Automatic Defect Detection in Structures

使用弱监督的锤击试验的自动化首先是在仅使用声学数据的初始框架中建立的。在实验室条件下对混凝土试块进行的实验表明，该方法非常有效地将特征空间转换为适合于区分缺陷和非缺陷锤击样本的特征空间。在这些结果之后，使用摄像机沿着可见的裂缝位置收集锤击样品的位置信息，即结构被锤子击中的位置。这被添加到前面提到的框架中，并为系统提供了关于其缺陷检测能力的增强的性能以及对人类用户产生的弱监督错误的稳健性。

Automation of the hammering test using weak supervision was first established in an initial framework only using acoustic data. Experiments conducted with concrete test blocks in laboratory conditions showed the approach extremely effective for conducting a transformation of the feature space into one appropriate for discriminating defect and non-defect hammering samples. Following those results, the position information of hammering samples, i.e., where the structure was hit with a hammer, was collected using a camera along visible crack positions. This was added into the previously mentioned framework and provided the system increased performance regarding its defect detection abilities as well as robustness against errors in weak supervision produced by human user.