Intelligent fault diagnosis of rotating machinery is essential for manufacturing reliability and predictive maintenance, yet deployment of deep learning models is limited by data scarcity: fault samples are rare, costly, and hazardous to obtain. Conventional synthetic data methods such as Generative Adversarial Networks and Variational Autoencoders often exhibit mode collapse, spectral distortion, and limited physical interpretability. This work presents MechaForge, a multi-strategy framework that employs Large Language Models (LLMs) as physics-guided generators for bearing fault time-series data. The approach is grounded in bearing kinematics, Motor Current Signature Analysis (MCSA), and the interpretation of in-context learning as implicit Bayesian inference. Within MechaForge, four progressively constrained tracks are defined: a real-data baseline, few-shot LLM mimicry, multi-stage semantic reasoning, and physics-guided generation with constraints on root mean square, kurtosis, and fault-band spectral energy. For direct benchmarking, conventional VAE- and GAN-based augmentation baselines are additionally evaluated under the same dataset split, synthetic-data budget, downstream CNN architecture, and evaluation metrics. Experiments on the Paderborn bearing dataset show that the Basic LLM track achieves the strongest performance under the present protocol (0.7862 accuracy, 0.7648 macro-F1), exceeding the added VAE and GAN baselines (both 0.7428 accuracy; 0.7202 and 0.7257 macro-F1, respectively), while a control experiment confirms that synthetic data provides discriminative structure rather than labeled noise. These results indicate the promise of LLM-based diagnostic augmentation under data scarcity in the present Paderborn setting, rather than a definitive demonstration of broad transferability across fault-diagnosis scenarios.

In edge computing systems, load prediction plays a crucial role in optimizing resource allocation and enhancing system performance. Traditional prediction models tend to perform well when handling loads with clear trends or periodic patterns. However, the dynamic nature of edge environments and the multidimensional complexity of data often challenge these models, leading to decreased accuracy and efficiency in complex scenarios. Recently, advances in patch-based time series prediction methods and frequency domain analysis techniques have significantly improved prediction accuracy in such environments. Against this backdrop, this paper proposes a frequency-domain based model: the Multi-Scale Patch Multilayer Perceptron Network (FMPNET), designed for high-precision load prediction in edge computing. By decomposing time series data into sub-bands of distinct frequencies using the Discrete Wavelet Transform (DWT), the FMPNET model captures both rapid fluctuations and long-term trends. The multi-scale patch mechanism enables short-term patches to identify local fluctuations and transient changes, while long-term patches focus on global trends. This design enhances the model’s capacity to grasp the global structure of the data, preserve fine-grained details, and improve adapt ability to multidimensional load patterns. Extensive experiments on real-world edge computing load datasets demonstrate that FMPNET achieves superior prediction accuracy and efficiency across various scenarios. Compared to benchmark methods, FMPNET not only significantly enhances prediction accuracy but also exhibits strong generalization capability, affirming its effectiveness in dynamic and complex edge environments.

With the rapid advancement of Industry 4.0 and intelligent manufacturing, the complexity of tasks and data processing in CNC systems continues to grow, rendering traditional centralized resource allocation methods inadequate in meeting real-time and efficiency demands, thereby creating system performance bottlenecks. To address this challenge, this paper proposes an adaptive task allocation method for CNC systems based on edge intelligence. By monitoring system states in real time, this approach leverages a greedy algorithm to adjust task priorities and optimize resource allocation. The proposed method significantly enhances task processing efficiency, enabling dynamic adjustments upon detecting anomalies, thus ensuring system stability and safety. Experiments conducted in real industrial scenarios validate the effectiveness of this method, demonstrating a notable improvement in resource utilization and response speed of CNC systems.

Precise dynamic models or fine tuning is necessary for traditional dynamical system controller. However, in many scenarios, precise dynamic models may be hard acquire or they just change by time. And aimless tuning doesn’t always yield ideal results. In this paper, we propose a model-free online controller based on reinforcement learning. We apply Q-learning algorithm and update new control value according to control values and real-time system feedbacks based on RLS iteratively in a model-free method. The controller is validated in gym simulation environment and compared with LQR controller. Results proved that our controller performs as well as LQR and can automatically adapt to model changes.

The rapid advancement of Industry 4.0 and intelligent manufacturing has elevated the demands for fault diagnosis in servo motors. Traditional diagnostic methods, which rely heavily on handcrafted features and expert knowledge, struggle to achieve efficient fault identification in complex industrial environments, particularly when faced with real-time performance and accuracy limitations. This paper proposes a novel fault diagnosis approach integrating multi-scale convolutional neural networks (MSCNNs), long short-term memory networks (LSTM), and attention mechanisms to address these challenges. Furthermore, the proposed method is optimized for deployment on resource-constrained edge devices through knowledge distillation and model quantization. This approach significantly reduces the computational complexity of the model while maintaining high diagnostic accuracy, making it well suited for edge nodes in industrial IoT scenarios. Experimental results demonstrate that the method achieves efficient and accurate servo motor fault diagnosis on edge devices with excellent accuracy and inference speed.

