Procedia CIRP最新文献_第8页

VLM-Based Agent for Image Processing Pipeline Generation in Micro-Assembly 基于vmm的微装配图像处理流水线生成代理

Procedia CIRP Pub Date : 2026-01-01 Epub Date: 2026-09-03 DOI: 10.1016/j.procir.2026.03.286

Rolf Wiemann, Niklas Terei, Annika Raatz

{"title":"VLM-Based Agent for Image Processing Pipeline Generation in Micro-Assembly","authors":"Rolf Wiemann, Niklas Terei, Annika Raatz","doi":"10.1016/j.procir.2026.03.286","DOIUrl":"10.1016/j.procir.2026.03.286","url":null,"abstract":"<div><div>Programming and setting up image processing for micro-assembly is a time-consuming and expert-intensive task, as pipelines must be manually designed and tuned. Commercial vision tools can reduce the required expertise but are costly and still depend on trained users, while deep learning approaches demand large annotated datasets that are rarely available in small-lot production. These limitations reduce flexibility and slow the commissioning of new products, especially in prototype and small-lot production environments. To address this challenge, this work presents a vision–language model (VLM)-based agent for the automatic generation of image processing pipelines. Integrated into a ROS 2-based robotic framework, the agent translates natural language requests into executable OpenCV pipelines, reducing manual programming effort and enabling intuitive human–robot interaction. The resulting pipelines remain transparent and easily modifiable when products or environmental conditions change. Experimental results show that large models such as GPT-5 autonomously generate accurate and robust pipelines, achieving subpixel localization performance comparable to commercial industrial tools. Smaller models remain effective for simpler detection tasks but exhibit reduced consistency. Future work will focus on enhancing spatial reasoning through geometric verification and deeper integration into autonomous micro-assembly programming.</div></div>","PeriodicalId":20535,"journal":{"name":"Procedia CIRP","volume":"146 ","pages":"Pages 463-468"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148939285","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Skill-based Programming of Micro-Assembly Robots 基于技能的微型装配机器人编程

Procedia CIRP Pub Date : 2026-01-01 Epub Date: 2026-09-03 DOI: 10.1016/j.procir.2026.03.235

Niklas Terei, Rolf Wiemann, Annika Raatz

引用次数: 0

A Personnel Allocation Planning Method in Back-Office Operations Considering Employee Proficiency 考虑员工熟练程度的后台业务人员分配计划方法

Procedia CIRP Pub Date : 2026-01-01 Epub Date: 2026-09-03 DOI: 10.1016/j.procir.2026.03.177

Nobuya Hayata, Nobutada Fujii, Noriaki Uenomachi, Kazuhito Uehara

引用次数: 0

Comparative Evaluation of Large Language Models for Regulatory Compliance Interpretation in Hydrogen Manufacturing 制氢行业法规符合性解释大语言模型的比较评价

Procedia CIRP Pub Date : 2026-01-01 Epub Date: 2026-09-03 DOI: 10.1016/j.procir.2026.03.307

Seval Eroglu Tanay, John Patsavellas, Konstantinos Salonitis

{"title":"Comparative Evaluation of Large Language Models for Regulatory Compliance Interpretation in Hydrogen Manufacturing","authors":"Seval Eroglu Tanay, John Patsavellas, Konstantinos Salonitis","doi":"10.1016/j.procir.2026.03.307","DOIUrl":"10.1016/j.procir.2026.03.307","url":null,"abstract":"<div><div>The rapid advancement of artificial intelligence (AI) is reshaping the competency required of the manufacturing workforce, particularly in regulation-intensive sectors such as hydrogen energy production. This study investigates how large language models (LLMs) can support engineers and technical staff in interpreting complex regulatory directives, with a focus on the Renewable Energy Directive (RED III) and the Atmosphères Explosibles (ATEX) Directive under European Union (EU) law. A comparative evaluation was conducted using three widely used LLMs, ChatGPT-4, Gemini, and Copilot, by analyzing their responses to compliance-related prompts derived from hydrogen manufacturing scenarios. The generated outputs were assessed by domain experts using a Likert-scale evaluation based on three criteria: clarity, relevance, and usefulness. In addition, automated readability metrics were applied to evaluate the accessibility of AI-generated regulatory explanations for engineering practitioners. The results indicate that LLMs can significantly support regulatory interpretation by transforming complex directive language into more accessible explanations for engineers. Among the evaluated models, Copilot achieved the highest readability and usefulness scores, while ChatGPT-4 demonstrated the most consistent performance across evaluation metrics. However, variability in response quality highlights the need for human expert oversight when applying AI tools in safety-critical regulatory contexts. These findings contribute to the growing body of research on AI-assisted regulatory compliance and highlight the potential of LLMs to support workforce upskilling and knowledge accessibility in manufacturing environments.</div></div>","PeriodicalId":20535,"journal":{"name":"Procedia CIRP","volume":"146 ","pages":"Pages 144-149"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148939419","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

