In recent years, artificial intelligence (AI) has achieved significant progress and remarkable advancements across various disciplines, including biology, computer science, and industry. However, the increasing complexity of AI network structures and the vast number of associated parameters impose substantial computational and storage demands, severely limiting the practical deployment of these models on resource-constrained edge devices. Although edge intelligence methods have been proposed to alleviate the computational and storage burdens, they still face multiple persistent challenges, such as large-scale model deployment, poor interpretability, privacy and security vulnerabilities, and energy efficiency constraints. This article systematically reviews the current advancements in edge intelligence technologies, highlights key enabling techniques including model sparsity, quantization, knowledge distillation, neural architecture search, and federated learning, and explores their applications in industrial, automotive, healthcare, and consumer domains. Furthermore, this paper presents a comparative analysis of these techniques, summarizes major trade-offs, and proposes decision frameworks to guide deployment strategies under different scenarios. Finally, it discusses future research directions to address the remaining technical bottlenecks and promote the practical and sustainable development of edge intelligence. Standing at the threshold of an exciting new era, we believe edge intelligence will play an increasingly critical role in transforming industries and enabling ubiquitous intelligent services.

Time sensitive networks support deterministic schedules over Ethernet networks. Due to their high determinism, high reliability and high bandwidth, they have been considered as a good choice for the backbone network of industrial internet of things. In industrial applications, the backbone network connects multiple industrial field networks together and has to carry massive real-time packets. However, the off-the-shelf time-sensitive network (TSN) switches can deterministically schedule no more than 1024 real-time flows due to the limited number of schedule table entries. The excess real-time flows have to be delivered by best-effort services because the switch only supports the two scheduling services. The best-effort services can reduce average delay, but cannot guarantee the hard real-time constraints of industrial applications. To make the limited number of schedule table entries support more real-time flows, first, we relax scheduling rules to reduce the requirement for schedule table entries and formulate the process of transmitting packets as a satisfiability modulo theories (SMT) specification. Then, we divide the SMT specification into multiple optimization modulo theories (OMT) specifications so that the execution time of solvers can be reduced to an acceptable range. Second, we propose fast heuristic algorithms that combine schedule tables and packet injection control to eliminate scheduling conflicts. Finally, we conduct extensive evaluations. The evaluation results indicate that, compared to existing algorithms, our proposed algorithm requires only one-twentieth the number of schedule entries to schedule the same flow set.

Time-Sensitive Networking (TSN) provides end-to-end data transmission with extremely low delay and high reliability on the basis of Ethernet. It is suitable for time-sensitive applications and will be widely used in scenarios such as autonomous driving and industrial Internet. IEEE 802.1Qbv proposes a time-aware shaper mechanism, which enables switches to control the forwarding of traffic in port queues according to pre-defined Gate Control List (GCL). The length of the GCL is limited, and the previous method of scheduling cycle with a hyper period may result in a larger GCL. Based on Satisfiability Modulo Theories (SMT), we propose a TSN scheduling method for industrial scenarios and develops a series of scheduling constraints. Different from the previous scheduling methods, the method proposed in this paper adopts the base period cycle to update GCL regularly, which can effectively reduce the number of time slots in GCL and make the configuration of GCL simpler and more efficient. In addition, compared with the traditional hyper period method, the method proposed in this paper can calculate the scheduling results faster while ensuring low latency and reducing the runtime effectively.

Time synchronization is one of the key technologies in Industrial Wireless Sensor Networks (IWSNs), and clustering is widely used in WSNs for data fusion and information collection to reduce redundant data and communication overhead. Considering IWSNs’ demand for low energy consumption, fast convergence, and robustness, this paper presents a novel Cluster-based Maximum consensus Time Synchronization (CMTS) method. It consists of two parts: intra-cluster time synchronization and inter-cluster time synchronization. Based on the theory of distributed consensus, the proposed method utilizes the maximum consensus approach to realize the intra-cluster time synchronization, and adjacent clusters exchange the time messages via overlapping nodes to synchronize with each other. A Revised-CMTS is further proposed to counteract the impact of bounded communication delays between two connected nodes, because the traditional stochastic models of the communication delays would distort in a dynamic environment. The simulation results show that our method reduces the communication overhead and improves the convergence rate in comparison to existing works, as well as adapting to the uncertain bounded communication delays.

In modern manufacturing, making accurate and timely decisions requires the ability to effectively handle multiple types of data. This paper presents a multimodal system designed specifically for smart manufacturing applications. The system combines various data sources including images, sensor data, and production records, using advanced multimodal large language models. This approach addresses common limitations of traditional single-modal methods, such as isolated data analysis and poor integration between different data types. Key contributions include a unified method for representing different data types, dynamic semantic tokenization for better data processing, strong alignment strategies across modalities, and a practical two-stage training method involving initial large-scale pretraining and later fine-tuning for specific tasks. Additionally, a novel Transformer-based model is introduced for generating both images and text, significantly improving real-time decision-making capabilities. Experiments on relevant industrial datasets show that this method consistently performs better than current state-of-the-art approaches in tasks like image–text retrieval and visual question answering. The results demonstrate the effectiveness and versatility of the proposed methods, offering important insights and practical solutions to enhance intelligent manufacturing, predictive maintenance, and anomaly detection, thus supporting the development of more efficient and reliable industrial systems.

Time synchronization is critical for wireless sensors networks in industrial automation, e.g., event detection and process control of industrial plants and equipment need a common time reference. However, cyber-physical attacks are enormous threats causing synchronization protocols to fail. This paper studies the algorithm design and analysis in secure time synchronization for resource-constrained industrial wireless sensor networks under Sybil attacks, which cannot be well addressed by existing methods. A node-identification-based secure time synchronization (NiSTS) protocol is proposed. The main idea of this protocol is to utilize the timestamp correlation among different nodes and the uniqueness of a node’s clock skew to detect invalid information rather than isolating suspicious nodes. In the detection process, each node takes the relative skew with respect to its public neighbor as the basis to determine whether the information is reliable and to filter invalid information. The information filtering mechanism renders NiSTS resistant to Sybil attacks and message manipulation attacks. As a completely distributed protocol, NiSTS is not sensitive to the number of Sybil attackers. Extensive simulations were conducted to demonstrate the efficiency of NiSTS and compare it with existing protocols.