Spatial Frequency Analysis for Improved Quality in Big Area Additive Manufacturing (BAAM)

Abstract

Process control in 3D printing (also known formally as Additive Manufacturing - AM) has largely been absent even in production systems. Simultaneously, computer vision has become more accessible with open source libraries (e.g. OpenCV, used successfully for traversing the state of California in an autonomous vehicle to win a DARPA Grand Challenge). 3D printing is particularly well suited to be enhanced by computer vision as fabrication is layer wise and predictable assuming correct operation. Big Area Additive Manufacturing (BAAM) — operating at significantly larger scales than traditional 3D printing — stands to benefit given the higher throughput of material (hundreds of pounds per hour) and the associated high costs of errant fabrication. Furthermore, minimum feature sizes in BAAM, such as individual layers, are sufficiently large to be analyzed with standard photography. With computer vision, sophisticated algorithms can be applied to identify problems early in the process that are not normally manifest until after process completion. Subtle and latent defects can be remediated before the onset of permanent damage or at a minimum the process can be aborted to avoid significant material loss. Fourier analysis can provide a useful perspective of the spatial periodicity of the layers of exposed surfaces during fabrication and this spectral information can inform the process of surface roughness, delamination, and deposition consistency in a data efficient manner. The large layer thickness of BAAM allow for Fourier analysis to be performed with standard photography. This paper explores the implementation and advantages of a low cost computer vision system that leverages OpenCV libraries operating on a Raspberry Pi Linux computer with simple yet high resolution photography — driven by the hypothesis that quality and yield of open source BAAM hardware can be dramatically enhanced.