CogAssem: A VLM-Driven Architecture Bridging Cognition and Execution for Intelligent Robotic Assembly in Flexible Manufacturing 柔性制造中智能机器人装配认知与执行的vlm驱动体系结构

Procedia CIRP Pub Date : 2026-01-01 Epub Date: 2026-10-05 DOI: 10.1016/j.procir.2026.02.130

Wencan Pei, Cong Ma, Xinyu Li, Liang Gao, Yiping Gao, Chunjiang Zhang

{"title":"CogAssem: A VLM-Driven Architecture Bridging Cognition and Execution for Intelligent Robotic Assembly in Flexible Manufacturing","authors":"Wencan Pei, Cong Ma, Xinyu Li, Liang Gao, Yiping Gao, Chunjiang Zhang","doi":"10.1016/j.procir.2026.02.130","DOIUrl":"10.1016/j.procir.2026.02.130","url":null,"abstract":"<div><div>Intelligent robotic assembly is increasingly essential for flexible manufacturing characterized by high variability in products and processes. As the dominant learning paradigm in robotic manipulation, reinforcement learning (RL) has shown strong capability in acquiring contact-rich assembly skills. However, RL-based methods mainly focus on motion control under fixed task settings and lack cognitive reasoning and planning abilities. Consequently, these methods cannot reason about variations in flexible manufacturing by perception understanding, nor can they adaptively combine reconfigurable assembly skills to meet flexible requirements, which renders current systems incapable of achieving intelligent assembly in flexible manufacturing that requires cognition-guided execution. To address these challenges, we propose CogAssem, a hierarchical architecture for intelligent robotic assembly in flexible manufacturing. It integrates a VLM-based cognitive chain of thought (CoT), a skill-adaptive execution module with online correction, and an experiential knowledge memory module, linked by an assembly operable representation (AOR). CogAssem establishes a closed cognition-execution loop in which VLM-based cognition guides adaptive skill execution through AOR, while successful executions contribute AOR as experiential knowledge that improves cognition for similar tasks, effectively bridging high-level cognition with low-level execution. Extensive experiments on multi-step peg-insertion tasks demonstrate CogAssem significantly outperforms existing baselines and achieves robust generalization to unseen peg shapes, reducing latency by approximately 39.8% through knowledge reuse and improving the success rates by 30% through online correction. These results validate the effectiveness of CogAssem, which marks a promising step toward intelligent robotic assembly.</div></div>","PeriodicalId":20535,"journal":{"name":"Procedia CIRP","volume":"147 ","pages":"Pages 135-140"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149095483","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Parameter optimisation for robotic disassembly of snap-fit plug connectors in end-of-life electric vehicle batteries 报废电动汽车电池卡扣式插头连接器机器人拆卸参数优化

Procedia CIRP Pub Date : 2026-01-01 Epub Date: 2026-10-05 DOI: 10.1016/j.procir.2026.02.191

Jun Huang, Mengying He, Quanyong Huang, Yilin Fang, Quan Liu, Zude Zhou, Huawei Song, Xiuyi Ao

