Statistical Elastodynamic Inversion for Pavement Nondestructive Evaluation

无损评价技术在评价道路、公路和机场路面结构承载能力方面引起了人们的极大兴趣。虽然人们对基于静力分析的反问题进行了大量的研究，但在考虑不确定性的情况下，基于动态分析的反问题研究却很少。该研究将建立一个集统计科学、优化理论和计算机模拟于一体的严密统一的路面无损检测理论框架，并将开发出面向实际工程应用的弹性动力反分析软件工具包。通过利用正演动力分析的整个时程进行反演，所提出的研究将提供更稳定和更准确的结果。通过在贝叶斯决策理论和马尔可夫链蒙特卡罗模拟的框架下，利用最大似然估计及其稳健估计来构造未知参数重构，该研究有望带来巨大的好处。这些措施包括允许自然地处理不确定性，允许减轻测量噪声和异常值的影响，允许将工程师的经验作为先验信息并入，以及允许使用所提供的参数不确定度来定量评估重建参数的准确性和可靠性。通过对各种经典和非经典优化算法和人工神经网络在无损检测中的时间域和变换域的研究，为项目的实施提供高效、准确和稳健的优化方法，并从实测的路面响应中提取更丰富的未知物理性质信息，所提出的方法框架可普遍适用于类似类型的无损检测，从而为其他领域增加新的科学知识。该项目将产生性价比高、快速可靠的路面无损检测技术，将产生显著的经济效益。作为一项跨学科的研究工作，该项目将为学生提供跨学科的接触和在分析、计算和实验工作方面的独特经验。

AbstractNon-destructive evaluation (NDE) has generated great interest in assessing structural capacity of road, highway and airfield pavements. While considerable effort has been devoted to static analysis based inversion, little has been done to investigate inverse problems based on dynamic analysis while taking uncertainty into account. The proposed research will establish a rigorous and unified theoretical framework for pavement NDE that integrates statistical science, optimization theory, and computer simulation, and will develop an elastodynamic inverse analysis software toolkit for practical engineering applications. By using the entire time history of forward dynamic analysis in inversion, the proposed research will provide more stable and more accurate results. By formulating unknown parameter reconstruction using maximum likelihood estimate and its robust counterpart within the framework of Bayesian decision theory and Markov Chain Monte Carlo simulation, the research promises to bring tremendous benefits. These include allowing uncertainty to be treated naturally, allowing the effect of measurement noise and outliers to be mitigated, allowing engineers' experience to be incorporated as a priori information, and enabling the accuracy and reliability of reconstructed parameters to be quantitatively assessed using the provided parameter uncertainty. By investigating a variety of classical and non-classical optimizations algorithm and artificial neural network in the context of NDE in both the time and the transformed domain, the research promises to develop efficient, accurate and robust optimization methods for the implementation of the project, and to retrieve much richer information of unknown physical properties from measured pavement responses.The proposed methodological framework can be generally applicable to similar types of NDE, thereby adding new scientific knowledge to other fields. The project will generate cost-effective, fast and reliable pavement NDE techniques, which will result in significant economic benefits. As an inter-disciplinary research effort, the project will provide students cross-disciplinary exposure and unique experience in analytical, computational and experimental work.