基於標記式與單應式矩陣演算法識別結構應力集中應用於裂縫破壞檢測

Abstract

本文以影像視覺方法進行結構裂縫破壞檢測，提出固定式和非固定相機檢測方法，以全域應變場集中位置有效定位結構裂縫，達到破壞檢測目的，為船體結構監測提出一個全新的方向。 在固定式相機監測方法中，以數位影像相關法建立全域應變場監測材料在降伏前所發生的應變，經過船體局部縮尺結構進行實驗驗證，驗證結果皆可在裂縫發生前預測破壞位置。為了更貼近實際應用層面，本研究探討固定式方法在船體黑暗環境中的使用情況，並進行參數討論提出有效之改善方法，其改善方法透過實驗驗證可行性。 為了達到監測方法不須長期架設固定相機的目的，本研究提出一非固定式相機監測方法，以自主開發之標記式與單應式矩陣演算法校正非固定姿態影像為固定影像。經過實驗驗證，本研究所開發之演算法在不同的相機姿態下成功於結構裂縫破壞前預測並定位裂縫位置，此外亦討論了標記點貼附於結構的使用情境，並透過實驗驗證其貼附式方法之裂縫檢測能力。

In this study, we proposed an image-based method aim to inspect structural crack damage. The method has been separated into fixed and mobile mode camera methods, which both can locate crack position at the concentration position by the structure full-field strain. With the outstanding behavior of proposed damage inspection method, it can point ship structure monitoring to a new direction. In the fixed camera inspection method, the digital image correlation method is used to establish a global strain field, which can measure the strain of the material before yielding. The crack prediction ability of the method is verified by the lab-scale partial ship structure experiment. The verification results can predict the damage position before the occurrence of crack. In order to get closer to the practical application, this study discuss the fixed method in dark environment. This study also proposes effective methods to improve the situations, and the methods had been verified the feasibility by experiment results. In order to achieve the requirement of crack inspection but no need to install the fixed camera, this study proposes a non-fixed camera monitoring method, which can transform the non-fixed pose image into a fixed image by the developed marker-based and homography matrix algorithm. As the result of the lab-scale structure experiment, the developed algorithm successfully predicted and located the crack location before the structural crack occurrence by different camera poses. This study also discusses the situation which the markers is needed to attach on the structure, and the crack inspection ability of attached method had been verified by experiment.