In the modern machine tool manufacturing scene, as the milling tool of CNC machine tool, the health of the tool directly affects the processing efficiency and product quality. Effective prognostic and health management of tool is critical. Precise monitoring of tool wear helps to avoid product quality problems caused by tool fault and improve production efficiency. Therefore, this paper constructs a tool fault diagnosis method based on deep learning. In order to effectively fuse the vibration signals features from different directions of machine tool spindle, we apply a variety of Channel Attention (CA) mechanisms to Multiscale Network (MSNet) to construct Multiscale-Channel Attention Network (MS-CA Net), and explore the performance gains of these modules in tool wear classification tasks. Among these modules, the CA blocks include Channel Attention Block (CAB), Squeeze and Excitation Block (SEB), and Efficient Channel Attention Block (ECAB). We use the three proposed networks to classify the tool wear status by identifying the vibration signal of the machine tool spindle. At the same time, in order to verify the performance of the proposed tool fault diagnosis method in the actual milling scene, this paper designs a tool wear test platform to collect sample data that meets actual industrial production scenarios, which uses a three-axis accelerometer to collect tool life cycle vibration signals, and a digital universal tool microscope to measure tool wear values. The experimental results show that, compared with the MSNet method, the tool fault diagnosis accuracy rate of the improved method is increased by 4.47%-7.38%.

Unmanned Aerial Vehicles (UAVs) communications appear to be one of the most promising paradigms for future wireless communication networks, because of their high flexibility in providing on-demand communication services. In this context, this paper investigates a fast and fine-grained UAV deployment scheme so as to improve the network throughput and meet the real-time communication demands of users. The key novelty of the proposed scheme lies in that the UAV deployment problem is formulated as a computer vision problem and a novel UAV deployment method, i.e., a convolutional neural network (CNN)-based UAV deployment method is proposed to solve it. By taking advantage of the classical CNN models such as VGG-Net, AlexNet, the UAV deployment position can be determined timely. Compared with the existing work, this work not only reduces the computational time overhead for determining the deployment positions of UAV, but also shortens the deployment response time of UAV. The superiority of the proposed UAV deployment scheme is investigated and verified. Simulation results demonstrate that the proposed scheme can provide a fast and on-demand UAV deployment solution while guaranteeing the throughput performance of the UAV network.

This study addresses the inefficiencies in how idle taxis determine their cruising strategy, which currently rely heavily on drivers' personal experiences. Such reliance often leads to inefficient roaming and delays in service. We propose a novel strategy for managing a small scale taxi-fleet, which is common for alliance business, utilizing real-time data and multi-agent reinforcement learning to predict potential travel demands and strategically direct idle taxis to high-demand zones. This method aims to optimize the utilization of idle taxis, reduce passenger wait times, and enhance the ride-hailing ecosystem's overall efficiency. Our approach integrates a spatial-temporal service request model with the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm. It innovates further by employing varied reward mechanisms informed by actual order data and the fluctuating of supply and demand, which does not only enhance the training efficiency in complex scenarios but also captures the dynamics across the service area, effectively managing supply-demand imbalances. Experimental results demonstrate that our strategy significantly outperforms several state-of-the-art methods, representing a major advancement in the optimization of transportation services. This study not only provides a more efficient framework for managing idle taxi fleets but also offers insightful implications for the future enhancement of the transportation sector.

In this paper, we propose an Industrial Semantic Communications (ISC) framework to ensure efficient and in-telligent communication in Industrial Internet-of-Things (IIoT) networks, and further apply the proposed framework to the task of tool wear monitoring. A semantic-aware tool wear monitoring network (TWM-Net) is designed to accurately identify the idling state from tool wear data, thus assisting the semantic data transmission process of the ISC framework. Then, a demo software system is implemented to verify the delay performance of the proposed ISC framework. Experimental results demonstrated that the proposed ISC framework can facilitate the semantic information exchange between edge industrial system and cloud platform together with the deep compressed sensing method, and significantly reduce the amount of data transmissions in IIoT networks. The source code and dataset is available through https://github.com/pnixsoft/TWM-Net.

