loading page

A physically informed neural network based on Basquin's formula for predicting fatigue life
  • +1
  • Afang JIN,
  • Dexin Yang,
  • Yong Dai,
  • Xiaochen Jia
Afang JIN
Xinjiang University

Corresponding Author:[email protected]

Author Profile
Dexin Yang
Xinjiang University
Author Profile
Yong Dai
Lifting machinery inspection station
Author Profile
Xiaochen Jia
Xinjiang University
Author Profile

Abstract

A physically informed neural network(PINN)for predicting material fatigue life is proposed. The Basquin formula and the Walker equivalent driving force model combined with the neural network custom blocks form physical information neurons to output preliminary lifetime results consistent with the S-N curve. The fully connected neural network evaluates the influence of temperature, notch stress coefficient and other factors that are not easy to directly participate in the calculation on fatigue life. The output of the physics module and the fully connected neural network is integrated by a single-layer fully connected neural network to generate the final life prediction result. The learning process of the obtained neural network embedded in the physical model is relatively stable, and the influence of outlier samples can be properly handled. The proposed method was validated using published experimental data. The prediction results are basically within 2 times the error bar, which meets the expectations of physical constraints and has certain extrapolation capabilities