1.National Key Laboratory of Aerospace Structure Mechanics and Control, Nanjing University of Aeronautics and Astronautics, Nanjing210016, China
2.National Key Laboratory of Strength and Structural Integrity, Aircraft Strength Research Institute of China, Xi’an710065, China
Citations
LÜ Jialong, XU Yusen, CHEN Jian, et al. Guided-wave-based quantile regression neural network for crack diagnosis in multi-fastener structures[J]. Aeronautical Manufacturing Technology, 2025, 68(21): 104–113.
Abstract
Accurate crack diagnosis in multi-fastener metallic structures is critical for instructing aircraft structural ground tests and ensuring in-service safety. However, heteroscedastic uncertainties in the relationship between crack length and guided-wave damage indices severely compromise diagnostic accuracy and reliability assessment. To address this, a multi-fastener-joint crack diagnosis method based on Quantile Regression Neural Network (QRNN) is proposed. The QRNN establishes a mapping model between guided-wave damage index and the crack length, where crack diagnosis result is determined through the median quantile point. Furthermore, by comprehensively leveraging the quantile outputs, the diagnostic reliability across different crack lengths is quantitatively characterized. A complex multi-layer stringer structure with multiple fastener joints was adopted as the testbed to validate the diagnostic capability and reliability assessment. Experimental results indicate that the proposed approach enables precise crack diagnosis in characteristic longeron fastener-joint areas, exhibiting Root Mean Squared Error (RMSE) 1.2 mm in the skin and RMSE of 2.2 mm in the stringer, with concurrent quantification of diagnostic reliability.
民用飞机机身、机翼等部件通常采用多种结构元件(如蒙皮、搭接板、长桁、加强片),通过多排铆钉连接的方式层叠装配而成[ SKORUPA A, SKORUPA M. Riveted lap joints in aircraft fuselage: Design, analysis and properties[M]. Dordrecht: Springer Netherlands, 2012. 1]。这些多钉连接部位会由于钉孔等部位出现应力集中,导致在交变载荷作用下容易出现疲劳裂纹损伤[ 孙侠生, 苏少普, 孙汉斌, 等. 国外航空疲劳研究现状及展望[J]. 航空学报, 2021, 42(5): 524791.SUN Xiasheng, SU Shaopu, SUN Hanbin, et al. Current status and prospect of overseas research on aeronautical fatigue[J]. Acta Aeronautica et Astronautica Sinica, 2021, 42(5): 524791. 2]。并且这些部件通常封闭且不可拆卸,难以通过传统无损检测手段进行损伤检测。结构健康监测(Structural health monitoring,SHM)[ OSTACHOWICZ W, GÜEMES J A. New trends in structural health monitoring[M]. Vienna: Springer Vienna, 2013. 3]通过集成在结构内部或表面的传感器获取与结构健康状态相关的信息,在线反映当前结构实际损伤状态,相较于传统无损检测方法更具有重要优势[ 田童, 李建乐, 邓德双, 等. 飞行器结构健康监测技术研究进展[J]. 航空制造技术, 2024, 67(13): 41–67, 98.TIAN Tong, LI Jianle, DENG Deshuang, et al. Research progress of structural health monitoring technology for aircraft[J]. Aeronautical Manufacturing Technology, 2024, 67(13): 41–67, 98. 4]。通过SHM技术准确获取多钉结构连接部位裂纹损伤的萌生和尺寸,对于指导民机结构地面试验和保障结构在役安全具有重要意义。
