Design Failure Mode and Effects Analysis (DFMEA) traditionally relies on static tables, limiting its ability to capture failure propagation across design hierarchies or support reasoning under out-of-distribution conditions. These constraints hinder knowledge reuse and evidence-based decision-making in early design phases. To address this, we formalize failure knowledge as a directed Knowledge Graph grounded in fault-tree logic and introduce AfGNN (Adaptive Failure Graph Neural Network). AfGNN achieves robust prediction on rare, long-tail failure patterns through three integrated innovations: (1) Causal-Enhanced Soft-Label Embedding (CESLE), which integrates semantic similarity with causal weights to distinguish genuine relationships from statistical correlations; (2) a Depth-Adaptive Causal Propagation Framework, synergizing dynamic subgraph sampling with depth-decay attention to balance efficiency and fidelity while suppressing noise in deep layers; and (3) a formalized computational workflow that transforms DFMEA into a reusable causal graph, enabling systematic reasoning over incomplete failure records. Evaluated on five public datasets and a self-constructed automotive failure KG, AfGNN surpasses all GNN-based baselines and competes with LLM-based methods on general benchmarks, while substantially outperforming all baselines on the domain-specific FMEA dataset (MRR, Hits@1, Hits@10). This framework enables engineers to reason about rare multi-failure cascading effects without relying on complete historical data, advancing failure knowledge management and reliability-oriented design decision-making.

The application of digitalization in manufacturing involves using sensors to collect and transmit large amounts of data in real-time. These complex, timestamped sequence data require effective analytical support to drive the generation of analysis summaries. Visual analysis helps researchers identify hidden patterns by intuitively displaying complex data, but its implementation relies on specialized knowledge and a deep understanding of data analysis techniques. Moreover, the analysis process often requires users to recall and process large amounts of information, which increases the complexity of the analysis. Therefore, there is an increasing demand for intelligent visual analysis technology. Large language models, with their powerful reasoning, summarization, and code generation capabilities, provide the possibility of visual analysis for general users. We propose a large language model-based multi-agent framework that integrates domain knowledge to achieve an automated data analysis workflow, from data acquisition to analysis summarization, helping users extract valuable analytical results from complex data and improving both efficiency and experience.

The assembly process information of wind turbines is usually scattered in process documents consisting of multi-modal information, such as 3D models, natural texts, and images. Therefore, the cost of maintaining data and extracting process knowledge is high while the efficiency is low. To solve this problem, a multi-modal knowledge graph-based modeling method for the assembly process knowledge of wind turbines is proposed with multi-source heterogeneous data. First, the concepts in multi-modal process knowledge graph of wind turbine (MPKG-WT) are defined by analyzing the process characteristics of wind turbines to complete the construction of ontology. Then, based on the characteristics of multi-source heterogeneous data and multi-modal information, data analysis, knowledge extraction, and semantic similarity calculation are leveraged to realize the automatic instantiation of the graph. Finally, taking the process data of a wind turbine enterprise as an example, MPKG-WT is constructed and verified by implementing an auxiliary system for process design. The results show that MPKG-WT is more informative than the single-modal graph, and the data in different modals can complement each other, which leads to significant improvements in the efficiency of process design.

Digital twin technology is a crucial driving technology for the realization of Industry 4.0, which enables simulation, analysis, and prediction by constructing geometric scenes that correspond to physical scenes. Current automation in scene modeling is limited, relying heavily on costly and inefficient manual operations. This hinders digital twin technology progress. This paper proposes a framework for a fast scene geometric modeling method that combines neural rendering and model retrieval for digital twins. Neural rendering techniques first train the collected image data, and the point cloud data of physical entities are rendered. Then based on the semantic mapping algorithm between the point cloud data of physical entities and the multi-view image data of 3D CAD models proposed in this study, the corresponding geometric models are retrieved from the geometric asset library by inputting the point cloud data, and all the retrieved geometric models are embedded in the geometric scene to complete the scene geometric modeling. Finally, in a case study of the digital twin-based scene construction for decommissioned lithium battery dismantling, the effectiveness of this method is demonstrated, which can improve the speed and reduce the cost of scene geometric modeling.

