Developing an Advanced Hybrid System of Structural Health Monitoring

工程研究界迫切需要一套可靠、高效、稳健的结构健康监测系统来支持桥梁等老化结构的监测。许多基于振动的方法被开发出来用于检测基础设施的损伤，这些方法大致分为基于数据和基于模型的方法。然而，能够考虑环境温度变化和桥梁温度分布不均匀等复杂问题的损伤检测方法仍有待发展。就准确性、损伤的定位和量化以及成本效率而言，没有最好的方法。现有的研究大多集中在一两项任务上，如损伤的检测或损伤的检测与定位。检测、定位、量化和补偿环境影响的最佳解决办法是将数据和基于模型的方法结合起来。这将提供量化的损坏信息，以帮助工程师做出适当的维修决策。这些问题正在推动一种新的集成混合损伤检测系统的开发，该系统可以解决现有方法的缺点。******我的研究项目主要集中在结构损伤检测系统的创新混合方法的发展，结合基于数据和基于模型的方法，以利用两者的固有优势。基于数据的方法使用概率/非概率模式分类器或监督和无监督机器学习方法将损伤敏感特征（如固有频率和模态振型）分类为“完整”或“损坏”。基于模型的方法采用目标结构的物理模型，通常采用有限元方法。通过将这两种方法结合起来并共享来自传感器的测量数据，可以获得一种混合方法，与传统的独立方法相比，该方法将以经济有效的方式提供量化的损伤信息，并进行早期检测和定位。然而，由于噪音、传感器故障和复杂的环境影响，误报的可能性很高，这意味着工程师仍然需要进行现场访问，以确认损坏已经发生。为了应对这一挑战，将开发一种新的基于计算机视觉的损伤检测方法，该方法使用低分辨率相机（类似于智能手机中使用的相机）、先进的图像处理技术和基于数据的方法中使用的机器学习方法，将图像分类为“完整”或“损坏”。这种损伤检测方法最终将与新的混合动力系统相结合。新的混合损伤检测系统与计算机视觉将提供可靠、准确和经济的信息，通过提供结构损伤的明确视觉记录和最佳共享测量传感器数据。

There is an essential and urgent need in engineering research communities for a reliable, efficient and robust structural health monitoring system to support the monitoring of aging structures like bridges. Many different vibration-based methods have been developed to detect damage in infrastructure, which are broadly categorized as data- and model-based approaches. However, damage detection methods that can account for the complexity of such problems as ambient temperature changes and non-uniform distribution of temperature across the bridges, still need to be developed. There is no best method in terms of accuracy, localization and quantification of damage, and cost efficiency. Most existing research has focused on only one or two tasks, such as detection or detection and localization of damage. A preferred solution to detect, localize, quantify and compensate for environmental effects would be to integrate both data and model-based methods. This would provide quantified damage information to help engineers make appropriate maintenance decisions. These concerns are driving the development of a new system of integrated hybrid damage detection systems that can address the shortcomings of existing methods.******My research program is focused on the development of innovative hybrid approaches to structural damage detection systems that combine data- and model-based methods in order to leverage the inherent strengths of both. The data-based method uses probabilistic/non-probabilistic pattern classifiers or supervised and unsupervised machine learning approaches to classify damage-sensitive features such as natural frequencies and mode shapes as ‘intact' or ‘damaged'. The model-based method uses physics model of the target structure, typically applying a finite element method. By combining the two approaches and sharing measured data from sensors, a hybrid method can be obtained that will provide quantified damage information with early detection and localization in a cost-efficient manner as compared with traditional independent approaches. However, the high possibility of false alarms due to noises, sensor malfunctions and complex environmental effects means that engineers would still have to make on-site visits to confirm that damage has occurred. To address this challenge, a novel computer-vision-based damage detection method will be developed using low resolution cameras (similar to those used in smartphones), advanced image processing techniques and machine learning methods used in the data-based approach to classify images as ‘intact' or ‘damaged'. This damage detection method will ultimately be combined with the new hybrid system. The new hybrid damage detection system with computer vision will supply information that is reliable, accurate and cost-efficient by providing an explicit visual record of structural damage and optimally sharing measured sensor data.