桥梁结构时频域地震非线性损伤统计诊断研究

Post-earthquake bridge safety assessment is an important and key part of the earthquake relief work, and thus earthquake-induced damage identification is the first vital step for the post-earthquake safety assessment of bridge structure. Hence, there lies great scientific and social significance in the research of basic theory for earthquake-induced bridge damage identification. Based on the bridge dynamic responses caused by the aftershock or ambient excitation, the identification of earthquake-induced nonlinear damage in the hidden parts of the bridge structure is investigated in this application. In light of the weak, nonlinear and non-stationary measured dynamic signals from the bridge structure, the test theory of weak and nonlinear signals is studied to purify and cleanse dynamic testing signals for improving the signal to noise ratio. The adaptive decomposition and time-frequency analysis methods of modern signal processing and analysis are improved to obtain the damage information-rich signals, and nonlinear time-varying bridge system is identified in the time-frequency domain, then in addition the statistical features library of bridge nonlinear time-varying system is set up in time-frequency domain. The semi-supervised statistical learning theory is ameliorated to identify the earthquake-induced damage in bridge structure, the damage indexes selection, sample library construction, selection and optimization of the identification algorithms for earthquake-induced bridge damage occurrence, location and severity detection are studied systematically, then the identification theory and method for earthquake-induced bridge nonlinear damage in time-frequency domain is established. Shaking-table testing model is adopted to verify the proposed methodology, and then finally the statistical identification for earthquake-induced bridge nonlinear damage is achieved. The proposal mainly focuses on the basic scientific problems in earthquake-induced bridge damage identification, and aims to promote the development of bridge damage identification with interdisciplinary nature and provide a solid theoretical basis for post-earthquake bridge structural safety assessment.

震后桥梁安全评估是抗震救灾工作的重要内容与关键环节，而地震引发的桥梁损伤诊断又是安全评估的基础，对桥梁地震损伤诊断的基础理论进行研究具有重要的科学和社会意义。本申请拟从余震或环境激励下的桥梁动力响应出发，对隐秘部位的桥梁地震非线性损伤进行诊断。针对桥梁地震损伤动力测试信号的微弱性、非线性和非平稳性，建立非线性、非平稳微弱信号检验理论并净化测试信号；改进自适应信号分解和时频分析方法，在时频域内识别桥梁结构非线性时变系统并建立其统计特征库；改进半监督统计学习理论，系统研究桥梁地震非线性损伤发生、损伤位置和程度诊断的特征选择、样本库构建、识别算法选择及优化，构建桥梁地震非线性损伤自适应时频域统计诊断理论和方法；通过振动台模型试验验证，实现桥梁地震非线性损伤的时频域统计诊断。本申请研究的是桥梁地震损伤诊断的基础科学问题，研究成果可促进桥梁损伤诊断这一交叉学科的发展，为桥梁震后安全评估提供理论支撑。

桥梁结构时频域地震非线性损伤统计诊断是集微弱信号处理、桥梁工程和机器学习为一体的研究课题，研究内容主要包括非线性非平稳微弱信号自适应处理、桥梁结构时频域特征库构建以及桥梁地震非线性损伤统计诊断。针对桥梁地震损伤特点，从地震引起的桥梁振动响应入手，改进了非线性非平稳微弱测试信号的自适应分解、净化、筛选和重构方法，实现了测试信号的自适应处理，并据此改进时频分析方法，有效识别了地震损伤的非线性时变桥梁系统。在构建时频域统计特征库过程中，为使地震损伤模式符合桥梁实际地震损伤情况，采用结合地震易损性分析的方法确定地震损伤模式，使损伤指标能更好地反映桥梁的损伤状态；对既有特征提取中的不足进行了相应改进，构建了具有可分性良好的特征库。针对现有模式识别存在的模式分类能力不足等问题，为实现对复杂桥梁地震损伤模式的准确识别，将深度学习引入桥梁地震非线性损伤统计诊断研究，构建了堆栈降噪自编码器、联合卷积神经网络&长短期记忆神经网络、多任务多层感知机等深度网络，对桥梁损伤样本库进行了训练和测试。最后依据振动台试验，对所提方法进行验证，实现了桥梁结构地震损伤状态的精确识别。研究成果能快速准确地诊断桥梁结构的地震损伤，可用于实际桥梁地震损伤诊断中，为桥梁震后安全评估提供了理论支撑。