在结构健康监测领域,已发展出多种裂纹监测方法[ 王彬文, 肖迎春, 白生宝, 等. 飞机结构健康监测与管理技术研究进展和展望[J]. 航空制造技术, 2022, 65(3): 30–41.WANG Binwen, XIAO Yingchun, BAI Shengbao, et al. Research progress and prospect of aircraft structural health monitoring and management technology[J]. Aeronautical Manufacturing Technology, 2022, 65(3): 30–41. QING X L, LI W Z, WANG Y S, et al. Piezoelectric transducer-based structural health monitoring for aircraft applications[J]. Sensors, 2019, 19(3): 545. 5-6]。其中,压电导波SHM方法通过压电元件在结构中激发导波,根据导波传播特性的变化识别损伤,具有区域监测、不可达部位监测等优势[ 王梓, 张少东, 胥静, 等. 金属结构裂纹损伤Lamb波定量化成像方法[J]. 振动, 测试与诊断, 2023, 43(6): 1183–1190.WANG Zi, ZHANG Shaodong, XU Jing, et al. Lamb wave based crack damage quantitative imaging method for metal material structure[J]. Journal of Vibration, Measurement & Diagnosis, 2023, 43(6): 1183–1190. 王化吉, 施利明, 戴玉山, 等. 直升机关键连接结构孔边裂纹导波监测方法[J/OL]. 机械强度, 2025: 1–8. [2025–07–08]. https://kns.cnki.net/kcms/detail/41.1134.TH.20250708.0939.002.html.WANG Huaji, SHI Liming, DAI Yushan, et al. Monitoring method-based guided wave for hole-edge crack in key attachment structure of helicopter[J/OL]. Journal of Mechanical Strength, 2025: 1–8. [2025–07–08]. https://kns.cnki.net/kcms/detail/41.1134.TH.20250708.0939.002.html. 杨斌, 胡超杰, 轩福贞, 等. 基于超声导波的压力容器健康监测Ⅰ: 波传导行为及损伤定位[J]. 机械工程学报, 2020, 56(4): 1–10.YANG Bin, HU Chaojie, XUAN Fuzhen, et al. Structural health monitoring of pressure vessel based on guided wave technology. Part Ⅰ: Wave propagating and damage localization[J]. Journal of Mechanical Engineering, 2020, 56(4): 1–10. 余孙全, 樊程广, 张翔, 等. 航天器结构中导波健康监测技术的若干进展[J]. 宇航学报, 2024, 45(4): 487–498.YU Sunquan, FAN Chengguang, ZHANG Xiang, et al. Guided waves-based structural health monitoring techniques for spacecraft: A review[J]. Journal of Astronautics, 2024, 45(4): 487–498. 7-10],非常适用于多钉结构这种损伤部位多、传感器布置空间少的结构。虽然基于压电导波的裂纹监测已有较多研究,但面向多钉结构的探索较少。Ihn等[ IHN J B, CHANG F K. Detection and monitoring of hidden fatigue crack growth using a built-in piezoelectric sensor/actuator network: II. Validation using riveted joints and repair patches[J]. Smart Materials and Structures, 2004, 13(3): 621–630. 11]采用压电传感器激励导波,通过损伤因子描述裂纹变化,研究了两块铝板搭接结构裂纹扩展。Yang等[ YANG J S, HE J J, GUAN X F, et al. A probabilistic crack size quantification method using in situ Lamb wave test and Bayesian updating[J]. Mechanical Systems and Signal Processing, 2016, 78: 118–133. 12]使用导波振幅和相位变化来表示两块板组成的铆接搭接结构中铆钉孔的裂纹扩展情况。Chen等[ CHEN J, WU W Y, REN Y Q, et al. Fatigue crack evaluation with the guided wave-convolutional neural network ensemble and differential wavelet spectrogram[J]. Sensors, 2022, 22(1): 307. 13]采用卷积神经网络处理导波信号,实现了铝板搭接结构裂纹诊断。Asadi等[ ASADI S, KHODAEI Z S, ALIABADI M H, et al. A baseline free methodology for crack detection in metallic bolted joints[J]. Advances in Fracture and Damage Mechanics, 2023, 2848: 020038. 14]研究了无基准的导波监测方法,用于加筋铝板的铆钉孔裂纹萌生监测。Liao等[ LIAO W L, SUN H, QIN X L, et al. A novel damage index integrating piezoelectric impedance and ultrasonic guided wave for damage monitoring of bolted joints[J]. Structural Health Monitoring, 2023, 22(5): 3514–3533. 15]提出了一种结合压电阻抗和导波的损伤指标,用于螺栓连接区域的裂纹评估。