{"title":"Parameter optimisation for robotic disassembly of snap-fit plug connectors in end-of-life electric vehicle batteries","authors":"Jun Huang, Mengying He, Quanyong Huang, Yilin Fang, Quan Liu, Zude Zhou, Huawei Song, Xiuyi Ao","doi":"10.1016/j.procir.2026.02.191","DOIUrl":"10.1016/j.procir.2026.02.191","url":null,"abstract":"<div><div>With the growth of the new energy sectors, recycling and remanufacturing end-of-life (EoL) electric vehicle (EV) batteries have become increasingly important. Disassembly is the initial step in this process. Snap-fit plug connectors are common components in EV batteries, which are currently disassembled manually. The accuracy of the machine vision system’s positioning affects the precision of robotic disassembly. This paper introduces a disassembly parameter optimisation method based on the Bees algorithm (BA) for robotic disassembly of snap-fit plug connectors. By combining six-axis force/torque sensor data with a limited set of experimental samples, the search direction could be iteratively optimised from the initial contact point to an effective disassembly pose of the robotic tool. Experimental results show that the method converges faster than that using proximal policy optimisation (PPO) with the same number of iterations. The proposed approach could improve disassembly robustness and adaptability, enabling the automation of snap-fit plug connector disassembly.</div></div>","PeriodicalId":20535,"journal":{"name":"Procedia CIRP","volume":"147 ","pages":"Pages 559-564"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149095791","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

An Intelligent Assembly Process Planning Framework Integrating Process Knowledge Graph and Large Language Model 集成工艺知识图和大语言模型的智能装配工艺规划框架

Procedia CIRP Pub Date : 2026-01-01 Epub Date: 2026-10-05 DOI: 10.1016/j.procir.2026.02.178

Yongqiang Wan, Zhicheng Huang, Chao Chen, Xinzheng Xu, Zhujia Li

{"title":"An Intelligent Assembly Process Planning Framework Integrating Process Knowledge Graph and Large Language Model","authors":"Yongqiang Wan, Zhicheng Huang, Chao Chen, Xinzheng Xu, Zhujia Li","doi":"10.1016/j.procir.2026.02.178","DOIUrl":"10.1016/j.procir.2026.02.178","url":null,"abstract":"<div><div>Traditional assembly process planning is a time-consuming and knowledge-intensive task that heavily relies on the experience and expertise of process designers, making it challenging to improve planning efficiency. To address this issue, this paper proposes an intelligent assembly process planning framework driven by the integration of an assembly process knowledge graph and a Large Language Model (LLM). First, a heterogeneous assembly process knowledge graph encompassing instance, rule, and resource knowledge is constructed to provide comprehensive knowledge support. Then, the knowledge encoded in the assembly process knowledge graph is leveraged to construct instruction fine-tuning dataset, which is then used to fine-tune the LLM for assembly sequence planning, enabling the generation and iterative optimization of assembly sequences. Additionally, a Retrieval-Augmented Generation (RAG) approach is developed to recommend and order appropriate assembly resources, such as tools, equipment, and fixtures, while satisfying the constraints imposed by assembly process design tasks. These recommended resources are incorporated into the generated process sequences to form a complete assembly process plan. Experiments on representative products demonstrate that the proposed framework can efficiently generate high-quality assembly process plans, confirming its validity and feasibility.</div></div>","PeriodicalId":20535,"journal":{"name":"Procedia CIRP","volume":"147 ","pages":"Pages 479-484"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149095861","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Mechanistic study of chip-induced secondary machining on hole-wall morphology in CFRP/Ti stack drilling CFRP/Ti叠层钻进中切屑诱导二次加工对孔壁形貌的机理研究

Procedia CIRP Pub Date : 2026-01-01 Epub Date: 2026-10-05 DOI: 10.1016/j.procir.2026.02.163

