A Gaussian process-based fatigue crack prediction method with update based on the feature deviation

Accurately predicting crack growth is essential for maintenance and health management in engineering. The unpredictable nature of crack propagation in real structures requires leveraging available data, such as condition monitoring, for estimation. Gaussian Process (GP), a Bayesian machine learning method, is adept at probabilistically modelling nonlinear relationships and handling uncertainty, thus facilitating straightforward implementation. A GP-based method for predicting crack growth in actual aircraft structure is proposed in this paper, incorporating updates based on feature deviation to enhance accuracy in crack propagation evaluation. The effectiveness of this approach is validated through fatigue tests on aluminium attachment lug specimens, affirming its reliability and practical utility.