Chen等[ CHEN J, XU Y S, YUAN S F, et al. Guided wave characteristic research and probabilistic crack evaluation in complex multi-layer stringer splice joint structure[J]. Sensors, 2023, 23(22): 9224. 16]通过试验与数值模拟,对多层长桁接头的导波传播特性开展了系统研究,提出适用于此类复杂结构的传感器布局准则,并建立了基于高斯过程模型的概率诊断方法。以上研究均证明了导波SHM面向多钉连接结构裂纹损伤诊断的应用潜力。但是这些研究大多采用带孔板或者简化的连接结构进行,通过提取导波信号的损伤因子进行裂纹损伤萌生的识别,或者采用多元线性回归进行损伤因子的拟合以进行定量化诊断。然而对于真实的复杂多钉连接结构,其多层结构特征、预紧力差异、以及多个铆钉和铆钉孔的耦合,使得结构裂纹的导波损伤因子具有强分散性,并且分散性的程度也会随裂纹尺寸的变化而变化,即裂纹长度–损伤因子具有异方差性。这种裂纹长度–损伤因子的异方差不确定性将严重影响结构裂纹诊断,以及对诊断结果的可靠性评估。因此,在进行损伤定量化诊断以及评估诊断结果的可靠性时必须考虑异方差性的问题。
对于考虑异方差的概率建模方法,其核心思想是通过增加一个模型对数据方差的变化建模表征。20世纪80年代,Bollerslev等[ BOLLERSLEV T. Generalized autoregressive conditional heteroskedasticity[J]. Journal of Econometrics, 1986, 31(3): 307–327. 17]提出了自回归条件异方差模型和广义自回归条件异方差模型。此后,在这两个模型的基础上,多种改进模型被提出[ PETRICĂ A C, STANCU S. Empirical results of modeling EUR/RON exchange rate using ARCH, GARCH, EGARCH, TARCH and PARCH models[J]. Romanian Statistical Review, 2017, 65(1): 57–72. 18]。此外,异方差高斯过程模型[ BINOIS M, GRAMACY R B, LUDKOVSKI M. Practical heteroscedastic Gaussian process modeling for large simulation experiments[J]. Journal of Computational and Graphical Statistics, 2018, 27(4): 808–821. 19]、分位数回归神经网络(Quantile regression neural network,QRNN)等无参数估计的概率统计模型也被应用于异方差建模。其中,QRNN结合了神经网络和分位数回归两个方面的优势,通过神经网络隐式学习输入–输出间的复杂映射,在无需预设函数形式的条件下即可获取准确的结果,在金融市场股票指数估计和预测、风力发电功率预测等领域有较多的应用,但在结构健康监测领域的研究还很少,特别是多钉结构裂纹诊断及可靠性评估的研究基本没有。
前文所述QRNN的分位点输出结果表征了诊断结果的不确定性分布。给定一个结构裂纹监测阈值,当监测结果大于阈值时,表示这个损伤可以被检出。定义结构存在损伤,系统诊断到损伤的概率为检出率(Probability of detection,POD),则此时POD定义为诊断结果大于检测阈值的概率,即图3中大于检测阈值的损伤概率密度函数曲线所围区域的面积。通常,该损伤概率密度被假设为高斯分布,因此POD可以用式(9)计算得出。
Fig.10 Architectural diagram of QRNN model for crack diagnosis
模型训练在搭载Windows 11操作系统的计算平台上进行,使用PyTorch 2.0作为模型框架,选择了自适应矩估计(Adaptive moment estimation,Adam)优化器,该优化器在实践中常常能够实现较快的收敛。在本研究中,设置学习率为0.01作为每次更新时步长的大小。本文将所有样本数据的80%作为模型的训练集数据,20%作为模型的测试集数据。数据集划分采取手动划分模式,将试件1~8的损伤因子–裂纹长度数据作为训练数据,试件9、10作为验证试件。观测τ={0.05,0.25,0.5,0.75,0.95}这5个目标分位点处的输出,训练迭代次数为500次,每次训练过程中使用批量大小32样本进行参数更新。对于损伤D1,其QRNN裂纹诊断模型训练以及分位点输出结果示意模型训练结果如图11所示。
图11 损伤D1的QRNN裂纹诊断模型训练以及分位点输出结果示意
Fig.11 Training and quantile output results of the QRNN crack diagnosis model for damage D1
SKORUPAA, SKORUPAM. Riveted lap joints in aircraft fuselage: Design, analysis and properties[M]. Dordrecht: Springer Netherlands, 2012.
[2]
孙侠生, 苏少普, 孙汉斌, 等. 国外航空疲劳研究现状及展望[J]. 航空学报, 2021, 42(5): 524791. SUNXiasheng, SUShaopu, SUNHanbin, et al. Current status and prospect of overseas research on aeronautical fatigue[J]. Acta Aeronautica et Astronautica Sinica, 2021, 42(5): 524791.
[3]
OSTACHOWICZW, GÜEMESJ A. New trends in structural health monitoring[M]. Vienna: Springer Vienna, 2013.
[4]
田童, 李建乐, 邓德双, 等. 飞行器结构健康监测技术研究进展[J]. 航空制造技术, 2024, 67(13): 41–67, 98. TIANTong, LIJianle, DENGDeshuang, et al. Research progress of structural health monitoring technology for aircraft[J]. Aeronautical Manufacturing Technology, 2024, 67(13): 41–67, 98.