Weakly rigid drilling systems such as robots are widely used in complex drilling scenarios because of their good flexibility and wide operating range. However, its weak rigidity would easily lead to flutter and burr. How to control burr is a hot and difficult point in the research. Based on digital twin technology, a weak rigid drilling system was proposed that consisted of entity model, virtual model, data processing module and drilling decision-making module. The behavior-rule model based on mechanism and data fusion was proposed to realize the state monitoring in the machining process. The proposed drilling decision-making module included a decision-information generation model andoptimization algorithms to make state optimization. The feasibility and effectiveness of the burr monitoring process and the decision-making process were verified by the comparison between the physical drilling experiment and the simulation experiment of the weak rigid machining system. The results showed that the average burr height could be reduced by about 14%.

为解决当前棉纱纱疵定量分析方法精度低、可靠性差的问题,提出一种基于异构集成学习的纱疵定量分析方法.首先,建立基于电容传感器的纱疵检测二维动态仿真模型,分析纱疵尺寸对纱疵信号的影响规律.其次,针对非平稳、非线性的纱疵信号难处理以及纱疵量化特征不明显的问题,采用时域分析方法,提取纱疵信号的时域参数作为纱疵定量分析特征,针对传统阈值和数值分析方法对纱疵定量精度低的问题,以支持向量机回归算法和径向基神经网络算法组成初级学习器,集成梯度提升决策树为元学习器,建立用于纱疵定量分析的异构集成学习算法模型.实验结果表明,本文方法比其它单模型回归拟合方法的检测准确率提升约10％,验证了本文方法对纱疵定量分析结果的可靠性.

设备点检记录是支撑故障原因分析与处理的重要信息来源,目前亟需对设备点检故障中的根因信息进行有效挖掘,以提升设备预防性维护的可靠性.鉴于此,首次将因果科学论引入制造领域,提出一种面向设备点检故障根因分析的因果知识建模方法.首先,从设备点检故障文档中提取事件知识,构建故障运维因果知识图谱;其次,定义故障运维因果知识规则,形成结构因果图模型;进而,设计一种基于ISPN的因果效应估计学习模型,对故障知识中混杂影响因素进行估计计算,挖掘出影响设备故障发生的语义关系,补全图谱节点间隐含的因果性语义链路;最后,以冶金设备点检故障文档的知识测试了所提方法,验证了因果知识模型估计设备故障根因知识间因果效应的可行性.

在汽车生产环节进行数字建模、系统仿真与优化对提升汽车的生产质量和效率具有重要意义.为了解决目前汽车制造企业普遍存在的因数据链断裂而导致的资源配置效率低下等难题,以汽车涂装车身缓存区(painted body storage,PBS)为研究对象,提出了一种新型的数字底座平台,来实现数据链整合和多源异构数据融合.同时,设计了一种针对PBS的车身调序策略,考虑了总装工艺对车序优化的约束,采用遗传算法获得了PBS出车序列,然后以逆序数对为参考指标,进行PBS车道排布.将基于数字底座的PBS系统应用于某汽车制造企业,应用效果验证了所提出方法和策略的有效性.研究结果为企业构建内部集成制造平台和设计具体车间单元提供了参考.

In computer aided design and manufacturing systems, the manufacturing feature recognition is a key technology. Aiming at the problems of poor scalability and robustness of traditional feature recognition technology, a manufacturing feature recognition method based on point cloud deep learning was proposed. A point cloud dataset of manufacturing features was constructed by sampling uniformly on the surface of manufacturing features. The K-nearest neighbor algorithm was used to construct a rotation-invariant representation of the point cloud, and a point cloud classification network incorporating geometric prior knowledge was proposed. For the point cloud data of the model with multiple features, an extraction method of the point cloud of manufacturing features and a separation method of intersecting features were proposed. Practical experiments were carried out to demonstrate the effectiveness of the proposed method, and the results illustrated that the method could effectively recognize single features and interacting features for CAD models.

In current small batch and customized production mode, the products change rapidly and the personal demand increases sharply. Human-robot cooperation combining the advantages of human and robot is an effective way to solve the complex assembly. However, the poor reusability of historical assembly knowledge reduces the adaptability of assembly system to different tasks. For cross-domain strategy transfer, we propose a human-robot cooperative assembly (HRCA) framework which consists of three main modules: expression of HRCA strategy, transferring of HRCA strategy, and adaptive planning of motion path. Based on the analysis of subject capability and component properties, the HRCA strategy suitable for specific tasks is designed. Then the reinforcement learning is established to optimize the parameters of target encoder for feature extraction. After classification and segmentation, the actor-critic model is built to realize the adaptive path planning with progressive neural network. Finally, the proposed framework is verified to adapt to the multi-variety environment, for example, power lithium batteries.

