SGER: A multivariate calibration of St-Id for applications to constructed facilities

摘要：在过去的十年中，结构识别（St-Id）作为结构健康监测（SHM）、基于性能的工程（PBE）和资产管理（AM）的核心和使能概念越来越重要。这三种模式现在被美国国会、联邦政府机构和ASCE认可为改善基础设施性能的关键。St- Id的目标是通过数学模型和实验响应数据的关联来推断结构系统的不可测量属性（如边界和连续性条件、材料刚度、损伤等）。潜在的假设是，结构的可测量响应（例如位移，应变，加速度等）与期望的不可测量属性之间存在足够的相关性，从而可以可靠地推断出它们。虽然被广泛接受，但我们目前对这一关键假设的验证几乎完全是由轶事证据提供的，既不全面也无法量化（例如，模态振型比频率对损伤更敏感）。这种知识差距既维持了对用户直觉和启发式的严重依赖，也排除了对与St-Id应用相关的总不确定性（随机和偏差）的可靠估计。为了解决这一知识差距，该研究被视为一种多变量校准，其中各种St-Id方法可靠地推断所需不可测量属性变化的能力将通过使用基于概率的？可辨认性指数吗?， Id（类似于可靠性指标）。该校准将在Drexel智能基础设施研究所（DI3）实验室的物理模型上进行，该模型的设计和建造与普通板梁桥相似。为了实现校准所需的控制，该模型能够模拟几种常见的边界条件、损伤情景和结构复杂性的来源（例如局部非线性），并能够配置多级冗余和不规则性（例如倾斜）。虽然本研究的主要基本目的是确定“可识别性”。对于基于物理的响应指数（如模态参数）和非基于物理的模型（如神经网络），确定模型更新程序推断不可测量属性绝对值的能力的第二个基本目标也将得到满足。智力价值：本研究的智力价值在于其目的是定量地建立响应指数/识别模型与构建系统的不可测量属性之间的因果关系。这种关系是St-Id过程的基础。尽管迄今为止St-Id的应用已经对这种关系提供了一些见解，但这些见解在本质上既不全面也不定量，这就妨碍了它们的推广。作者认为，这种知识差距代表了从已建成设施中提取可靠且广泛适用的性能指标的根本障碍，这阻碍了SHM， PBE和AM的有意义应用。所提出的努力将是对构建系统的这种关系进行严格调查的第一次尝试，并代表了迈向更可靠和科学的St-Id形式的必要的第一步。更广泛的影响：拟议的研究将使St-Id社区广泛参与第一个损伤检测基准研究，该研究明确表示了建筑设施的现实方面（例如结构复杂性）以及获取数据的现实方法（例如环境监测和爬行速度卡车测试）。这项研究的结果将通过通讯、会议和期刊以及通过开发一个互动网站广泛传播。本研究的综合教育部分将把DI3实验室的物理模型纳入德雷克塞尔大学和卡内基梅隆大学（DU和CMU）的本科生结构分析课程中。这项工作的目标将是展示构建系统的结构行为的不太了解但现实的方面，并提供具体的经验，说明不确定性在实际结构的分析中所起的作用。杜大学的学生将定期访问该模型，以检查结构复杂性对传统结构分析主题（如影响线、弹性挠度的确定等）应用的影响。CMU的学生将研究类似的现象；然而，他们与模型的互动将通过互动网站得到促进。Douglas A. Foutch，结构系统和结构危害缓解项目主管

Abstract:Within the last decade, structural identification(St-Id) has enjoyed increasing significance as a core and enabling concept for structural health monitoring (SHM), performance-based engineering (PBE), and asset management (AM) ? three paradigms that are now recognized by the US Congress, Federal Government Agencies and the ASCE as critical for improving infrastructure performance. The goal of St- Id is to infer un-measurable attributes of structural systems (e.g. boundary and continuity conditions, material stiffness, damage, etc) through the correlation of mathematical models and experimental response data. The underlying assumption is that there is a sufficient correlation between the measurable responses of the structure (e.g. displacement, strain, acceleration, etc.) and the desired un-measurable attributes, such that they can be inferred reliably. Although widely accepted, our present validation of this key assumption is provided nearly exclusively by anecdotal evidence and is neither comprehensive nor quantifiable (e.g. mode shapes are more sensitive than frequencies to damage). This knowledge gap both sustains a heavy reliance on user intuition and heuristics, and precludes reliable estimates of the total uncertainty (random and bias) associated with applications of St-Id.To address this knowledge gap, the proposed research is cast as a multivariate calibration in which the ability of various St-Id approaches to reliably infer changes in desired un-measurable attributes will be examined and quantified through the use of a probability-based ?identifiability index?, Id (analogous to the reliability index). This calibration will be carried out on a physical model in the laboratory at the Drexel Intelligent Infrastructure Institute (DI3), which was designed and constructed with similitude to common slab-on-girder bridges. To enable the control needed for this calibration, the model was fabricated with the ability to simulate several common boundary conditions, damage scenarios, and sources of structural complexities (e.g. local nonlinearities) as well as the ability to be configured with multiple levels of redundancies and irregularities (such as skew). Although the primary fundamental objective of this study is to determine the ?identifiability? of physics-based response indices (e.g. modal parameters) and non-physics- based models (e.g. neural networks), a secondary fundamental objective of identifying the ability of model updating procedures to infer absolute values of un-measurable attributes will also be satisfied.Intellectual Merit: The intellectual merit of this study lies in its aim to quantitatively establish the causal relationship between response indices/identified models and un-measurable attributes of constructed systems ? a relationship that serves as the underpinning for the St-Id process. While St-Id applications to date have provided some insight into this relationship, these insights are neither comprehensive nor quantitative in nature, which precludes their generalization. The authors believe that this knowledge gap represents a fundamental barrier to the extraction of reliable and widely applicable performance measures from constructed facilities, which hinders meaningful applications of SHM, PBE, and AM. The proposed effort will be the first attempt to rigorously investigate this relationship for constructed systems and represents a necessary first-step towards a more reliable and scientific form of St-Id.Broader Impacts: The proposed research will engage the St-Id community at large in the first damage detection benchmark study that explicitly represents realistic aspects of constructed facilities (e.g. structural complexities) as well as realistic approaches to acquiring data (e.g. ambient monitoring and crawl-speed truck testing). The results of this research will be broadly disseminated through newsletters, conferences and journals as well as through the development of an interactive website. The integrated educational component of this study will incorporate the physical model in the DI3 laboratory into undergraduate structural analysis courses at both Drexel and Carnegie Mellon Universities (DU and CMU). The objective of this effort will be to demonstrate less-understood but realistic aspects of the structural behavior of constructed systems and provide concrete experiences that illustrate the role uncertainty plays in the analysis of real structures. Students at DU will visit the model periodically to examine the impact of structural complexities on the application of traditional structural analysis topics such as influence lines, determination of elastic deflections, etc. Students at CMU will examine similar phenomena; however, their interaction with the model will be facilitated through the interactive website.Douglas A. Foutch, Program Director, Structural Systems and Hazard Mitigation of Structures