This study proposes a residual modeling approach for machine tool dynamics based on the Transformer architecture, aiming to enhance the predictive accuracy of traditional kinematic models within complex machine tool systems. While conventional Denavit-Hartenberg models excel at describing geometric configurations, they struggle to capture non-Markovian characteristics and intricate dynamic behaviors arising from factors such as thermal deformation, tool wear, and frictional hysteresis. To address this limitation, we leverage the Transformer model's powerful sequence learning capabilities and its aptitude for capturing long-range dependencies, constructing a compensatory model that learns the residuals between the D-H model and the actual dynamic responses. The D-H framework serves as structured prior knowledge, while the Transformer is tasked with modeling the unexplained dynamic deviations, particularly those governed by nonlinear effects requiring longterm memory and contextual awareness. Experimental results demonstrate that this method significantly improves the fidelity of dynamic modeling, thereby providing a more robust digital foundation for high-precision machining and intelligent control.

Addressing the urgent needs for real-time, reliable and interpretable diagnosis of CNC machine tools in intelligent manufacturing, this paper proposes an agent-based edge-cloud collaborative fault diagnosis system. The system designs a multisource heterogeneous data acquisition architecture that integrates multi-type sensor data such as vibration and temperature, and achieves efficient preprocessing through edge computing. To improve diagnosis accuracy and real-time performance, this paper constructs a lightweight MCGNN_MSTransformer model that integrates channel graph perception, multi-scale convolution, local Transformer, and frequency domain branches to effectively model multi-class fault features under complex working conditions. In experiments with six typical faults of servo motors, the model achieved a test set accuracy of 99.02 %, significantly outperforming comparative methods such as 1D-CNN, BiLSTM, and ResNet1D. Meanwhile, this paper introduces a knowledge graph reasoning mechanism that associates model outputs with fault mechanisms, phenomena, and treatment methods to generate interpretable diagnostic reports and maintenance recommendations. The research results show that the system can achieve a closed-loop process from data acquisition, feature extraction, intelligent diagnosis to knowledge-driven decision making, providing an efficient and scalable solution for intelligent health management of CNC machine tools.

Computer Numerical Control (CNC) machine tools are pivotal in modern manufacturing, yet their machining performance, precision, and service life are frequently compromised by anomalies in axis motion arising from assembly quality issues or wear due to prolonged use. To address the challenge of accurate anomaly detection in CNC machine tools, this study proposes a novel Cross-Modal Gated Multi-Scale Transformer (XG-MST) model, leveraging multi-axis vibration and current signals for supervised learning. Evaluated on a self-collected dataset derived from real-world CNC machine tool operations, the model achieves a detection accuracy of over 95%. Further validation through small-batch multi-axis simulation experiments on the open-source Paderborn Bearing Dataset yields an accuracy of over 90%. These results underscore the XG-MST algorithm's superior performance and significant practical value for enhancing machine tool quality inspection and predictive maintenance in manufacturing.

In recent years, the deep integration of advanced information technology and advanced manufacturing technology has gradually become one of the main ways to achieve smart manufacturing. The computer numerical control (CNC) system is the basic equipment for machining and manufacturing, and the quality and efficiency of the system’s machining are the basis for supporting and ensuring smart manufacturing. However, the G-code used in CNC machining is usually generated with computer-aided manufacturing (CAM) according to a static model, and its tool path is relatively rough, with uneven adjacent paths and bad points in the path causing machining defects. To solve these problems, a modeling approach combining the basic elements of the intelligent CNC system with the human-cyber-physical system (HCPS) model is proposed, and a digital solution for tool path optimization is further proposed, integrating the redesign process of CAM tool path into cyber application. In addition, the process of tool path optimization is processed in steps, and a pipelined processing flow is established to accelerate the optimization process. Finally, the effectiveness of the proposed method is demonstrated using an example of process file optimization for a pentagram convex rib model.

In recent years, artificial intelligence technology has seen increasingly widespread application in the field of intelligent manufacturing, particularly with deep learning offering novel methods for recognizing geometric shapes with specific features. In traditional CNC machining, computer-aided manufacturing (CAM) typically generates G-code for specific machine tools based on existing models. However, the tool paths for most CNC machines consist of a series of collinear motion commands (G01), which often result in discontinuities in the curvature of adjacent tool paths, leading to machining defects. To address these issues, this paper proposes a method for CNC system machining trajectory feature recognition and path optimization based on intelligent agents. This method employs intelligent agents to construct models and analyze the key geometric information in the G-code generated during CNC machining, and it uses the MCRL deep learning model incorporating linear attention mechanisms and multiple neural networks for recognition and classification. Path optimization is then carried out using mean filtering, Bézier curve fitting, and an improved novel adaptive coati optimization algorithm (NACOA) according to the degree of unsmoothness of the path. The effectiveness of the proposed method is validated through the optimization of process files for gear models, pentagram bosses, and maple leaf models. The research results indicate that the CNC system machining trajectory feature recognition and path optimization method based on intelligent agents can significantly enhance the smoothness of CNC machining paths and reduce machining defects, offering substantial application value.

