On the efficiency of machine learning for fatigue assessment of post-processed additively manufactured AlSi10Mg. (July 2022)

• Record Type: Journal Article
• Authors: Maleki, E.; Bagherifard, S.; Razavi, S.M.J.; Bandini, M.; du Plessis, A.; Berto, F.; Guagliano, M.
• Is Part Of: International journal of fatigue. Volume 160(2022)
• Publication Date: 2022-07
• Subjects: Laser powder bed fusion -- Notch -- Shot peening -- Fatigue -- Heat treatment -- Machine learning; Materials -- Fatigue -- Periodicals; Materials -- Fatigue; Periodicals; 620.1122
• DOI: 10.1016/j.ijfatigue.2022.106841
• Languages: English
• ISSNs: 0142-1123

Abstract

Highlights: ML was used to model the fatigue behavior of V-notched LPBF AlSi10Mg samples. Fatigue life was predicted and analyzed using sensitivity and parametric analyses. Enhancing the number of neurons in each layer enhanced the efficiency of the SNNs. the accuracy of the predicted results was improved by increasing the number of hidden layers in the developed NNs. Pre-trained DNNs with SAE demonstrated the best performance with an accuracy of more than 99%. Considering parameters impacted by SP, surface modification factor had the highest effects for improving fatigue life. Abstract: Laser powder bed fusion (LPBF) is receiving widespread attention for its capability to build components with complex geometries. Post-processing can address the adverse effects of various imperfections exhibited in LPBF parts in their as-built state, including inhomogeneous microstructure, tensile residual stresses and poor surface quality. In a recent experimental study, we investigated the influences of different post-processing techniques including heat treatment and shot peening as well as their combination on rotating bending fatigue behavior of V-notched LPBF AlSi10Mg samples. Herein, we further examined those samples regarding the specific parameters that directly influence fatigue performance with the aim to develop a deep learning based approach by means of artificial neural network. The effect of yield stress, ultimate tensile strength, elongation, porosity, microhardness, compressive residual stresses, and surface roughness and morphology were assessed and implemented in the model. Fatigue behavior of the samples was predicted and analyzed using sensitivity and parametric analyses. The obtained results reveal the high potential of deeply learned neural network for unlocking the role of post-processing on fatigue performance of LPBF AlSi10Mg samples.