Structural Defect Recognition; Structural Damage Assessment and Story Drift Estimation
指導老師：張家銘 教授
專案經理：許舜翔
結構損傷診斷與層間變位估計 材料受損可從其變形得出相對應的損傷程度。在結構健康監測中，也可以由觀測結構 的層間位移來評斷該樓層是否超過設計限制，並達到損傷標準。然而，位移量測計的 安裝需有參考點的設置，導致在真實結構中難以實現。透過加速規量測，並經過前處 理與分析可以估計動態位移的變化，進而觀測結構物的健康與否。
Structural Damage Assessment and Story Drift Estimation
The seriousness of material damage can be inferred from the corresponding deformation. For structural health monitoring, story drift of structures can also be utilized to examine if the floor exceeds its designed structural capacity and is damaged by the drift. Although displacement meters can directly acquire the data of story drift, the installation that requires an additional setup of reference points can be arduous in real-world cases. On the contrary, the indirect data from accelerometers, which are more easily installed, can be exploited to calculate the displacement through data processing and analysis. Based on the results, the health of structures can finally be estimated.
結構物劣化辨識 發展基於電腦視覺的深度學習模型，涵蓋有影像分類、物件偵測或影像分割等不同定 位精度的模型演算法，針對構造物目視檢測等實務外觀檢查工作，以深度學習模型 (CNN、Transformer 等)或其他影像分析方法(影像拼貼、視角校正等)，輔助工程 師於龐大影像資料中紀錄損害或劣化相關資訊，包含其位置與尺寸，應用之構造物類 別有一般建築、橋梁、隧道等。
Structural Defect Recognition
Our research group focuses on developing deep learning models of popular computer vision topics, including image classification, object detection, and image segmentation. The topics indicate different levels of object localization. The deep learning models (e.g., convolutional neural network (CNN) and Vision Transformer) are combined with digital image processing methods (e.g., image stitching and camera calibration) to improve visual inspection of structures in practice. The vision solutions for identifying defects on the surfaces of structures can help engineers efficiently handle a large amount of image data from visual inspection and obtain information about defects. The information consists of positions and scales of defects. We have applied and examined the above methods in various types of structures, such as buildings, tunnels, and bridges.