Application of tabular data synthesis using generative adversarial networks on machine learning-based multiaxial fatigue life prediction

Machine learning has gradually developed into a new and effective scheme for fatigue life prediction. The novelty of this work is the proposal and verification of using virtual synthetic multiaxial fatigue data as input of machine learning models. First, the data generated by tabular generative adversarial networks are applied to machine learning models for life prediction. Then based on equivalent stress (strain) amplitude-life relationship curve, a multiaxial fatigue data generation evaluation metric is proposed. Finally, the effect of the generated sample size on the predictions of machine learning models is investigated. The method is demonstrated on 5 multiaxial fatigue data sets. The results indicate the synthetic data help machine learning models arrive at good life prediction ability. Using this method will help expand the application of machine learning-based multiaxial fatigue life prediction. • A multiaxial fatigue data augmentation method is proposed through tabular GANs. • Tabular GANs synthetic data are applied to machine learning for life prediction. • A synthetic data evaluation metric for multiaxial fatigue data is proposed. • The method extends the application of ML in multiaxial fatigue life prediction.