Artificial Intelligence aided Design and Manufacturing

Leader of the group:  Dr. Vahid Babaei

Group's website

Open Positions - PhD candidates

We are looking for motivated researchers to start as PhD candidates in the Artificial Intelligence aided Design and Manufacturing Group. The candidates will research the exiting and rising field of computational manufacturing. The group has a particular focus on appearance manufacturing using a wide range of advanced manufacturing devices, among which 3D printers. Appearance manufacturing concerns the fabrication of objects with a given appearance and has a variety of applications from medical devices and cultural heritage preservation to stop-motion animation. We have a keen interest in using deep learning methods in the context of manufacturing, where we have taken successful initial steps.

Strong applicants with a master degree (or those approaching the end of their master studies) in computer science or a related field are encouraged to apply. Research background in one or more of the following areas is necessary: computer graphics (with a preference for rendering), geometry processing, image processing or machine learning. Solid skills in mathematics and related programming languages (C++ / Python / OpenGL) are also required.

Please send your application or any inquiry to Dr. Vahid Babaei ( The application should include a CV, motivation letter and copy of transcripts. Names of two references agreed to write a recommendation letter are also required.

Vision and Research Strategy

We focus on inventing new computational tools that release the full potential of digital fabrication hardware, such as additive manufacturing (also known as 3D printing). With the immense growth of the manufacturing hardware in resolution, scale and speed, the algorithm complexity increases even more dramatically. Our group therefore aims at developing hardware-aware, scalable algorithms for advanced manufacturing.
We have a particular interest in visual appearance of objects and strive for better algorithms that help manufacturing products with novel and useful appearance characteristics. Design for manufacturing of objects with high-fidelity appearance specifications is a key engineering task. Thus, there remains a huge opportunity for revolutionizing the computational appearance design and manufacturing, and significantly improving the appearance quality of many products. The results of this research will immediately enable numerous applications in rapid prototyping and manufacturing of end-use products. This spans several application domains from medical devices and surgical training, to cultural heritage preservation and anti-counterfeiting.

Research Areas and Achievements

Appearance 3D Printing. High-fidelity appearance reproduction using 3D printers is a main area of our research. Recently, we introduced a color reproduction framework for 3D printed objects with challenging geometric features while compensating for their, undesired, subsurface scattering. The engine of this framework is a light transport simulator relying on the Monte Carlo volume rendering. In a later work, we replaced this component by a hierarchical deep neural network, obtaining more than 2 orders of magnitude speed up.
Gloss is arguably the most prominent visual attribute after the color. Although there has been considerable progress in color reproduction, 3D fabrication of objects with controlled gloss has not received a lot of attention. The main reason is that the available fabrication hardware have limited capability for modifying the spatially-varying gloss of objects. In another work, we addressed these limitations and proposed a new printing hardware based on piezo-actuated needle valves capable of jetting highly viscous varnishes. Based on the new hardware setup, we presented a complete pipeline for controlling the gloss of a given 2.5 D object, from printer calibration to manufacturing of models with spatially-varying gloss.
One of the main challenges of appearance fabrication is rooted in materials. Real-life objects are made of a variety of materials, most of which cannot be used by digital fabrication. To this end, we proposed to adapt the limited channels of digital printers to the appearance of the input. Particularly, given a watercolor painting, we ask what are the best ink set, from within a large ink library, that reproduce this painting optimally? We introduced a novel algorithm for ink selection problem using mixed integer programming. Our physically inspired problem formulation results in a mixed-integer linear programming whose continuous relaxation is convex and scales gracefully with larger problems. The results of this part have been published in: SIGGRAPH (1), SIGGRAPH ASIA (2), and Eurographics (1).

Novel Computational Fabrication Products. Most of the time, digital hardware offer capabilities which are never utilized, because the necessary software and algorithms are underdeveloped. Recently, our group has invented new computational tools capable of releasing the full potential of fabrication technologies, such as additive manufacturing (also known as 3D printing) or laser-material processing. For example, we developed algorithms for design and fabrica-tionofthemoiréeffect.Byprintingasetoflenticularlenseson3Dsurfaces,theobtainedmoiré that can function as a visual security tag. In a second work, we designed and 3D printed highresolution volumetric light-field displays thanks to a neural parameterization of the design.
In addition to 3D printing, we are paying attention to new digital manufacturing platforms, such as laser marking, a rapidly growing technology with many applications in object identification, customization, and authentication. While mostly a monochromatic method, some important metals exhibit a range of colors when treated with laser, as a result of complex physicochemical phenomena. Unfortunately, the analytical relationship between the device’s design space (laser parameters) and performance space (e.g., marked colors) is unknown. Assuming a blackbox model, in a recent work, we designed a measurementbased, data-driven performance space exploration. We explored different performance criteria including the color gamut and marking resolution by consecutive marking and measuring. We uncover the process’s Pareto front by formulating a multi-objective optimization and solving it using a tailored evolutionary algorithm. Thanks to this work we enabled, for the first time, high-resolution, full-color images using laser marking. The results of this part have been published in: SIGGRAPH (1), SIGGRAPH ASIA (1), and Optics Express (1).