[5]
王彬文, 肖迎春, 白生宝, 等. 飞机结构健康监测与管理技术研究进展和展望[J]. 航空制造技术, 2022, 65(3): 30–41. WANGBinwen, XIAOYingchun, BAIShengbao, et al. Research progress and prospect of aircraft structural health monitoring and management technology[J]. Aeronautical Manufacturing Technology, 2022, 65(3): 30–41.
[6]
QINGX L, LIW Z, WANGY S, et al. Piezoelectric transducer-based structural health monitoring for aircraft applications[J]. Sensors, 2019, 19(3): 545.
[7]
王梓, 张少东, 胥静, 等. 金属结构裂纹损伤Lamb波定量化成像方法[J]. 振动, 测试与诊断, 2023, 43(6): 1183–1190. WANGZi, ZHANGShaodong, XUJing, et al. Lamb wave based crack damage quantitative imaging method for metal material structure[J]. Journal of Vibration, Measurement & Diagnosis, 2023, 43(6): 1183–1190.
[8]
王化吉, 施利明, 戴玉山, 等. 直升机关键连接结构孔边裂纹导波监测方法[J/OL]. 机械强度, 2025: 1–8. [2025–07–08]. https://kns.cnki.net/kcms/detail/41.1134.TH.20250708.0939.002.html. WANGHuaji, SHILiming, DAIYushan, et al. Monitoring method-based guided wave for hole-edge crack in key attachment structure of helicopter[J/OL]. Journal of Mechanical Strength, 2025: 1–8. [2025–07–08]. https://kns.cnki.net/kcms/detail/41.1134.TH.20250708.0939.002.html.
[9]
杨斌, 胡超杰, 轩福贞, 等. 基于超声导波的压力容器健康监测Ⅰ: 波传导行为及损伤定位[J]. 机械工程学报, 2020, 56(4): 1–10. YANGBin, HUChaojie, XUANFuzhen, et al. Structural health monitoring of pressure vessel based on guided wave technology. Part Ⅰ: Wave propagating and damage localization[J]. Journal of Mechanical Engineering, 2020, 56(4): 1–10.
[10]
余孙全, 樊程广, 张翔, 等. 航天器结构中导波健康监测技术的若干进展[J]. 宇航学报, 2024, 45(4): 487–498. YUSunquan, FANChengguang, ZHANGXiang, et al. Guided waves-based structural health monitoring techniques for spacecraft: A review[J]. Journal of Astronautics, 2024, 45(4): 487–498.
[11]
IHNJ B, CHANGF K. Detection and monitoring of hidden fatigue crack growth using a built-in piezoelectric sensor/actuator network: II. Validation using riveted joints and repair patches[J]. Smart Materials and Structures, 2004, 13(3): 621–630.
[12]
YANGJ S, HEJ J, GUANX F, et al. A probabilistic crack size quantification method using in situ Lamb wave test and Bayesian updating[J]. Mechanical Systems and Signal Processing, 2016, 78: 118–133.
[13]
CHENJ, WUW Y, RENY Q, et al. Fatigue crack evaluation with the guided wave-convolutional neural network ensemble and differential wavelet spectrogram[J]. Sensors, 2022, 22(1): 307.
[14]
ASADIS, KHODAEIZ S, ALIABADIM H, et al. A baseline free methodology for crack detection in metallic bolted joints[J]. Advances in Fracture and Damage Mechanics, 2023, 2848: 020038.
[15]
LIAOW L, SUNH, QINX L, et al. A novel damage index integrating piezoelectric impedance and ultrasonic guided wave for damage monitoring of bolted joints[J]. Structural Health Monitoring, 2023, 22(5): 3514–3533.
[16]
CHENJ, XUY S, YUANS F, et al. Guided wave characteristic research and probabilistic crack evaluation in complex multi-layer stringer splice joint structure[J]. Sensors, 2023, 23(22): 9224.
PETRICĂA C, STANCUS. Empirical results of modeling EUR/RON exchange rate using ARCH, GARCH, EGARCH, TARCH and PARCH models[J]. Romanian Statistical Review, 2017, 65(1): 57–72.
[19]
BINOISM, GRAMACYR B, LUDKOVSKIM. Practical heteroscedastic Gaussian process modeling for large simulation experiments[J]. Journal of Computational and Graphical Statistics, 2018, 27(4): 808–821.