This paper considers smooth strongly convex and strongly concave stochastic saddle point (SSP) problems. Suppose there is an arbitrary oracle that in expectation returns an ϵ -solution in the sense of certain gaps, which can be the duality gap or its weaker variants. We propose a general PB-SSP framework to guarantee an ϵ small duality gap solution with high probability via only 𝒪 (log1/p·poly(logκ ) ) calls of this oracle, where p∈ (0,1) is the confidence level and κ is the condition number. When applied to the sample average approximation (SAA) oracle, in addition to equipping the solution with high probability, our approach even improves the sample complexity by a factor of poly(κ ) , since the high-probability argument enables us to circumvent some key difficulties of the uniform stability analysis of SAA.

Edge content caching has emerged as a promising solution against network latency by pre-caching popular contents at the edge of networks. However, the edge content caching and wireless resource allocation are intertwined with each other. The content downloading latency depends not only on the edge content caching strategy, but also highly on the wireless resource allocation strategy. To cope with them as a whole, this work aims at jointly optimizing the edge content caching and power allocation to minimize the content downloading latency. In doing so, a joint edge content caching and power allocation problem is formulated. But, this problem is NP-hard and difficult to be solved. Solving it with the traditional approaches will cause quite long computational delay, which cannot meet the real-time requirements of resource scheduling. To address this challenge, a novel deep learning (DL)-enabled joint edge content caching and power allocation framework is proposed. In particular, the formulated problem is firstly transformed into a classification problem in DL field. After that, the convolutional long short-term memory networks (ConvLSTM) are utilized to capture the temporal-spatial features of content requests, and the fully-connected networks (FC) are introduced to explore the users' location differences. Finally, the temporal-spatial features of content requests and the location differences are fused to make intelligent decisions for joint edge content caching and power allocation. Simulation results show that the proposed joint edge content caching and power allocation strategy outperforms the state-of-the-art baselines at about 2.3% in terms of the average content downloading latency while guaranteeing real-time resource scheduling.

Edge caching has emerged as a promising technique against latency caused by explosive growth of mobile data traffic through caching popular contents at the edge networks. However, the dynamically changing content popularity nature and limited caching capacity make it challenging to design an effective caching scheme to reduce latency. To solve this, a learning-based hierarchical edge caching (LHEC) scheme is proposed in this work. We first propose a novel deep learning architecture, namely Stacked Autoencoder-Long Short Term Memory Network (SAE-LSTMNet) to capture both the correlation of the request patterns among different content and the periodicity in time domain to improve the prediction accuracy of the content popularity. Then, to predict the popularity of these newly-added contents, a dynamic content catalog is introduced and a similarity-based content popularity prediction (SCPP) approach is proposed. Based on the content popularity prediction, a hierarchical edge caching optimization problem is formulated to minimize the average content downloading latency. Since the formulated problem is NP-hard and difficult to be solved, a low-complexity algorithm is proposed to obtain the near-optimal solutions. Simulation results show that the proposed content popularity prediction approach outperforms up to 6.36% in terms of the mean absolute error compared with the state-of-the-art methods and the proposed LHEC scheme reduces the average downloading latency at about 5.3%~7.9% compared with those existing caching schemes.

This work investigates proactive edge caching for device-to-device (D2D)-assisted wireless networks, where user equipment (UE) can be selected as caching nodes to assist content delivery to reduce the content transmission latency. In doing so, there are two challenges: 1) how to precisely get the user’s preference to cache the proper contents at UEs and 2) how to replace the contents cached at UEs when there are new popular contents emerging. To address these, we develop a user preference learning-based proactive edge caching (UPL-PEC) strategy. In the strategy, we first propose a novel context and social-aware user preference learning method to precisely predict user’s dynamic preferences by jointly exploiting the context correlation among different contents, the influence of social relationships and the time-sequential patterns of user’s content requests. Specifically, the bidirectional long short-term memory networks are adopted to capture the time-sequential patterns of the user’s content requests. And, the graph convolutional networks are developed to capture the high-order similarity representation among different contents from the constructed content graph. To learn the social influence representation, an attention mechanism is designed to generate the social influence weights to users with different social relationship. Based on the learned user preference, a proactive edge caching architecture is proposed to integrate the offline caching content placement and the online caching content replacement policy to continuously cache the popular contents at UEs. Simulation results show that the proposed UPL-PEC strategy outperforms the existing similar caching strategies at about 3.13%–4.62% in terms of the average content transmission latency.

