Application of data science approach to fatigue property assessment of laser powder bed fusion stainless steel 316L

The adaptive neuro-fuzzy inference system (ANFIS) was applied for fatigue life prediction of laser powder bed fusion (L-PBF) stainless steel 316L. The model was evaluated using a dataset containing 111 fatigue data derived from 14 independent S-N curves. By using porosity fraction, tensile strength and cyclic stress as the inputs, the fuzzy rules defining the relations between these parameters and fatigue life were obtained for a Sugeno-type ANFIS model. The computationally derived fuzzy sets agree well with understanding of the fatigue failure mechanism, and the model demonstrates good prediction accuracy for both the training and test data. For parts made by the emerging L-PBF process where sufficient knowledge of the material behavior is still lacking, the ANFIS approach offers clear advantage over classical neural network, as the use of fuzzy logics allows more physically meaningful system design and result validation.