Intelligent manufacturing is the main direction of Industry 4.0, pointing towards the future development of manufacturing. The core component of intelligent manufacturing is the computer numerical control (CNC) system. Predicting and compensating for machining trajectory errors by controlling the CNC system’s accuracy is of great significance in enhancing the efficiency, quality, and flexibility of intelligent manufacturing. Traditional machining trajectory error prediction and compensation methods make it challenging to consider the uncertainties that occur during the machining process, and they cannot meet the requirements of intelligent manufacturing with respect to the complexity and accuracy of process parameter optimization. In this paper, we propose a hybrid-model-based machining trajectory error prediction and compensation method to address these issues. Firstly, a digital twin framework for the CNC system, based on a hybrid model, was constructed. The machining trajectory error prediction and compensation mechanisms were then analyzed, and an artificial intelligence (AI) algorithm was used to predict the machining trajectory error. This error was then compensated for via the adaptive compensation method. Finally, the feasibility and effectiveness of the method were verified through specific experiments, and a realization case for this digital-twin-driven machining trajectory error prediction and compensation method was provided.

Intelligent manufacturing has garnered widespread attention due to its potential to enhance production efficiency and product quality. However, effective resource management remains crucial and challenging, particularly in real-time data processing and task optimization. In response to the shortcomings of existing task offloading solutions regarding network latency and resource limitations, this paper introduces a novel framework for task offloading based on reinforcement learning. This framework dynamically enhances the execution efficiency of neural network tasks and resource allocation within the synchronized control of numerical control systems. Furthermore, this study incorporates digital twin technology further to augment the system’s response speed and processing capabilities. Empirical research has demonstrated the significant advantages of the proposed method in reducing computational delay and enhancing processing efficiency. These improvements enhance operational flexibility and ensure efficient system performance in resource-constrained environments. Experimental results indicate that the proposed algorithm surpasses various existing technologies in latency optimization within the experimental benchmark environment.

With the development of smart factory, the requirements of multi-objective, real-time and intelligent are put forward for the dynamic flexible job shop scheduling problem (DFJSP). In this paper, the flexible job shop intelligent scheduling system is constructed, and the workshop level industrial network topology, scheduling information integration model and its main functions are designed. A soft dueling double deep Q-networks (SDDDQN) method is proposed to solve the DFJSP with random job arrival to minimize the total machine load and maximize the utilization rate of machine tools. Through the gear production scheduling case test, the results show the effectiveness, generalization and timeliness of the flexible job shop scheduling system based on the SDDDQN.

随着科学技术的发展和生产工艺的进步，智能化数控系统是必然的发展趋势。智能化数控系统的软硬件平台是算法运行的基础。因此，从系统的整体效能出发，阐述数控系统智能化软硬件平台、软件架构、轨迹规划方法。同时测试软件模块在智能化数控系统架构中的主要功能，并分析试验测试结果。本文提出了一种智能化轨迹规划方法，包括算法的智能选取、智能前瞻、动态规划等模块。根据加工路径信息，自适应地选取合适的算法进行轨迹规划，提高表面质量或加工效率。智能前瞻模块自动调整线段两端的速度可达性，对拐角速度有一个下压的过程。动态轨迹规划根据外界信号和当前的加工状况，按原有的加减速规律实时地调整速度曲线，在单段内响应或者跨段响应。在中科数控系统的智能化软硬件平台上进行验证，试验结果表明，所提出的方法能够在实时响应外界环境变化的同时，缩短加工时间。

In recent years, digital twin (DT) technology has gradually become the primary way to achieve the intelligence of CNC systems. However, with the development of next-generation information technologies such as artificial intelligence (AI) and its wide application in CNC systems, the limitation of computing power and network resources has become one of the urgent problems that must be solved by the DT of CNC systems. To address these problems, a theoretical modeling method for CNC systems based on its hierarchical structure is proposed first, and the edge intelligence (EI) technology is introduced to support the deployment of DT models. Meanwhile, a model partitioning method and a model selection algorithm are proposed to support real-time model response in the model deployment process. In addition, an application case of EI-driven DT of CNC system is given to diagnose and predict the tool wear during machining processes.