激光焊接在航空领域具有广泛的应用场景,基于视觉的激光焊接缺陷识别对于产品质量的提高至关重要.针对当前基于深度学习的激光焊接缺陷识别方法存在可解释性差的问题,提出了一种融合多尺度特征的类激活映射(MSF-CAM)方法.在训练阶段,以VGG16为骨架模型并将监督信息施加于多个尺度以促进模型对多尺度特征的学习.在测试阶段,对输出类别在多个尺度上的激活图进行叠加,并以此作为模型的判断依据.多尺度特征的融入不但增强了模型的可解释性,而且还提高了激光焊接缺陷识别的准确性.试验结果表明:MSF-CAM在测试集上的准确率为98.12％,识别单幅图像耗时8.28 ms.此外,MSF-CAM可以从边缘、轮廓这种初级特征的角度对模型的决策依据提供人类更容易理解的解释.

为了保证复杂产品在关键装配阶段的一次成功率和质量一致性，提出了一种基于数字孪生的增强现实（Augmented Reality，AR）多人协作装配方法。首先提出了基于数字孪生的AR多人协作装配架构，并依据协作装配的复杂性对装配序列进行过程分解。其次，将装配过程上下文数据构造为装配工艺知识图谱，明确了协同过程装配信息的传递与迭代机制，提高了AR协作装配的工况自适应性。在此基础上，采用基于根锚点的协同方法，实现多人装配过程的AR场景协同，并依赖装配工艺知识图谱提供的实时知识实现数据协作，保证了协同装配的质量和效率。最后，以某钢厂复杂轧机的装配为例，建立了基于增强现实的多人协作数字孪生装配系统，验证了论文方法的有效性和可行性。

In the multi-variety and small batch manufacturing workshop, digital twin model is mostly established for specific scenarios. Due to its lack of adaptive ability under working conditions, the prediction accuracy of machining quality is insufficient. To solve this problem, an adaptive transferring method of the digital twin model is proposed. By building the transferable digital twin model, the online prediction of machining quality based on the fusion of mechanism and algorithm model is realized. The transferring process and strategy of the digital twin model are proposed. Based on the analysis and calculation of characteristic data, the source model to be transferred is selected. At the same time, in combination with the transfer learning theory, the transfer of digital twin models is realized under simple and complex changing conditions. Taking drilling as an example, the drilling experiment platform is built and the feasibility of digital twin model transfer is verified. The results show that the model can keep the mean absolute error of prediction less than 1.5% under changing working conditions. This method provides a new idea to improve the adaptive ability of digital twin models.

为解决制造过程中多学科、多物理量、多尺度、动态时变等因素带来的虚实融合困难问题,以航天薄壁件旋压成型加工过程为对象,提出一种数字孪生高保真建模方法.该方法分别从几何、机理和数据3个层次描述和定义模型,并基于构建的模型研究了旋压加工过程知识获取和数据虚实映射过程.为评估所建模型的有效性,给出了一种数字孪生模型的保真度评估方法.通过案例验证了所提数字孪生高保真建模方法的有效性和模型的可评估性.

在制造系统中获取并利用知识辅助决策,发展认知学习能力,已成为发展新一代智能制造的核心需求,而知识图谱为满足这一需求提供了成熟的条件.知识图谱与制造系统的融合应用成为今后发展的必然趋势.为全面深入地了解制造领域知识图谱的研究现状并探究其发展应用前沿,在充分搜集整理现有文献的基础上,首先从发文量趋势、文章关键词、研究的关键技术和涉及的案例场景等多个维度进行了统计和分析,然后根据现有研究成果的共性总结出制造领域知识图谱的三个应用层次:语义关联、定性决策和综合决策,最后总结了制造领域知识图谱面临的高质量制造知识获取、复杂制造知识表示、行业制造图谱开放共享等一系列复杂挑战,并对制造领域知识图谱在应用层次、动态时效性和服务对象等方面的未来发展进行展望.

