Defect‐induced fatigue scattering and assessment of additively manufactured 300M-AerMet100 steel: An investigation based on experiments and machine learning

•The physics-informed machine learning strategy is proposed for fatigue assessment.•Experimental investigations on the 300M-AerMet100 steel specimens are carried out.•Fractographic analysis shows that the internal defects affect fatigue behaviour.•The CDM based fatigue models and ANN model are presented for numerical prediction. Additive manufacturing (AM) has attracted much attention recently for its immanent advantages. Assessment of the fatigue performance for AM treated materials becomes vital for both material science and engineering applications. In this study, we extensively investigate the fatigue performance of AM processed 300M-AerMet100 steel by combining experiments, numerical simulations and machine learning. We conduct experiments to obtain fatigue curves as calibration and to determine the parameters used in the theoretical models. Continuum damage mechanics-based fatigue models are presented and numerically implemented to generate sufficient training data for machine learning. We then employ a multi-layer perceptron neural network model to predict the fatigue life of the AM processed 300M-AerMet100 steel. Experimental results show that there are scatters in the fatigue data, which may be caused by the small cracks induced by the laser cladding process via fractographic analyses. Numerical results show that a good prediction of fatigue life can be achieved by combining the continuum damage mechanics-based fatigue models and the multi-layer perceptron neural network model. This work provides a systematic prediction platform for the fatigue performance of the AM fabricated 300M-AerMet100 steel.