Abstract
In this paper, a hybrid computational framework that combines the state-of-the art machine learning algorithm (i.e., deep neural network) and nonlinear finite element analysis for efficient and accurate fatigue life prediction of rotary shouldered threaded connections is presented. Specifically, a large set of simulation data from nonlinear FEA, along with a small set of experimental data from full-scale fatigue tests, constitutes the dataset required for training and testing of a fast-loop predictive model that could cover most commonly used rotary shouldered connections. Feature engineering was first performed to explore the compressed feature space to be used to represent the data. An ensemble deep learning algorithm was then developed to learn the underlying pattern, and hyperparameter tuning techniques were employed to select the learning model that provides the best mapping, between the features and the fatigue strength of the connections. The resulting fatigue life predictions were found to agree favorably well with the experimental results from full-scale bending fatigue tests and field operational data. This newly developed hybrid modeling framework paves a new way to realtime predicting the remaining useful life of rotary shouldered threaded connections for prognostic health management of the drilling equipment.
Recognition of Free-form Features for Finite Element Meshing using Deep Learning
