This paper studies the multi-modal collaborative optimization problem of Internet of Things (IoT) information logistics supply chain based on Federated Learning and Digital Twin (FL-DT) architecture, proposes a distributed optimization framework that integrates federated learning and decision tree, and uses multi-modal data collected by IoT technology to improve supply chain efficiency. The experiment uses real-time IoT data and order, inventory and historical logistics data for analysis. The results show that the FL-DT architecture significantly optimizes supply chain performance: order processing time is shortened by 33.33%, inventory turnover rate is increased by 26.15%, logistics cost is reduced by 20.33%, response time is compressed by about 40% and order fulfillment accuracy rate is increased from 92% to 97%. Multi-modal data fusion makes the model accuracy reach 91.2%, which is 7.8% higher than that of single-modal, the transportation efficiency is increased by 23% and the carbon emission is reduced by 14%. The digital twin simulation shows that the transportation timeliness deviation is only −0.7%, but the inventory cost deviation is [Formula: see text]2.4%, and the fitting degree is 86.4%. The dynamic prediction needs to be improved. The FL-DT architecture has small loss fluctuations in abnormal events, which is better than traditional systems. The amount of communication data is reduced by 82% to 12.7 MB/round, demonstrating high efficiency and robustness. This study provides a new method for IoT supply chain optimization, verifies its potential in efficiency improvement, cost reduction and intelligent transformation, and can be expanded to more fields in the future.

Artificial Intelligence (AI) is changing the manufacturing process speedily; however, there is little empirical data on the adoption rates, technology emphasis, and performance results achieved, especially in developing economies. The study of AI adoption trends in developing economies is critical because these countries have special infrastructural and ability limitations that determine technology adoption and performance. This study fills this gap by surveying 209 manufacturing companies in India and examines the relationships between AI technologies, their areas of application, and multidimensional performance outcomes using descriptive, inferential, and multivariate analysis. Results indicate that robotics, IoT, computer vision, and machine learning are the most commonly utilized technologies, with the majority of their applications being in production, predictive maintenance, supply chain, and design activities. Firms with large size are more likely to adopt, but the size of a firm is not the decisive factor that determines performance improvement. Instead, market dynamics in an industry are more influential, particularly time-to-market, innovation, and responsiveness to customers. The results are relevant because they provide empirical associations between adoption tendency and technology orientation with quantifiable performance gains, which support the objective of the study to advise AI-context-dependent strategies in manufacturing. The novelty of the paper is not only an integrated empirical framework of the tools-applications–multidimensional outcomes but also a change of debate on adoption to an industry-centered view of adoption as opposed to size-centered view. The paper contains delicate organizational guidelines and precise actions concerning the spread of inclusive AI in emerging economies.

In the context of global energy green transformation, the increase in installed capacity of photovoltaic systems is accompanied by efficiency losses and safety hazards caused by component failures. Traditional monitoring relies on threshold alarms and regular inspections, and the false alarm rate of distributed scenarios reaches 12–18%, making it difficult to cope with dynamic failures. This study innovatively proposes a hybrid twin model framework for real-time monitoring and fault prediction of photovoltaic systems, integrating the physical mechanism of electric thermal coupling with deep learning data-driven methods, constructing a multimodal data collaborative mapping digital image, and combining it with an edge cloud collaborative architecture. Empirical results of a 30 kW photovoltaic array show that under shadow occlusion and hotspot fault scenarios, the model detection accuracy is 95.6% and the false alarm rate is 3.2%, which shortens the response time by 47 ms compared to traditional LSTM. To address issues such as inverter harmonic anomalies, a time series attention mechanism is used to extract frequency domain features, achieving 2–4 h of early warning (accuracy rate of 89.3%). The average latency of 10,000 data points analysis in 5G environment is 82 ms, meeting real-time monitoring requirements and providing a high-precision and low-latency technical path for photovoltaic intelligent operation and maintenance.

