Application of Machine Learning (ML) for the Prediction of Stress Concentration and Fatigue Life

Bridges are subjected to cyclic loading during their lifetime. Such cyclic loads can cause fatigue failures in the bridge structure in places with high stress concentration. One of such areas where the stress concentration can be quite high is in the connections. There are mainly two ways of joining steel members, one is by bolting and the other is by welding. When bridges are connected by bolting, during the working life of a bridge, it is expected that any bolted connection may eventually become a bearing-type bolted connection even if it was originally designed to be slip-critical. So, because of this bearing condition, high stresses are created and fatigue failure of the connections are more likely to originate from spots where the bolt bear against the plates or other parts of the connection. Therefore, it is necessary to understand the factors that contribute to increased stress concentration because then we can design bridge connections to avoid or reduce the stress concentration and hence increase the fatigue life of such connections. We can also use the stress concentration to predict the remaining life of bridge connections. So, this research investigated the predictors of the stress concentration factor and fatigue life of a bridge connection using machine learning (ML) algorithms. The dataset used was from testing of several bolted connections, and it contained a total of 31 instances. From the ML models, it was observed that parameters such as the stagger, and the gage distance were predictors of the stress concentration and that the stress concentration can be used to predict the fatigue life of such bolted bridge connection. Furthermore, the research also investigated the best ML algorithm for predicting the stress concentration for such small dataset.