Multiaxial fatigue prediction and uncertainty quantification based on back propagation neural network and Gaussian process regression. (March 2023)

• Record Type: Journal Article
• Authors: Gao, Jingjing; Wang, Jun; Xu, Zili; Wang, Cunjun; Yan, Song
• Is Part Of: International journal of fatigue. Volume 168(2023)
• Publication Date: 2023-03
• Subjects: Multiaxial loading -- Fatigue life prediction -- Back propagation neural network -- Gaussian process regression -- Uncertainty quantification; Materials -- Fatigue -- Periodicals; Materials -- Fatigue; Periodicals; 620.1122
• DOI: 10.1016/j.ijfatigue.2022.107361
• Languages: English
• ISSNs: 0142-1123

Abstract

Highlights: A data-driven method, BPNN-GPR method, is proposed for multiaxial fatigue life prediction. The BPNN-GPR method can simultaneously predict the multiaxial fatigue life and quantify the uncertainty. The BPNN-GPR method has better performance compared with the five conventional methods. The extrapolation ability of the BPNN-GPR method for unknown multiaxial non-proportional loading paths is shown. Abstract: Engineering structures are often suffering multiaxial stress which leads to multiaxial fatigue failures. Accurate and reliable multiaxial life prediction is a challenging issue in fatigue analysis due to the intricate deterioration mechanisms of the fatigue failure and the large uncertainty in the material parameters at the microscopic scale. Moreover, conventional multiaxial fatigue life prediction models are empirical or semi-empirical. To tackle this problem, a data-driven method is presented, which combines the back propagation neural network (BPNN) and the Gaussian process regression (GPR). The BPNN-GPR method can predict the multiaxial fatigue life and quantify the uncertainty simultaneously. This method is validated using six materials including the uniaxial, multiaxial proportional and multiaxial nonproportional loading cases. All the predicted lives fall within a life factor of ± 3, which indicates that the BPNN-GPR method has the satisfactory capability for multiaxial fatigue life prediction. In addition, results also show the prediction capability of the BPNN-GPR method for unknown multiaxial nonproportional loading paths.