基于准分布式长标距FBG和深度信念网络的组合梁损伤识别研究

The conventional damage identification methods and inspection measures have certain drawbacks in the contexts of complex engineering structures and stochastic inspecting environment, and the inspection accuracy is also limited. In this paper, a method of using fiber Bragg grating (FBG) sensors to the damage identification of steel-concrete composite girder, which own high precision and are capable of doing dynamic inspection, is proposed..Defects were deliberately set in different locations in the testing girder to simulate the possible damages that would occur in the different parts of the composite girder. The distributed long-gauge FBG sensors were used to inspect the macro-strain of the girder before and after the damage inspection, and the maximum value of the macro-strain difference before and after the damage inspection were used for damage alarming and positioning. Meanwhile, along with the testing data, the influential degree of a series of parameters on the load bearing capacity of the steel-concrete composite girder was studied, including the decrease of the stiffness of the steel girder, concrete cracking, anti-shearing connection stiffness on different interfaces, and the sliding effect of the steel-concrete interface. .Then, the finite element software was used to establish the model of the girder to do calculation. The testing and calculated results were contrastively analyzed to modify the model, aiming to gain the strains in the testing points under the damaging conditions that were not included in the testing via the calculation done by the most accurate finite element model. Based on which, the deep belief network training samples and the non-parametric damage identification model based on deep belief networks were established, with the intension of using deep neural network to realize the identification and judgment of the targets by directly using the original data of the targets, and applying the damage identification method based on the new theories in artificial intelligence to the steel-concrete composite girder.

针对当前传统的损伤识别方法和测试手段对于实际复杂工程结构和随机的测量环境存在一定的不足，且精度不高，本项目提出将拥有高精度和动态测试能力的光纤光栅传感器应用于组合梁的损伤识别。.通过在试验梁不同的位置设置“缺陷”来模拟组合梁不同部位可能会发生的损伤，通过分布式长标距光纤光栅传感器测试损伤前后的宏应变，利用构件损伤前后的宏应变差值的极大值进行损伤预警和损伤定位。同时结合试验数据，研究钢梁刚度降低、混凝土开裂、不同界面抗剪连接刚度、钢混界面滑移对组合梁各项受力指标的影响程度。.采用有限元计算软件进行建模分析，并将试验结果与计算结果进行对比分析，反复修正计算模型；以最准确的有限元模型计算得到多种损伤状态下的测点应变值，建立深度信念网络训练样本，建立基于深度信念网络的结构非参数化损伤识别模型，通过深层次神经网络直接从原始数据对对象进行认识与判断，将基于人工智能新兴理论的损伤识别方法应用于组合梁。

本项目将环保的橡胶材料对剪力钉根部进行外包，形成橡胶-剪力钉组合剪力连接件，进行推出试验。由于橡胶的设置，橡胶-剪力钉群试件破坏时具有更好的延性，呈弯剪破坏，橡胶-剪力钉群试件的极限抗剪承载力均大于普通剪力钉群的极限抗剪承载力；橡胶降低了剪力钉群的抗剪刚度，使得单枚剪力钉的受力趋于均匀。本次试验的结果为该类橡胶-剪力钉组合剪力连接件中橡胶的设置和应用提供了试验和理论参考。.本项目将普通直形贯穿钢筋弯折后，放入开孔板中，利用钢筋良好的抗拉性能，提高连接件的承载力。在两种不同形态的贯穿钢筋上布置光栅应变传感器，对加载全过程中贯穿钢筋的应变值进行动态监测，用以分析不同形态贯穿钢筋的力学性能差异。详尽研究了两种不同形态贯穿钢筋PBL剪力键的破坏形态、抗剪承载力及荷载-滑移规律。研究结果表明，弯折钢筋PBL剪力键的抗剪承载力较普通PBL剪力键提高了10.42%，抗剪刚度提高了10.67%，其抗剪性能优于传统PBL剪力键。弯折贯穿钢筋能有效的提高PBL剪力键的延性，延长结构使用寿命。.本项目所研究得到的基于深度学习的混凝土表观病害分类器可以针对混凝土单个病害图像进行智能分类，通过预处理对原始数据集的扩充，模型的预训练，再经过迁移学习的优化，准确率达到了91.3%，对混凝土破损露筋病害的识别准确度达到了97.6%，可以满足实际工程中混凝土病害智能检测的需要．在本文的研究工作中还基于Python语言完成了混凝土病害智能识别软件的开发，该软件可对用户输入的多张图片进行批量识别，自动生成文档，且可以运用于移动终端．在后续的研究工作会引入主动学习的方法对数据进行标记，扩大数据集，进一步提高模型的精度。.本项目利用模态曲率差指标成功地识别了结构的损伤区域。根据模态曲率差的峰值，确定了结构的疑似损伤单元。利用改进的BP神经网络，以曲率差输入参数，建立相应的损伤识别程序，对确定的疑似损伤单元进行了精确损伤定位及损伤程度判定。