Predicting bearing capacity of angle steel bolted connections using machine learning based on experimental and numerical database

The angle steel members in the transmission tower are mostly connected by bolts. The bearing capacity of angle steel bolted connections is related to many factors (e.g., material parameters, geometry of angel steel, and distribution of bolts) and is difficult to fit into conventional formulas. This paper utilizes machine learning techniques to predict the bearing capacity of angle steel bolted connections with different design parameters. Experimental studies and numerical simulations were performed to investigate the failure modes and bearing capacities, and meanwhile to establish a database for machine learning. Three different machine learning models, i.e., Back Propagation Neural Network (BPNN), Random Forest (RF), and Gradient Boosting (GB), were trained to predict the bearing capacity of steel bolted connections. The predicted accuracy of machine learning models was compared with existing industrial standards. The results show that the failure of angle steel bolted connections mainly happened in the hole wall. The bearing capacity of angle steel bolted connections is mainly influenced by the bolt end distance and bolt edge distance. In the machine learning prediction of bearing capacity, the R-2 correlation coefficients for RF and GB machine learning models, compared to the original data, are all above 0.98. The predicting errors of machine learning models are within 10 %, while the maximal predicting error of existing industrial standards are 38.2 %similar to 58.2 %. This demonstrates the feasibility of the proposed machine learning technology in accurately predicting the bearing capacity of angle steel bolted connections.