Strut Diameter Uncertainty Prediction by Deep Neural Network for Additively Manufactured Lattice Structures

Abstract

Abstract Additive manufacturing (AM) introduces geometric uncertainties on the fabricated strut members of lattice structures. These uncertainties result in deviations between the modeled and fabricated geometries of struts. The use of deep neural networks (DNNs) to accurately predict the statistical parameters of the effective strut diameters to account for the AM-introduced geometric uncertainties with a small training dataset for constant process parameters is studied in this research. For the training data, struts with certain angle and diameter values are fabricated by the material extrusion process. The geometric uncertainties are quantified using the random field theory based on the spatial strut radius measurements obtained from the microscope images of the fabricated struts. The uncertainties are propagated to the effective diameters of the struts using a stochastic upscaling technique. The relationship between the modeled strut diameter and the characterized statistical parameters of the effective diameters are used as the training data to establish a DNN model. The validation results show that the DNN model can predict the statistical parameters of the effective diameters of the struts modeled with angles and diameters different from the ones used in the training data with good accuracy even if the training data set is small. Developing such a DNN model with small data will allow designers to use the fabricated results in the design optimization processes without requiring additional experimentations.