This study investigates the main barriers and overcoming strategies for the implementation of digital twin (DT) technology in the manufacturing industry. The research, conducted in a two-step process, first employed a modified Delphi method with seasoned industry experts to identify and classify main barriers. Later, the best-worst method (BWM) ranked main barriers and mitigation strategies. The most critical challenges determined are the integration of legacy systems, high implementation cost, and data quality. Retrofitting hardware, rolling out cloud services, and edge computing were determined as the best solutions. The findings give real-world insights for organizations seeking to deploy DTs in a prioritized set of tasks and activities applicable to the manufacturing context.

With the widespread application of cloud computing technology, the energy consumption problem of data centers is becoming increasingly severe. This research is dedicated to presenting a highly efficient and smart solution for optimizing energy usage in cloud computing systems, especially since effectively managing energy consumption through resource allocation and task scheduling has now become a critical challenge in the field. In view of this, the study first constructs a resource allocation model based on redundant resource recovery to dynamically identify and recover redundant resources. Second, a task scheduling optimization method based on dual deep reinforcement learning is established, which utilizes directed acyclic graphs to model task dependencies and optimizes task execution order and resource allocation strategies through reinforcement learning agents. In the testing of the resource allocation optimization model, when the number of tasks is 500, the redundant resource recovery rate is 87.0%, and the computational complexity is 16.7 s. The response time in the regional resource environment is 2.28 s, and the load adaptability is 95.8%. In the task scheduling method test, the proposed method has a memory occupancy of 85 MB and a scheduling delay of 0.13 s in a high-bandwidth environment of 300 Mbps. The results show that the proposed method outperforms traditional scheduling strategies in terms of resource utilization and energy consumption, and contributes to achieving the goal of green cloud computing. The research aims to provide an efficient and intelligent solution for energy optimization in cloud computing systems.

This paper addresses a dynamic job shop scheduling problem (DJSSP) based on a real stone paper factory, incorporating parallel machines, order rejection, tardiness costs, machine deterioration, and inventory costs. Orders arrive at variable times, with some potentially rejected due to capacity constraints. Also, raw materials must be purchased at appropriate times and quantities. The model optimizes three objectives: profit maximization, inventory cost variance minimization, and machine workload variance minimization. A heuristic solution method based on a double-network deep reinforcement learning (DRL) algorithm, complemented by an artificial neural network for inventory management, was developed. Evaluation against public benchmarks and other DRL algorithms highlights the impact of inventory cost variance and workload balancing. Real-case evaluation demonstrated that multi-objective optimization reduced inventory cost variance by 9.8%, machine workload variance by 21.87%, and raw material holdings by 21.7%, while slightly decreasing profit (3.2%) and slightly increasing purchase frequency (4.7%). Sensitivity analysis highlighted the individual effects of each objective on profit, workload, and makespan, confirming the practical benefits of the proposed approach.

Under the background of digital transformation in the manufacturing supply chain, frequent order fluctuations and uncertain resource allocation make it difficult for traditional static scheduling to balance timeliness and utilization. This paper raises a flexible allocation optimization model for Industrial Internet of Things (IIoT) routing scheduling and manufacturing supply chain cloud platform under order demand fluctuations. The model builds a routing scheduling framework by combining routing optimization and Genetic Algorithm (GA), introduces double deep [Formula: see text]-Network and proximal policy optimization to achieve dynamic scheduling of task selection and machine allocation, and enhances the global optimization ability of flexible allocation at the supply chain level through an Improved Random Drift Particle Swarm Optimization (IRDPSO) algorithm. Experimental results show that the proposed model achieves a minimum average end-to-end delay of 89.4 ms, a maximum link utilization of 91.2%, an order completion rate of 95.1%, and an on-time delivery rate of 92.3%. The minimum task delay rate is only 7.7%, and the single scheduling overhead is only 0.27 s, demonstrating the stability and efficiency of the model. Overall, this model achieves efficient and stable collaborative optimization of routing scheduling and flexible allocation under complex demand fluctuation scenarios and provides effective support for improving response speed and delivery reliability of the manufacturing supply chain.