Kuangqing Hu, Shunuan Liu

{"title":"Mechanistic study of chip-induced secondary machining on hole-wall morphology in CFRP/Ti stack drilling","authors":"Kuangqing Hu, Shunuan Liu","doi":"10.1016/j.procir.2026.02.163","DOIUrl":"10.1016/j.procir.2026.02.163","url":null,"abstract":"<div><div>CFRP/Ti heterogeneous stacks are extensively applied in critical aircraft structures due to their superior mechanical and physical properties. Stack drilling is an effective approach to improve machining efficiency and hole-making accuracy; however, the pronounced disparity in physical properties between CFRP and Ti alloys poses significant challenges in achieving consistent hole quality during drilling. In this study, the secondary machining effect of Ti alloy chips on the CFRP hole wall during CFRP/Ti stack drilling is systematically investigated. An analytical strategy combining incomplete drilling experiments with energy-dispersive X-ray spectroscopy (EDS) is employed to elucidate the interaction mechanisms between Ti alloy chips and the CFRP hole wall under dry drilling and cryogenic drilling conditions. The results indicate that Ti alloy chips primarily interact with the smeared matrix on the machined CFRP surface. Under dry drilling conditions, matrix failure is dominated by plastic deformation, and the action of Ti alloy chips promotes the formation of a new smeared matrix layer on the hole wall. In contrast, under cryogenic drilling conditions, matrix failure is governed by brittle behavior, whereby the interaction with Ti alloy chips leads to further removal of the original smeared matrix, resulting in a combined morphology characterized by block-like matrix debris and exposed fibers.</div></div>","PeriodicalId":20535,"journal":{"name":"Procedia CIRP","volume":"147 ","pages":"Pages 372-377"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149095925","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Non-destructive Testing System for High-curvature Laminated Composite Workpieces 高曲率层合复合材料工件无损检测系统

Procedia CIRP Pub Date : 2026-01-01 Epub Date: 2026-02-18 DOI: 10.1016/j.procir.2025.09.046

Banghai Zhang , Deteng Liu , Ziyan Luo , Jingde Li , Xiansheng Qin , Zhanxi Wang

引用次数: 0

Embodied AI Agent Framework for (Re)Programming Robotic Tasks in Flexible Assembly using LLMs and VLMs 基于llm和vlm的柔性装配机器人任务（再）编程的嵌入式AI代理框架

Procedia CIRP Pub Date : 2026-01-01 Epub Date: 2026-02-18 DOI: 10.1016/j.procir.2025.08.187

Lucas Manassés Pinheiro de Souza , Abhishek Deshmukh , Kedar Rajpathak , Alexander Moriz , Amon Göppert , Robert H. Schmitt

{"title":"Embodied AI Agent Framework for (Re)Programming Robotic Tasks in Flexible Assembly using LLMs and VLMs","authors":"Lucas Manassés Pinheiro de Souza , Abhishek Deshmukh , Kedar Rajpathak , Alexander Moriz , Amon Göppert , Robert H. Schmitt","doi":"10.1016/j.procir.2025.08.187","DOIUrl":"10.1016/j.procir.2025.08.187","url":null,"abstract":"<div><div>The increasing product variability and skilled labor shortages in manufacturing intensify the need for more flexible and adaptive automation solutions, particularly in assembly systems. However, existing robotic automation typically requires expert-level programming, preventing adaptability to new tasks or product variants. This work presents a modular embodied AI agent framework that enables non-programming operators to implement and reconfigure robotic tasks through intuitive natural language (NL) commands in text or voice format. By integrating open-source Large Language Models (LLMs) and Visual Language Models (VLMs) with a Robot Operating System 2 (ROS2)-based stack, the agent translates user instructions into perception, grasp generation, and motion planning actions. The framework is evaluated in real-world experiments across four scenarios: low-level motion control, vision-guided cable grasping, visual scene feedback, and task-cycle recording with replay for scalable execution. Results show 100% command parsing accuracy for basic motions, reliable visual feedback, and 70% success in dual-cable manipulation, with failures mainly from trajectory planning or self-collisions. VLM inference latency proved highly hardware-dependent, with near real-time performance on high-end GPUs but significant slowdowns on consumer devices, highlighting deployment challenges. These findings show the potential of embodied AI agents to bridge NL interaction with robotic execution, lowering barriers to deploying adaptive, operator-friendly robotic systems in dynamic manufacturing.</div></div>","PeriodicalId":20535,"journal":{"name":"Procedia CIRP","volume":"139 ","pages":"Pages 59-65"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147425106","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0