On the need to adopt strain-based probabilistic approach in predicting fatigue life

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

This paper presents a comparison of fatigue life prediction capabilities among three deterministic and one probabilistic strain–life models. Most existing life prediction models are deterministic even though the fatigue process is stochastic in nature. Therefore, variability in experimental fatigue life data is not statistically justified when the deterministic approach is adopted. In this study, the Coffin–Manson, Smith–Watson–Topper, Morrow, and Weibull regression models through the use of Miner’s rule for damage accumulation process are each compared with the experimental data. Both available in the literature, two sets of experimental data for variable amplitude loading and random loading spectrum were used to explore the prediction accuracies of the models. Simulations were based on models to obtain additional results for different levels of maximum strain. They were conducted to investigate the maximum strain–life behaviour under random loading. Fatigue life comparisons were then performed, and the scatter band for type of load, level of strain amplitude, and each model separately were used for evaluations. The results show that the difference percentage in comparison with experimental data varies, but all the predictions are within the acceptable limit, that is, 50% when the factor of 2 was used. Weibull regression model has the lowest normalized-root-mean square error value (0.07) and thus has highest prediction capability among the models. Therefore, the probabilistic model is a better alternative for strain–life prediction compared with deterministic models.

Original languageEnglish
Article number82
JournalJournal of the Brazilian Society of Mechanical Sciences and Engineering
Volume40
Issue number2
DOIs
Publication statusPublished - 1 Feb 2018

Fingerprint

Fatigue of materials
Miners
Random processes
Mean square error
Loads (forces)
Statistical Models

Keywords

  • Deterministic method
  • Fatigue life
  • Probabilistic method
  • Random
  • Strain–life

ASJC Scopus subject areas

  • Mechanical Engineering

Cite this

@article{2c34adc203f947ecbb03bf5071f1f25a,
title = "On the need to adopt strain-based probabilistic approach in predicting fatigue life",
abstract = "This paper presents a comparison of fatigue life prediction capabilities among three deterministic and one probabilistic strain–life models. Most existing life prediction models are deterministic even though the fatigue process is stochastic in nature. Therefore, variability in experimental fatigue life data is not statistically justified when the deterministic approach is adopted. In this study, the Coffin–Manson, Smith–Watson–Topper, Morrow, and Weibull regression models through the use of Miner’s rule for damage accumulation process are each compared with the experimental data. Both available in the literature, two sets of experimental data for variable amplitude loading and random loading spectrum were used to explore the prediction accuracies of the models. Simulations were based on models to obtain additional results for different levels of maximum strain. They were conducted to investigate the maximum strain–life behaviour under random loading. Fatigue life comparisons were then performed, and the scatter band for type of load, level of strain amplitude, and each model separately were used for evaluations. The results show that the difference percentage in comparison with experimental data varies, but all the predictions are within the acceptable limit, that is, 50{\%} when the factor of 2 was used. Weibull regression model has the lowest normalized-root-mean square error value (0.07) and thus has highest prediction capability among the models. Therefore, the probabilistic model is a better alternative for strain–life prediction compared with deterministic models.",
keywords = "Deterministic method, Fatigue life, Probabilistic method, Random, Strain–life",
author = "M. Mahmud and Shahrum Abdullah and {Mohd Ihsan}, {Ahmad Kamal Ariffin} and {Mohd Nopiah}, Zulkifli",
year = "2018",
month = "2",
day = "1",
doi = "10.1007/s40430-018-1000-4",
language = "English",
volume = "40",
journal = "Journal of the Brazilian Society of Mechanical Sciences and Engineering",
issn = "1678-5878",
publisher = "Brazilian Society of Mechanical Sciences and Engineering",
number = "2",

}

TY - JOUR

T1 - On the need to adopt strain-based probabilistic approach in predicting fatigue life

AU - Mahmud, M.

AU - Abdullah, Shahrum

AU - Mohd Ihsan, Ahmad Kamal Ariffin

AU - Mohd Nopiah, Zulkifli

PY - 2018/2/1

Y1 - 2018/2/1

N2 - This paper presents a comparison of fatigue life prediction capabilities among three deterministic and one probabilistic strain–life models. Most existing life prediction models are deterministic even though the fatigue process is stochastic in nature. Therefore, variability in experimental fatigue life data is not statistically justified when the deterministic approach is adopted. In this study, the Coffin–Manson, Smith–Watson–Topper, Morrow, and Weibull regression models through the use of Miner’s rule for damage accumulation process are each compared with the experimental data. Both available in the literature, two sets of experimental data for variable amplitude loading and random loading spectrum were used to explore the prediction accuracies of the models. Simulations were based on models to obtain additional results for different levels of maximum strain. They were conducted to investigate the maximum strain–life behaviour under random loading. Fatigue life comparisons were then performed, and the scatter band for type of load, level of strain amplitude, and each model separately were used for evaluations. The results show that the difference percentage in comparison with experimental data varies, but all the predictions are within the acceptable limit, that is, 50% when the factor of 2 was used. Weibull regression model has the lowest normalized-root-mean square error value (0.07) and thus has highest prediction capability among the models. Therefore, the probabilistic model is a better alternative for strain–life prediction compared with deterministic models.

AB - This paper presents a comparison of fatigue life prediction capabilities among three deterministic and one probabilistic strain–life models. Most existing life prediction models are deterministic even though the fatigue process is stochastic in nature. Therefore, variability in experimental fatigue life data is not statistically justified when the deterministic approach is adopted. In this study, the Coffin–Manson, Smith–Watson–Topper, Morrow, and Weibull regression models through the use of Miner’s rule for damage accumulation process are each compared with the experimental data. Both available in the literature, two sets of experimental data for variable amplitude loading and random loading spectrum were used to explore the prediction accuracies of the models. Simulations were based on models to obtain additional results for different levels of maximum strain. They were conducted to investigate the maximum strain–life behaviour under random loading. Fatigue life comparisons were then performed, and the scatter band for type of load, level of strain amplitude, and each model separately were used for evaluations. The results show that the difference percentage in comparison with experimental data varies, but all the predictions are within the acceptable limit, that is, 50% when the factor of 2 was used. Weibull regression model has the lowest normalized-root-mean square error value (0.07) and thus has highest prediction capability among the models. Therefore, the probabilistic model is a better alternative for strain–life prediction compared with deterministic models.

KW - Deterministic method

KW - Fatigue life

KW - Probabilistic method

KW - Random

KW - Strain–life

UR - http://www.scopus.com/inward/record.url?scp=85041015919&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85041015919&partnerID=8YFLogxK

U2 - 10.1007/s40430-018-1000-4

DO - 10.1007/s40430-018-1000-4

M3 - Article

AN - SCOPUS:85041015919

VL - 40

JO - Journal of the Brazilian Society of Mechanical Sciences and Engineering

JF - Journal of the Brazilian Society of Mechanical Sciences and Engineering

SN - 1678-5878

IS - 2

M1 - 82

ER -