Machine learning method for predicting the fatigue life of sucker rods

Abstract

As an important part of the rod pump production system, the safety of sucker rods directly affects the normal production of the oil production system. In order to predict the fatigue life of defective sucker rods, a novel method based on data-driven machine learning (ML) is proposed. Firstly, the mechanical properties of sucker rods with defects are tested and the damage mechanism of sucker rods is analyzed from the fracture morphology. Then the finite element method (FEM) is used to analyze the stress distribution of the defective sucker rod. Based on the SN curve of sucker rods obtained from the experiment, the fatigue life of defective sucker rods is calculated by the nominal stress method. A large sample dataset consisting of rod diameter, defect diameter, defect depth, axial load and fatigue life is established to construct the training set and test set of ML models. Different from the traditional elastic–plastic mechanics or FE methods, this method can effectively calculate the fatigue life of sucker rods by inputting working parameters. The results show that BPNN has good generalization ability by comparing the prediction results of three ML models, and can effectively predict the fatigue life of defective sucker rods under different working conditions, which provides a feasible method for rod safety monitoring.