• Online only
  • New
First page LCF10-2026-047
search
  • First page LCF10-2026-047

Neural-Network-Based Low Cycle Fatigue Life Estimation of Aluminum Alloys - Effect of Input Variables

    Aluminum alloy components with high specific strength are widely used in automotive engines to improve fuel efficiency by reducing weight. These components are subjected to severe service conditions involving simultaneous cyclic mechanical and thermal loading. Therefore, reliable evaluation methods of fatigue resistance need to be developed. In particular, because fatigue life decreases due to the creep deformation over one-half of the melting temperature, a practical method that can estimate fatigue life under complex loading conditions is required.

    Low cycle fatigue (LCF) life prediction of metallic materials has traditionally been based on strain-based relationships, such as the Coffin-Manson law, as well as physics-based models. Some of the…

€25.00
VAT included
Quantity

 

Datenschutzbedingungen (bearbeiten im Modul "Kundenvorteile")

 

Lieferbedingungen (bearbeiten im Modul "Kundenvorteile")

 

Rücksendebedingungen (bearbeiten im Modul "Kundenvorteile")

    Aluminum alloy components with high specific strength are widely used in automotive engines to improve fuel efficiency by reducing weight. These components are subjected to severe service conditions involving simultaneous cyclic mechanical and thermal loading. Therefore, reliable evaluation methods of fatigue resistance need to be developed. In particular, because fatigue life decreases due to the creep deformation over one-half of the melting temperature, a practical method that can estimate fatigue life under complex loading conditions is required.

    Low cycle fatigue (LCF) life prediction of metallic materials has traditionally been based on strain-based relationships, such as the Coffin-Manson law, as well as physics-based models. Some of the authors have proposed a fatigue damage law that employs the inelastic strain obtained from LCF tests, separated into plastic and creep components. The applicability of the damage law was validated through simulations using a constitutive model of cyclic viscoplasticity [1]. However, in these empirical and physics-based approaches, a large number of material parameters must be identified to describe damage evolution under different temperatures and loading conditions. This requirement may limit their applicability to generalized fatigue life prediction under complex service conditions.

    With recent advances in computational capabilities, machine learning approaches, particularly neural networks, have attracted attention for fatigue life estimation due to their high flexibility in nonlinear regression problems. Based on experimental data, neural-network-based models can directly capture complex interactions among multiple factors without explicitly defining constitutive equations. Most of the previous studies, however, have mainly focused on improving prediction accuracy or optimizing network architecture. In contrast, the effect of input variables on prediction performance and the physical interpretation of the internal weights of trained models have not been sufficiently investigated. The selection of input variables is therefore a crucial problem for enhancing both the reliability and physical consistency of machine-learning-based fatigue models.

    In this study, a method for fatigue life estimation based on a neural network is developed using LCF test results of aluminum alloys. LCF tests are conducted under different loading conditions and test temperatures. Input variables are selected from temperature, strain amplitude, loading time as test conditions, and elastic strain amplitude, inelastic strain amplitude, and inelastic strain energy density as experimental results. The prediction results obtained from the trained model using these input variables in different combinations are validated to quantify the effect of the input variable on the estimations. Furthermore, the contributions of individual input variables to fatigue life estimation are clarified. The clarification is conducted by analyzing the connection strengths between the input layer and the hidden and output layers of the trained network.

Reference
LCF10-2026-047

Title
Neural-Network-Based Low Cycle Fatigue Life Estimation of Aluminum Alloys - Effect of Input Variables
Author(s)
T. Hayashibe, K. Sasaki, K. Ohguchi, K. Fukuchi, S. Honda, Y. Tsubota, W. Nagai, K. Ohsato, N. Shinya
DOI
10.48447/LCF10-2026-047
Event
Tenth International Conference on Low Cycle Fatigue (LCF-10)
Year of publication
2026
Publication type
conference paper (PDF)
Language
English
Keywords
Aluminum alloy,Low-cycle fatigue,Fatigue life estimation,Neural network