Deep convolutional generative adversarial network for generation of computed tomography images of discontinuously carbon fiber reinforced polymer microstructures.

Computed tomography images are of utmost importance when characterizing the heterogeneous and complex microstructure of discontinuously fiber reinforced polymers. However, the devices are expensive and the scans are time- and energy-intensive. Through recent advances in generative adversarial networks, the instantaneous generation of endless numbers of images that are representative of the input images and hold physical significance becomes possible. Hence, this work presents a deep convolutional generative adversarial network trained on approximately 30,000 input images from carbon fiber reinforced polyamide 6 computed tomography scans. The challenge lies in the low contrast between the two constituents caused by the close proximity of the density of polyamide 6 and carbon fibers as well as the small fiber diameter compared to the necessary resolution of the images. In addition, the stochastic, heterogeneous microstructure does not follow any logical or predictable rules exacerbating their generation. The quality of the images generated by the trained network of 256 pixel × 256 pixel was investigated through the Fréchet inception distance and nearest neighbor considerations based on Euclidean distance and structural similarity index measure. Additional visual qualitative assessment ensured the realistic depiction of the complex mixed single fiber and fiber bundle structure alongside flow-related physically feasible positioning of the fibers in the polymer. The authors foresee additionally huge potential in creating three-dimensional representative volume elements typically used in composites homogenization.

The Mechanical Properties of 3D-Printed Polylactic Acid/Polyethylene Terephthalate Glycol Multi-Material Structures Manufactured by Material Extrusion.

The mechanical properties of polylactic acid (PLA), polyethylene terephthalate glycol (PETG), and PLA/PETG structures manufactured using the multi-material additive manufacturing (MMAM) method were studied in this work. Material extrusion additive manufacturing was used to print PLA/PETG samples with various PLA and PETG layer numbers. By varying the top and bottom layer numbers of two thermoplastics, the effect of layer number on the mechanical properties of 3D-printed structures was investigated. The chemical and thermal characteristics of PLA and PETG were investigated using Fourier transform infrared spectroscopy and differential scanning calorimetry. Tensile and flexural strength of 3D-printed PLA, PETG, and PLA/PETG samples were determined using tensile and three-point bending tests. The fracture surfaces of the samples were evaluated using optical microscopy. The results indicated that multi-material part containing 13 layers of PLA and 3 layers of PETG exhibited the highest ultimate tensile strength (65.4 MPa) and a good flexural strength (91.4 MPa). MMAM was discovered to be a viable way for producing PLA/PETG materials with great mechanical performance.

Experimental investigation and development of a deep learning framework to predict process-induced surface roughness in additively manufactured aluminum alloys.

A deep learning framework is developed to predict the process-induced surface roughness of AlSi10Mg aluminum alloy fabricated using laser powder bed fusion (LPBF). The framework involves the fabrication of round bar AlSi10Mg specimens, surface topography measurement using 3D laser scanning profilometry, extraction, coupling, and streamlining of roughness and LPBF processing data, feature engineering to select the relevant feature set and the development, validation, and evaluation of a deep neural network model. A mix of core and contour-border scanning strategies are employed to fabricate four sets of specimens with different surface roughness conditions. The effects of different scanning strategies, linear energy density (LED), and specimen location on the build plate on the resulting surface roughness are discussed. The inputs to the deep neural network model are the AM process parameters (i.e., laser power, scanning speed, layer thickness, specimen location on the build plate, and the x,y grid location for surface topography measurements), and the output is the surface profile height measurements. The proposed deep learning framework successfully predicts the surface topography and related surface roughness parameters for all printed specimens. The predicted surface roughness ( S a ) measurements are well within 5% of experimental error for the majority of the cases. Moreover, the intensity and location of the surface peaks and valleys as well as their shapes are well predicted, as demonstrated by comparing roughness line scan results with corresponding experimental data. The successful implementation of the current framework encourages further applications of such machine learning-based methods toward AM material development and process optimization.