As Transformer-based models have achieved impressive performance on various time series tasks, Long-Term Series Forecasting (LTSF) tasks have also received extensive attention in recent years. However, due to the inherent computational complexity and long sequences demanding of Transformer-based methods, its application on LTSF tasks still has two major issues that need to be further investigated: 1) Whether the sparse attention mechanism designed by these methods actually reduce the running time on real devices; 2) Whether these models need extra long input sequences to guarantee their performance? The answers given in this paper are negative. Therefore, to better copy with these two issues, we design a lightweight Period-Attention mechanism (Periodformer), which renovates the aggregation of long-term subseries via explicit periodicity and short-term subseries via built-in proximity. Meanwhile, a gating mechanism is embedded into Periodformer to regulate the influence of the attention module on the prediction results. Furthermore, to take full advantage of GPUs for fast hyperparameter optimization (e.g., finding the suitable input length), a Multi-GPU Asynchronous parallel algorithm based on Bayesian Optimization (MABO) is presented. MABO allocates a process to each GPU via a queue mechanism, and then creates multiple trials at a time for asynchronous parallel search, which greatly reduces the search time. Compared with the state-of-the-art methods, the prediction error of Periodformer reduced by 13% and 26% for multivariate and univariate forecasting, respectively. In addition, MABO reduces the average search time by 46% while finding better hyperparameters. As a conclusion, this paper indicates that LTSF may not need complex attention and extra long input sequences. The code has been open sourced on Github.

为提高机床刀具磨损故障诊断精度,借助三轴加速度传感器采集机床刀具磨损振动数据,并以此为基础,提出一种基于长短时记忆网络与多尺度卷积神经网络集成的机床刀具故障诊断算法,挖掘机床刀具在不同故障模式下的空时域多尺度特征,实现机床刀具磨损状态的精准识别.同时,为满足实际产线对故障诊断的高实时性要求,设计一种适用于机床刀具故障诊断的云边端协同架构,通过底层产线、边缘节点与工业云平台的高效协同,实现机床刀具故障的及时预警.研究结果表明,所提云边端协同的机床刀具故障智能诊断系统可实现机床刀具磨损状态的精准识别,同时可缩短故障诊断任务的完成时间.

In an era of rapid technological advancements, keeping up with the iterative updates of factory technology in Intelligent manufacturing presents a daunting challenge. Against this backdrop, this paper investigates the performance of the ECA-YOLOv5n model, a novel approach for appearance quality detection in air conditioner external units. Building upon an open-source dataset, ECA-YOLOv5n model integrates the ECA-Net attention mechanism into the YOLOv5-nano model. ECA-Net effectively captures channel-wise dependencies and studies discriminative features, allowing for seamless integration into the convolutional neural networks of the YOLOv5 architecture without the need for extensive modifications. Furthermore, the efficient channel attention mechanism of ECA-Net demands fewer parameters and computations in comparison to other attention mechanisms, conferring a significant advantage in terms of computational efficiency. Experimental results show that the newly proposed model achieves approximately 98% precision, 99% recall, and 99% mAP on the dataset. The ECA-YOLOv5n model reduces storage usage and minorly increases detection speed in comparison to other YOLO models.