In response to the low efficiency of traditional environmental perception and logistics distribution Vehicle Routing Problems (VRPs) in the logistics transportation process, a complex event-based environment perception model based on the Internet of Things is studied and designed. Combining the interpolation method and adaptive ant colony algorithm, a hybrid VRP algorithm is designed. The results showed that the optimal solutions of the design method on the four instances were 59.93%, 61.82%, 90.34%, and 91.75%, respectively. Compared with other methods, the maximum improvement in the optimal results was 13.6% and 23.5%, proving that the design method had good solving ability. By using design methods for path planning, the average transportation costs for each week within four weeks were 0.9 × 10 4 yuan, 1.1 × 10 4 yuan, 1.4 × 10 4 yuan, and 1.5 × 10 4 yuan, respectively, which were significantly lower than other algorithms. This indicates that the design method can effectively reduce transportation costs. The design method has high practicality in logistics transportation and provides strong technical support for the sustainable development of the logistics industry.

Although welding is an essential production operation, automated welding systems still struggle to produce consistently high-quality welds. Artificial intelligence (AI) and sensor technology developments in recent years have created new opportunities to increase the effectiveness and quality of welding. The state-of-the-art in sensor technologies and AI-based automation for welding operations is reviewed in this study. Numerous sensing techniques, such as force, arc, temperature, vision, infrared, and audio sensors, have been developed to track and manage welding operations in real-time. The development of intelligent welding systems (IWS) and adaptive intelligent welding production has been made possible by the integration of these sensors with cutting-edge information processing technologies, such as machine learning and deep learning. The applications of sensing technologies in welding seam extraction, welding path identification, welding tracking, and quality diagnostics are covered in this study. The research also emphasizes how crucial it is to choose the right neural network topologies and optimize them for particular goals in process monitoring and welding fault identification. Additionally covered is the incorporation of the attention mechanism notion into the information processing of welding sensing. The review concludes that sensing technology’s capacity to imitate the welder’s superior sensory and cognitive abilities is critical to the success of intelligent welding. Existing AI and sensor-integrated welding systems face challenges such as high complexity, cost, and adaptability issues. This review aims to systematically explore how the integration of AI and advanced sensor technologies can overcome existing limitations in automated welding systems. It identifies the most effective sensor-AI combinations for real-time weld monitoring, fault detection, and adaptive process control. These advancements offer practical benefits such as enhanced weld quality, reduced human intervention, and improved efficiency in industrial applications like automotive and aerospace manufacturing.

The Internet of Things (IoT) is rapidly transforming modern smart environments, but its distributed nature and resource-constrained devices pose significant challenges to cybersecurity. To address these problems, this manuscript proposes an intrusion detection method based on Binarized Simplicial Convolutional Neural Networks (IoT-IDS-Bi-SCNN), which integrates advanced preprocessing and feature selection for enhanced IoT security. The raw data from the CIC-IDS2017 dataset are first preprocessed using an Adaptive Fast Desensitized Kalman Filter (AFDKF) to impute missing values and improve data reliability. Later, the most discriminative features are chosen using the Secretary Bird Optimization Algorithm (SBOA) to lower dimensionality and boost learning effectiveness. These features are fed into the Bi-SCNN model, which leverages higher-order simplicial structures and binary operations for efficient and accurate intrusion classification. The proposed model classifies traffic into multiple categories, including Benign, Brute Force, DoS, DDoS, Web Attack, Heartbleed, Infiltration, Botnet and Port Scan. Experimental findings indicate that the proposed IoT-IDS-Bi-SCNN achieves a detection rate of 98.75%, recall of 98.89%, accuracy of 98%, precision of 98.5% and F1-score of 98.77%, while maintaining a low computational time of 1.65 s and outperforms state-of-the-art models like an actual-time intrusion detector for IoT nodes by deep learning (ID-IoT-LSTM), improving IoT network security by a deep learning-powered intrusion detection system (DLP-IDS-FFNN), and a new deep learning-based intrusion detection system for IoT DDoS safety (IDS-IoT-DNN). These findings indicate the proposed method’s suitability for resource-aware IoT intrusion detection.