特种设备的安全运行关乎国计民生，失效分析是保障其本质安全的重要手段。针对传统失效分析高度依赖专家经验、主观性强且知识传承困难等局限，本文旨在系统介绍人工智能技术在特种设备失效分析领域的应用现状与发展路径，以期为提升失效分析的客观性与智能化水平提供理论参考。通过梳理现有研究，本文从3大关键技术切入：一是基于计算机视觉与卷积神经网络的断口智能识别，实现失效模式的快速客观分类；二是基于知识工程的失效原因分析，利用知识图谱整合碎片化领域知识，构建可查询、可推理的知识系统；三是基于多源信息融合与知识图谱推理的失效智能预测，实现从“被动查询”到“主动预测”的跨越。在此基础上，重点介绍了北京航空航天大学研究团队在失效断口智能识别分类与管线钢氢脆知识图谱构建这2项前沿工作中的具体实践成果。结果表明，基于改进的双分支网络模型可显著提升典型断口形貌的识别精度，而构建的管线钢氢脆知识图谱成功将散落文献中的隐性知识转化为结构化、可查询的显性知识系统。人工智能技术正推动失效分析从依赖个人经验的人工模式迈向数据驱动与知识引导相结合的新型范式。未来发展方向在于推动“感知-认知-决策”技术的深度融合，构建以智能体为核心的一体化智能失效分析系统，并倡导行业共建开放协同的数据与标准基础设施。
Abstract
The safe operation of special equipment is crucial to the national economy and people’s livelihoods, with failure analysis serving as a core means to ensure its intrinsic safety. Traditional failure analysis heavily relies on expert experience, suffering from limitations such as strong subjectivity and difficulties in knowledge transfer. This paper aims to systematically introduce the application status and development pathways of artificial intelligence (AI) technology in the field of special equipment failure analysis, providing a theoretical reference for enhancing the objectivity and intelligence level of failure analysis. By synthesizing existing research, this paper examines three key technological approaches: Firstly, intelligent fracture recognition based on computer vision and convolutional neural networks, enabling rapid and objective classification of failure modes. Secondly, failure cause analysis based on knowledge engineering, utilizing knowledge graphs to integrate fragmented domain knowledge and construct queryable and inferable knowledge systems. Thirdly, intelligent failure prediction based on multi-source information fusion and knowledge graph reasoning, achieving a leap from “passive query” to “active prediction”. Building on this foundation, the paper highlights specific practical achievements of the Beihang University research team in two advanced areas: intelligent recognition and classification of fracture surfaces and the construction of a knowledge graph for hydrogen embrittlement in pipeline steels. The results demonstrate that an improved dual-branch network model significantly enhances the recognition accuracy of typical fracture morphologies, while the constructed knowledge graph successfully transforms tacit knowledge scattered across the literature into a structured, queryable explicit knowledge system. This paper concludes that AI technology is driving failure analysis from a traditional model relying on personal experience toward a novel paradigm combining data-driven approaches with knowledge guidance. Future development directions lie in promoting the deep integration of “perception-cognition-decision” technologies, constructing integrated intelligent failure analysis systems centered on intelligent agents, and advocating for industry-wide collaboration in building open and synergistic data and standard infrastructure.
关键词
特种设备 /
失效分析 /
人工智能 /
断口识别 /
知识图谱 /
失效预测
Key words
Special equipment /
Failure analysis /
Artificial intelligence /
Fracture recognition /
Knowledge graph /
Failure prediction
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 国家市场监督管理总局.市场监管总局关于2024年全国特种设备安全状况的通报[EB/OL]. (2025-05-07)[2026-03-04]. https://www.samr.gov.cn:5100/tzsbj/qktb/tb/art/2025/art_1b31b22095744942af16cf24cec957ac.html.
[2] 吴洋,郑宏伟,程佳,等.人工智能大语言模型在特种设备检验检测中的应用探讨[J].中国特种设备安全,2025,41(S1):122-126.
[3] 国务院江苏响水“3·21”特别重大爆炸事故调查组. 江苏响水天嘉宜化工有限公司“3·21”特别重大爆炸事故调查报告[R]. 北京: 应急管理部, 2019.
[4] 深圳欢乐谷“10·27”设备碰撞事故调查组. 广东省深圳市“10·27”一般大型游乐设施碰撞事故调查报告[R].深圳:深圳市人民政府,2024.
[5] 宋明大,曹怀祥,汪立新.承压设备失效分析思路及方法[J].中国特种设备安全,2010,26(12):9-12.