This work investigates proactive edge caching for D2D-assisted wireless networks, where user equipments (UEs) can be selected as caching nodes to assist content delivery. The objective of this work is to achieve a trade-off between the cost for providing caching services and the content transmission latency. Doing so, there are two challenges: 1) Which UEs can be selected as caching nodes; 2) How to place contents on these selected UEs without user’s privacy disclosure. To address these, a novel community detection and attention-weighted federated learning based proactive edge caching (CAFLPC) strategy is proposed. In the strategy, we first group UEs into different communities based on both the mobility and social properties of UEs, and then select important users (IUs) as caching nodes for each community by considering the social importance of UEs. To determine how to place the popular contents in these selected IUs, an attention-weighted federated learning (AWFL) based content popularity prediction framework is proposed. It integrates the attention-weighted federated learning with Bidirectional Long Short Term Memory Network (AWFL_BiLSTM) to achieve a higher content popularity prediction accuracy while protecting user’s privacy. Considering the imbalance of UEs’ active levels and local computing capacities, an attention-weighted aggregation mechanism is proposed to improve the training efficiency and prediction accuracy. Simulations results show that the proposed CAFLPC strategy outperforms the compared existing caching strategies at about 2.2%-35.1% in terms of the transmission latency reduced by per unit cost.

Digital simulation is a prevalent tool to evaluate the performance of complex industrial systems. In this paper, we consider the discrete optimization via simulation (DOvS) problems, where the design space is an integer lattice. In addition, we are interested in leveraging cheap low-fidelity models to enhance optimization efficiency. An innovative Gaussian Markov random field (GMRF), which adaptively evolves along with the optimization process, is introduced to exploit the spatial and inter-model relationships among the objective function values of designs. We then propose the Multi-Fidelity Gaussian Markov Improvement Algorithm (MFGMIA) under the Bayesian global optimization framework. The numerical experiments show that it can achieve significant performance improvement by properly using low-fidelity information.

刀具故障预测与健康管理(Prognostic and health management,PHM)是机床制造领域的关键问题.作为数控机床的"牙齿",刀具的健康状态直接影响着机床加工效率和产品质量.借助大数据与人工智能(AI)技术实现对刀具运行状态的实时精准监测,目前已成为学术界和工业界关注的热点问题.然而,刀具高质量全寿命周期数据的匮乏,严重制约了机械装备PHM技术的理论研究与工程应用.为解决上述问题,开展了面向刀具全寿命周期的数控加工中心立铣刀多工况试验与数据采集工作,并将获取的试验数据集面向全球学者公开发布.该数据集共包含2种工况下的立铣刀全寿命周期振动信号,且明确标注了刀具主后刀面最大VB值、主后刀面1/2ap(背吃刀量)处VB值、主后刀面SVB值,副后刀面最大VB值、副后刀面SVB值等5种标签,可为PHM领域基于AI的刀具故障诊断与预测性维护研究提供数据支撑.

With the development of Industrial Internet of Things (IIoT), the ever growing mismatch between the numerous tasks generated in real industrial scenarios and the limited computing ability is enlarging the system delay. How to schedule the system tasks to enhance the system efficiency has become extremely significant. Along this line, a task scheduling scheme for MEC-enabled IIoT systems is proposed in this work to minimize the total delay of the whole system. Since there exist time overlaps between different tasks while the tasks are conducted in parallel, it is difficult to accurately model the process of the tasks in terms of delay. To solve this, a novel modeling method is proposed to transform the optimization problem of minimizing the total delay into minimizing the unprocessed data volumes. To solve the formulated problem, we formulate the process of the tasks being executed as a Markov decision process (MDP) and propose a modified deep reinforcement learning (DRL) algorithm. To evaluate the performance of our proposed task scheduling scheme, intensive experiments have been conducted. The results show that our proposed scheme achieves better performance than some existing schemes. In the end, the scalability and availability of our scheme are tested.

As the “teeth” of CNC machine tools, the tool health status directly affects the machining efficiency and the quality of products. However, due to the complex and variable process of tool wear, the tool remaining useful life is difficult to predict. To address the above challenges, we investigate a deep learning-based tool remaining useful life prediction method in this work. First, the time-frequency features of the tool monitoring signal are extracted and fused using time-frequency analysis and wavelet transform methods. And then, a long short-term memory network based tool remaining useful life prediction method is proposed to accurately predict the remaining tool useful life, which can guide the update and optimization of equipment maintenance plans, thus effectively improving the productivity of enterprises, reducing production costs and ensuring product quality. To evaluate the performance of the proposed method, experiments are conducted based on the open dataset of tool health prediction for high-speed CNC machine tools. The experimental results show that the proposed tool remaining useful prediction method based on time-frequency features fusion and long and short-term memory network performs the best prediction performance in terms of mean absolute error, mean squared error, R-squared and other performance indicators compared with the existing classical methods.

