Composite lay-up (a core step in the process of making a composite part) is traditionally a labor-intensive process. The process requires skilled technicians to create the parts needed using specialized tools and equipment. This is often slow and expensive, which limits the quantity of parts that composite manufacturers can make. The Composites & Automated Solutions group at Fives has developed a technology that allows their customers to create composite parts using a robotic fiber placement head. This technology provides a lower-cost entry point into the composite lay-up process, making it easier for manufacturers to create the parts they need.

Over the past 30 years, the use of glass and carbon-fiber reinforced composites in aerospace and other high-performance applications has soared along with the broad industrial adoption of composite materials. Key to the strength and versatility of these hybrid, layered materials in high-performance applications is the orientation of fibers in each layer. Recent innovations in additive manufacturing (3-D printing) have made it possible to finetune this factor, thanks to the ability to include within the CAD file discrete printer-head orientation instructions for each layer of the component being printed, thereby optimizing strength, flexibility, and durability for specific uses of the part. These 3-D-printing toolpaths (a series of coordinated locations a tool will follow) in CAD file instructions are therefore a valuable trade secret for the manufacturers. However, a team of researchers from NYU Tandon School of Engineering led by Nikhil Gupta, a professor in the Department of Mechanical and Aerospace Engineering showed that these toolpaths are also easy to reproduce--and therefore steal--with machine learning (ML) tools applied to the microstructures of the part obtained by a CT scan.

Exemplar-based models have achieved great success on localizing the parts of semi-rigid objects. However, their efficacy on highly articulated objects such as humans is yet to be explored. Inspired by hierarchical object representation and recent application of Deep Convolutional Neural Networks (DCNNs) on human pose estimation, we propose a novel formulation that incorporates both hierarchical exemplar-based models and DCNNs in the spatial terms. Specifically, we obtain more expressive spatial models by assuming independence between exemplars at different levels in the hierarchy; we also obtain stronger spatial constraints by inferring the spatial relations between parts at the same level. As our method strikes a good balance between expressiveness and strength of spatial models, it is both effective and generalizable, achieving state-of-the-art results on different benchmarks: Leeds Sports Dataset and CUB-200-2011.