Executive Summary : | Connections are the most critical structural components in a structure. Bolted connections have gained popularity and widely been used in metal structures due to the ease of assembly and disassembly. Bolted connections are vulnerable to loosening when subjected to dynamic loads and sometimes even static service loads and hence cause structural failures which sometimes can be catastrophic in nature. Very limited studies in our country were reported wherein the concept of deep learning method combined with machine vision can lead to automated bolt loosening detection without relying on manual feature extraction process. The main objective of the present proposal is to develop the feasible autonomous solution to monitor/ assess the condition of the bolted connections using the machine vision based deep learning techniques to avoid connection failures. The methodology consists of ; the first stage capturing the bolted connection images (using the smart phone camera) of the laboratory test setup and establishing the datasets for healthy and loosened states of the bolts. Loosening of the bolts is induced by rotating the bolts to known rotation angle or number of turns. secondly deep learning algorithm (Faster R-CNN) will be applied to train, test and evaluate the data. The testing and evaluation is done for various shooting distances, shooting angles and different lighting conditions. Finally, this application will be implemented for real time images captured /collected from the drone mounted smart phone. The proposed methodology can be applied to real time bolted steel structure in collaboration with Kirby Building systems, a national level steel structures company specialized in Pre-engineered buildings. |