Prognostic and health management is a key issue in the field of machine tool manufacturing. As the “teeth” of CNC machine tools, their health status directly affects the machining efficiency and the quality of products. Accurate monitoring of tool wear can help to avoid product quality problems caused by tool failure and improve productivity. In this paper, we investigate the deep learning-based tool fault diagnosis approach. First, a new data-driven tool fault diagnosis method based on improved multiscale network and feature fusion (IMSNet-F) is proposed to recognize and classify the tool wear condition. It can increase the efficiency of the process and make it possible to replace the tool before catastrophic wear occurs. And then, a tool wear experimental system is designed to verify the performance of the proposed tool fault diagnosis method in a real production scenario. Besides, based on the tool wear experimental system, a data set of vibration signals used to detect tool wear conditions is constructed and publicly released. Experimental results show the proposed approach can improve the tool fault diagnosis accuracy by 2.2% compared to existing methods.

Edge caching has been explored as an effective technology to alleviate the heavy traffic burden of the backhaul and avoid transmission congestion in vehicular networks. However, high mobility of vehicles could lead to repetitive content caching, resulting in high system cost. Because content popularity changes very frequently in vehicular networks, to provide better service for vehicle users, it is essential to update content frequently. This leads to expensive update cost at the same time. To reduce such cost, we propose a mobility-aware cost effective edge caching strategy, in which vehicle mobility, file encoding technology and dynamic content popularity are jointly taken into consideration. To reduce the complexity of formulated problem, deep reinforcement learning (DRL) approach is adopted. Simulation results show that the proposed mobility-aware coded edge caching strategy can dramatically reduce the system cost (up to 36% compared with classic caching algorithm).

Wireless traffic prediction has drawn increasing research interests as it can provide guidance to the network optimization. With the predicted information, one can preassign the resources on demand and perform network congestion control adaptively. The network efficiency is therefore enhanced. However, the wireless traffic prediction in the context of mobile scenario, such as Internet of Vehicles (IoVs), is still a challenge issue. The mobile nature of devices, which dynamically changes the topology of network, would brings difficulties to the prediction. This paper focuses on the deep learning based wireless traffic prediction in the IoVs scenario. We first propose a novel method to match up the movement- and communication-behavior of users, by merging two independent datasets on the trajectories of vehicles and communication traffic volumes together. Then a novel STeP-UNet is proposed, in which the SpatioTemporal Partial (STeP) Convolutional Neural Network module is embedded to capture cross-domain features of the wireless traffic pattern, and the UNet structure is utilized to realize the skipping connection from front layer to back layer to fuse different resolutions. Experimental results confirms the promising performance of the proposed model, where 4%~8% performance improvement over other benchmark methods can be achieved.

Heterogeneous network (HetNet) is a promising solution to satisfy the unprecedented demand for higher data rate in the next generation mobile networks. Different from the traditional single-layer cellular networks, how to provide the best service to the user equipments (UEs) under the limited resource is an urgent problem to solve. In order to efficiently address the above challenge and strive towards high network energy efficiency, the joint optimization problem of user association and power control in orthogonal frequency division multiple access (OFDMA) based uplink HetNets is studied. Considering the non-convex and non-linear characteristics of the problem, a multi-agent deep Q-learning Network (DQN) method is studied to solve the problem. Different from the traditional methods, such as game theory, fractional programming and convex optimization, which need more and accurate network information in practice, the multi-agent DQN method requires less communication information of the environment. Moreover, for the communication environment dynamics, the maximum long-term overall network utility with a new reward function while ensuring the UE’s quality of service (QoS) requirements is achieved by using the multi-agent DQN method. Then, according to the application scenario, the action space, state space and reward function of the multi-agent DQN based framework are redefined and formulated. Simulation results demonstrate that the multi-agent DQN method has the best performance on convergence and energy efficiency compared with the traditional reinforcement learning (Q-learning).