structure decentralized control theory is proposed. On the basis of this method, a rotor-side adaptive terminal sliding mode controller for doubly-fed induction generator based on synchronous rotating coordinate system is designed. Based on RTDS, a real-time simulation platform for transient stability control of “wind-thermal bundled” transmission system was developed, and real-time simulation experiments were performed. Experimental results prove that the proposed method in this paper can effectively suppresses the frequency and voltage fluctuations of the system, significantly improve the transient stability of the system and enhance the robustness of the system during grid faults.

Insulator fault in the transmission lines is the main factor of power transmission accident. The images captured from the aerial inspection can be utilized to detect the fault of insulators for further maintenance. For automatic transmission lines inspection system, the insulator fault detection is an interesting and challenging task due to the complex background and diversified insulators. In this paper, we propose a novel insulator fault detection method based on multi-level perception for aerial images. The multi-level perception is implemented by an ensemble architecture which combines three single-level perceptions. These single-level perceptions include the low level, middle level, and high level that are named by the attention to the insulator fault. They detect the insulator fault in the entire image, multi-insulator image, and single-insulator image, respectively. To address the filtering problem in the combination of three single-level perceptions, an ensemble method is proposed for generating the final results. For training the detection models employed in the multi-level perception, a powerful deep meta-architecture so-called single shot multibox detector (SSD) is utilized. The well-trained SSD models can automatically extract high quality features from aerial images instead of manually extracting features. By using the multi-level perception, the advantages of global and local information can achieve a favorable balance. Moreover, limited inspection images are fully utilized by the proposed method. Fault detection recall and precision of the proposed method are 93.69% and 91.23% testing in the practical inspection data, and insulator fault under various conditions can be correctly detected in the aerial images. The experimental results show that the proposed method can enhance the accuracy and robustness significantly.

The detection of insulators with cluttered backgrounds in aerial images is a challenging task for an automatic transmission line inspection system. In this paper, we propose an effective and reliable insulator detection method based on a deep learning technique for aerial images. In the proposed deep detection approach, the single shot multibox detector (SSD), a powerful deep meta-architecture, is incorporated with a strategy of two-stage fine-tuning. The SSD-based model can realize automatic multi-level feature extractor from aerial images instead of manually extracting features. Inspired by transfer learning, a two-stage fine-tuning strategy is implemented using separate training sets. In the first stage, the basic insulator model is obtained by fine-tuning the COCO model with aerial images, including different types of insulators and various backgrounds. In the second stage, the basic model is fine-tuned by the training sets of the specific insulator types and specific situations to be detected. After the two-stage fine-tuning, the well-trained SSD model can directly and accurately identify the insulator by feeding the aerial images. The results show that both the porcelain insulator and composite insulator can be quickly and accurately identified in the aerial images with complex background. The proposed approach can enhance the accuracy, efficiency, and robustness significantly.

Aiming at the Low voltage ride through capability of doubly-fed wind farm when it is merged into a weak grid, this paper researches the theory of rotor excitation control and proposed an additional control model of DFIG grid-connected rotor flux based on structure decentralization theory. In this model, the Low voltage ride through capability of the doubly-fed wind farm is enhanced by designing the d-axis and q-axis adaptive terminal sliding mode controllers for the synchronous rotating coordinate system. Simulation results show that the proposed control model not only can better realize the rapid regulation of reactive power of doubly-fed wind farms, but also improve the ability of the system to resist grid voltage fluctuations, and improve the transient stability of power systems. It has certain theoretical significance and practical application value.

The unmanned surface vehicle (USV) is usually required to perform some tasks with the help of static and dynamic environmental information obtained from different detective systems such as shipborne radar, electronic chart, and AIS system. The essential requirement for USV is safe when suffered an emergency during the task. However, it has been proved to be difficult as maritime traffic is becoming more and more complex. Consequently, path planning and collision avoidance of USV has become a hot research topic in recent year. This paper focuses on dynamic obstacle avoidance and path planning problem of USV based on the Ant Colony Algorithm (ACA) and the Clustering Algorithm (CA) to construct an auto-obstacle avoidance method which is suitable for the complicated maritime environment. In the improved ant colony-clustering algorithm proposed here, a suitable searching range is chosen automatically by using the clustering algorithm matched to different environmental complexities, which can make full use of the limited computing resources of the USV and improve the path planning performances firstly. Second, the dynamic searching path is regulated and smoothed by the maneuvering rules of USV and the smoothing mechanism respectively, which can effectively reduce the path length and the cumulative turning angle. Finally, a simulation example is provided to show that our proposed algorithm can find suitable searching range according to different obstacle distributions, as well as accomplish path planning with good self-adaptability. Therefore, a safe dynamic global path with better optimize performances is achieved with the help of multi-source information.

Aiming at the problem of high voltage disconnection when doubly-fed wind farms are merged into weak power grids, this paper researches the theory of rotor excitation control and proposed an additional control model of DFIG grid-connected rotor flux based on structure decentralization theory. In this model, the high voltage ride-through capability for grid-connected wind farms is enhanced by designing the d-axis and q-axis adaptive terminal sliding mode controllers of the synchronous rotating coordinate system. Using the simulation software of MATLAB/Simulink, this paper established a simulation model of wind farm high voltage ride through composed of 2MW doubly fed wind turbines and carried out the off-line simulation of the whole process of high voltage traversing. Real-time simulation experiment of high voltage ride-through was also conducted on a self-developed real-time simulation platform for grid-connected doubly-fed wind farms. Real-time simulation results prove the accuracy of theoretical and offline simulation analysis and the feasibility of control strategy.

In this paper, we study a cooperative network-coding (NC) media-access control for the two-way two-hop relay network specified with several helpers besides a default relay node. The idle helpers can help forward relay packets. The considered protocol focuses on the channel-based time-division multiple access (TDMA) principle and is called a two-way network-coding-based cooperative relaying TDMA (TW-NCCR) protocol. As the main contribution, we analytically derive the throughput performance of TW-NCCR, and further verify it by simulation. Then, we compare it with a two-way cooperative relaying TDMA (TW-CR) protocol, which is the counterpart of TW-NCCR but applies no network coding. Both the theoretical analysis and simulation results demonstrate that TW-NCCR provides superior throughput performance compared with TW-CR, especially when the two-way relay network is symmetric. On the other hand, no matter how weak they are, helpers are able to improve the performance of asymmetric networks using TW-NCCR, especially when the non-reciprocation packet generation rate of the relay node is small.