数字孪生正在制造系统中发挥重要作用,然而在面向人机协助完成的复杂制造场景中,人-机-环境及其构成的数字孪生系统呈现出任务异构复杂、环境动态多变及其交互实时等特点.目前欠缺人-机-环境共融的数字孪生协同过程中智能方法相关研究,尤其是数字孪生模型在协同中的迁移和强化,以满足制造系统的鲁棒性和自适应能力.提出面向人-机-环境共融的数字孪生协同技术,从环境和任务两个核心来展开数字孪生协同的人机共融科学问题.首先给出协作装配环境的数字孪生体系,以虚拟装配的形式为人-机-任务交互提供理解;建立相应的空间模型与协同模型,为共融的孪生协同提供理论支持;最后,以最典型的人机共融制造场景(装配任务)为案例,在决策层基于迁移学习算法为机器人提供装配操作指引,同时通过强化学习算法优化机器人的具体执行动作.在不同型号产品的人机协同装配任务中,均可以生成相应的人机协作装配规划方案,证明了所提方法的可行性.

纺织行业因涉及面大、生产过程复杂、自动化和数字化水平发展不均衡,对如何分阶段实现智能制造缺乏顶层规划.从"智能"角度分析了纺织制造系统的演化进程,并据此提出面向认知的新一代纺织智能制造体系.该体系以数据为要素,以知识图谱为集成方式,以信息的感知和认知为智能驱动,切合了智能制造的内在驱动力.以棉纺生产为具体对象,分析了棉纺生产全生命周期的感知和认知过程,并对过程监控、任务实时调度、产品质量优化及设备运行和维护等关键技术进行分析.结果表明,面向认知的纺织智能制造体系促进了纺织生产系统的流程优化,并提升了产品质量和生产效率.

为解决高精密产品装配过程中虚拟仿真分析与物理装配脱节导致的装配效率较低、装配质量一致性较差的问题,提出一种数字孪生驱动的高精密产品智能化装配方法.构建了包括装配全要素的高精密产品数字孪生体;针对当前装配工艺文档可读性差、知识关联关系弱等问题,提出一种基于知识图谱的装配工艺表达方式,并利用知识图谱的可推理性对装配工艺进行动态调整;针对产品质量控制问题,提出一种操作—状态—质量三层结构质量控制策略.以某型号汽车发动机缸体单元装配为例,验证了所提方法的实用性.

In the production process of aerospace structural parts, there coexist batch production tasks and research and development (R&D) tasks. Personalized small-batch R&D and production tasks lead to frequent emergency insertion orders. In order to ensure that the task is completed on schedule and to solve the flexible job shop dynamic scheduling problem, this paper takes minimization of equipment average load and total completion time as optimization goals, and proposes a dual-loop deep Q network (DL-DQN) method driven by a perception-cognition dual system. Based on the knowledge graph, the perception system realizes the representation of workshop knowledge and the generation of multi-dimensional information matrix. The cognitive system abstracts the scheduling process into two stages: resource allocation agent and process sequencing agent, corresponding to two optimization goals respectively. The workshop status matrix is designed to describe the problems and constraints. In scheduling decision, action instructions are introduced step by step. Finally, the reward function is designed to realize the evaluation of resource allocation decision and process sequence decision. Application of the proposed method in the aerospace shell processing of an aerospace institute and comparative analysis of different algorithms verify the superiority of the proposed method.

为解决人工评估复杂工艺表格的相似性用于工艺重用设计存在效率低、精度差等问题,提出一种图神经网络组合算法,以有效提取工艺表格的结构、语义等特征以度量相似性.首先提出改进Mask R-CNN算法用以进行表格检测,包括距离变换突出表格特征、Confluence算法提高检测精度、角点定位调整检测框以实现精准定位,同时利用光学字符识别(OCR)技术提取表格文本信息.然后,针对提取的关键单元信息,分别建模工艺表格的结构特性图网络与语义关系图网络.进一步,提出图神经网络组合算法提取图网络模型的结构特征与节点属性,并转化成低维实值向量,以支撑提出的一种联合相似度综合评估方法,实现度量工艺表格语义相似性.最后,经实验分析表明了所提方法的有效性,并以工艺重用实例验证了方法的可行性.