This study presents a novel autonomous vision-based bolt-looseness detection method for splice bolted connections. The method is sequentially designed with a Faster regional convolutional neural network-based bolt detector, an automatic distortion corrector, an adaptive bolt-angle estimator, and a bolt-looseness classifier. The robustness of the method is demonstrated by detecting loosened bolts in a lab-scale bolted joint under sharp capturing angles and different lighting conditions. Next, the method is applied to detect loosened bolts in a realistic joint of the Dragon Bridge in Danang, Vietnam. The bolt detector shows the training, validation, and testing accuracy of 98.85%, 97.48%, and 93%, respectively. Loosened bolts in the lab-scale and real-scale joints are well detected with precisely-estimated loosening severities, even for a sharp perspective angle. The method also shows a high level of adaptability with low-brightness images. Therefore, the method has great potentials for autonomous monitoring of in-situ bolted connections.

这项研究提出了一种新颖的基于自主视觉的用于螺栓连接的螺栓松动检测方法。该方法是使用基于Faster区域卷积神经网络的螺栓检测器，自动变形校正器，自适应螺栓角度估计器和螺栓松动度分类器顺序设计的。通过在锐利的捕获角度和不同的照明条件下检测实验室规模的螺栓连接中松动的螺栓，证明了该方法的鲁棒性。接下来，该方法用于检测越南岘港龙桥的真实接缝处的螺栓松动。螺栓检测器的训练，验证和测试准确度分别为98.85％，97.48％和93％。通过精确估计的松动强度，可以很好地检测到实验室级和实际级接头的螺栓松动，即使是锐利的视角 该方法还显示出对低亮度图像的高度适应性。因此，该方法具有自主监测原地螺栓连接的巨大潜力。