[6] 应急管理部办公厅. 应急管理部办公厅关于河南省三门峡市河南煤气集团义马气化厂 “7·19”重大爆炸事故的通报（应急厅函〔2019〕447号）[EB/OL]. (2019-07-25)[2026-03-16]. https://www.mem.gov.cn/gk/tzgg/tb/201907/t20190726_325359.shtml.
[7] 闫涵.金属断口图像识别方法研究[D].大连：大连交通大学,2020.
[8] 罗艳波,赵登超,袁玉杰.高精度图像识别关键技术应用进展与挑战[J].计算机时代,2026(2):7-10.
[9] LECUN Y, BENGIO Y, HINTON G. Deep learning[J].Nature, 2015, 521(7 553): 436-444.
[10] 王楚涵.基于N-Net 的金属断口图像识别及FPGA硬件实现[D].大连：大连交通大学,2023.
[11] YAN H, ZHONG C Q, LU W, et al. Metal fracture recognition: a method for multi-perception region of interest feature fusion[J]. Applied Intelligence, 2023, 53(20): 23 983-24 007.
[12] OSOVSKI S, TSOPANIDIS S. Unsuper vised machine learning in f ractography: Evaluat ion and interpretation[J]. Materials Characterization,2020,182:111 551.
[13] 郑强,许振彬.面向特种设备的大语言模型-知识图谱双向推理优化与幻觉抑制方法[J].数据采集与处理, 2025,40(3):647-658.
[14] 毕鑫,聂豪杰,赵相国,等.面向知识图谱约束问答的强化学习推理技术[J].软件学报,2023,34(10):4 565-4 583.
[15] 沈华,熊开宇,闫斌,等.知识图谱的用户兴趣向量化方法及应用[J].南昌大学学报(理科版),2020,44(6):610-616.
[16] 沈英汉,江旭晖,王元卓,等.时态知识图谱的推理研究综述[J].计算机学报,2023,46(6):1 272-1 301.
[17] 杨观赐,许彪,罗可欣,等.知识图谱技术综述:构建,推理及典型应用[J].贵州大学学报（自然科学版）,2025,42(2):1-10.
[18] 许文倩,黄栋,陈照春,等.基于特种设备知识库的AI交流机器人研究与设计[J].中国特种设备安全,2023,39(2):26-29.
[19] 黄小双,刘必林,张英,等.知识图谱技术在渔业领域中应用研究进展[J/OL].上海海洋大学学报,1-17[2026-04-03].http://link.cnki.net/urlid/31.2024.s.20260108.1400.008.
[20] 李迦龙,郭中华,周鹏.基于医学知识图谱的智能医疗系统[J].西北工程技术学报（中英文）, 2025,24(4):365-371.
[21] 雷菁.知识图谱AI技术在计算机领域的应用研究[J].信息与电脑,2026,38(3):56-58.
[22] 金国清.油气管道完整性管理的知识图谱研究[D].北京：中国石油大学(北京),2021.
[23] 姚金楠,杨新健,刘立磊.基于大模型RAG应用的特种设备科普AI问答数字人的研发[J].中国特种设备安全,2025,41(S1):93-97.
[24] LIU C, LI X, GE J, et al. A deep learning framework based on attention mechanism for predicting the mechanical properties and failure mode of embedded wrinkle fiber-reinforced composites[J].Composites Part A, 2024, 186:108 401.
[25] 马高，王瑶.基于机器学习的钢管混凝土剪力墙破坏模式预测与解释[J].地震工程与工程振动，2022， 42（3）：143-152.
[26] BAO H, WU S, WU Z, et al. A machine-learning fatigue life prediction approach of additively manufactured metals[J]. Engineering Fracture Mechanics, 2020,242:107 508.
[27] BHANDARI U， RAFI M R， ZHANG C Y， et al. Yield strength prediction of high-entropy alloys using machine learning[J]. Materials Today Communications， 2021， 26: 101 871.
[28] 李绪尧.基于机器学习的腐蚀管道剩余强度预测[D].杭州：杭州电子科技大学，2024.
[29] LI Z, ZHAO T, ZHANG J, et al. Fusing image and physical data for fatigue life prediction of nickel-based single crystal superalloys[J]. Engineering Failure Analysis, 2024, 162:108 343.
[30] 李桌汉,有移亮,赵子华,等.人工智能技术在失效分析领域的应用[J].航空材料学报,2024,44(5):1-16.
基金
国家科技重大专项（2025ZD0619400）
