Build surface study of single-layer raster scanning in selective laser melting: Surface roughness prediction using deep learning. (September 2022)

• Record Type: Journal Article
• Authors: Fotovvati, Behzad; Chou, Kevin
• Is Part Of: Manufacturing letters. Volume 33(2022)Supplement
• Page Start: 701
• Page End: 711
• Publication Date: 2022-09
• Subjects: Selective laser melting -- Surface roughness -- Machine learning -- Artificial neural network -- Titanium alloy; Manufacturing industries -- Periodicals; Production engineering -- Periodicals; Manufacturing industries; Periodicals; 670
• DOI: 10.1016/j.mfglet.2022.07.088
• Languages: English
• ISSNs: 2213-8463

Abstract

Abstract: Selective laser melting (SLM) is a widely used powder-bed fusion additive manufacturing (AM) process for the fabrication of parts from metal powders in a variety of industries such as aerospace, medical, automotive, etc. Despite significant improvements in the design flexibility and mechanical performance, the poor predictability in surface finish, and yet oftentimes with large variability, remains a major challenge in the SLM use. Numerous factors affect the surface roughness of SLM-manufactured parts, which have been reported in the literature, but mostly for bulk samples composed of several layers. In this work, single-layer raster scanning of Ti6Al4V samples are designed and fabricated. The influence of the four most dominant SLM process parameters, i.e., laser power, scanning speed, hatch spacing, and layer thickness on sample surface roughness is thoroughly investigated using a fractional factorial design. Surface roughness data, acquired by white-light interferometry, from 216 data sets are then used to train a machine learning model with the back-propagation method and predict the surface roughness based on the input process parameters. The results show that the laser power is the most significant parameter in determining the top surface roughness of samples. Interestingly, although the investigated samples are single layer raster scanning areas on a solid SLM-built sample with the same parameter set, the layer thickness has a contribution of 10% to 15% in the variations of the surface roughness of the single layers. Furthermore, the machine learning algorithm achieves reasonable predictability, showing a coefficient of determination of 98.8% for a separate 32 testing data set.