With the development of intelligent manufacturing, machine tools are considered the “mothership” of the equipment manufacturing industry, and the associated processing workshops are becoming more high-end, flexible, intelligent, and green. As the core of manufacturing management in a smart shop floor, research into the multi-objective dynamic flexible job shop scheduling problem (MODFJSP) focuses on optimizing scheduling decisions in real time according to changes in the production environment. In this paper, hierarchical reinforcement learning (HRL) is proposed to solve the MODFJSP considering random job arrival, with a focus on achieving the two practical goals of minimizing penalties for earliness and tardiness and reducing total machine load. A two-layer hierarchical architecture is proposed, namely the combination of a double deep Q-network (DDQN) and a dueling DDQN (DDDQN), and state features, actions, and external and internal rewards are designed. Meanwhile, a personal computer-based interaction feature is designed to integrate subjective decision information into the real-time optimization of HRL to obtain a satisfactory compromise. In addition, the proposed HRL framework is applied to multi-objective real-time flexible scheduling in a smart gear production workshop, and the experimental results show that the proposed HRL algorithm outperforms other reinforcement learning (RL) algorithms, metaheuristics, and heuristics in terms of solution quality and generalization and has the added benefit of real-time characteristics.

With the increase in the number of networking devices in the industrial field, the total amount of data increases, resulting in huge broadband pressure and power consumption, making the traditional centralized processing outdated, and giving rise to the development of edge computing in the industrial field. To ensure the security of the overall network, the trust concept of the access terminal computer numerical control (CNC) system is introduced as an important basis for improving the manufacturing capability of CNC machine tools, so the inclusion of computing power and trust mechanism in the intelligent CNC system is a good solution to the lack of computing and analysis resources and the ability to reduce the security risks of the CNC system. In this paper, we mainly design a trust level evaluation method for the intelligent CNC system based on edge intelligence, which incorporates the trust evaluation in the edge computing environment, and effective plans and coordinates the cloud, edge, and final resources in the intelligent CNC system. We designed the trust degree model evaluation and constructed the security trust level of the CNC intelligent system with the collaboration between the cloud and the edge. Finally, the experimental scheme is validated and tested in a real environment.

The production process of a smart factory is complex and dynamic. As the core of manufacturing management, the research into the flexible job shop scheduling problem (FJSP) focuses on optimizing scheduling decisions in real time, according to the changes in the production environment. In this paper, deep reinforcement learning (DRL) is proposed to solve the dynamic FJSP (DFJSP) with random job arrival, with the goal of minimizing penalties for earliness and tardiness. A double deep Q-networks (DDQN) architecture is proposed and state features, actions and rewards are designed. A soft ε-greedy behavior policy is designed according to the scale of the problem. The experimental results show that the proposed DRL is better than other reinforcement learning (RL) algorithms, heuristics and metaheuristics in terms of solution quality and generalization. In addition, the soft ε-greedy strategy reasonably balances exploration and exploitation, thereby improving the learning efficiency of the scheduling agent. The DRL method is adaptive to the dynamic changes of the production environment in a flexible job shop, which contributes to the establishment of a flexible scheduling system with self-learning, real-time optimization and intelligent decision-making.

In transonic wind tunnel, anomalous data that are often referred to as outliers or anomalies have severe impact on system identification. To address such a problem, outliers should be detected and new substitutions should be provided before system identification. The combined request for outlier detection and compensation makes it suitable to develop a regression-based outlier mining algorithm. To enhance the effectiveness of traditional regression-based algorithm, this paper proposes a novel one based on ensemble learning. In our outlier ensemble, the base regression models are learnt on a two-level ensemble structure. The aim of the first level is to enhance the robustness to unknown outliers by homogeneous ensemble. The goal of the second level is to improve the robustness to base regression model. In order to verify the effectiveness of the proposed hybrid outlier ensemble, we use several real-world datasets from transonic wind tunnel and compare it with several underlying competitors. The experimental results have shown that the proposed outlier ensembles could outperform its competitors with respect to both outlier mining and the improvement of system identification.

As a key technology to improve the national equipment level and manufacturing level, CNC technology is of great significance to the development of the national strategic height. The CNC system is an important foundation for improving the manufacturing capacity of CNC machine tools, so its performance evaluation has an important influence on the development of the CNC system. This paper designs a performance evaluation platform for CNC system based on micro-services. The platform mainly includes three levels: data collection layer, data service layer and data application layer. The data collection layer is used to collect the machining trajectory data of the tested numerical control system, the data service layer is used to store the data and provide the data to the upper data application layer, and the data application layer uses the Pearson product-moment correlation coefficient method and the network stress test. To analyze the performance of the CNC system. Finally, the experimental test shows that the platform has good reliability for the performance evaluation of the numerical control system, which can make the analysis and evaluation of the numerical control system more convenient, accurate and effective for users.