Machine-aided cleanroom assembly of SRF structures

The modern linear particle accelerators use superconducting radiofrequency (SRF) cavities for achieving extremely high-quality factors (Q) and higher beam stabilities. The assembly process of the system, although with a much more stringent cleanness requirement, is very similar to the ultrahigh vacuum (UHV) system operation procedure. Humans, who are conventionally the operators in this procedure, can only avoid contaminating the system by wearing proper sterile personal protection equipment to avoid direct skin contact with the systems, or dropping lint or dander. However, humans unavoidably make unintentional mistakes that can contaminate the environment - cross contamination of the coverall suits during wearing, slippage of masks or goggles, damaged gloves, and so forth. Besides, humans are limited when operating heavy weights, which may lead to incorrect procedures, or even worse, injury. Euclid Techlabs, LLC submits this Phase I SBIR proposal for the development of a viable and cost-effective machine vision and automation assisted robotics system to dominate the assembly process in a cleanroom environment, for SRF systems and beyond. Our object recognition algorithms are supported by machine learning to identify SRF structures and specify actions for the robotic arm to mate parts together. Our proposed complete assembly system has the unique combination of image processing, object identification, spatial position and orientation marking, robotic arm control and manipulation, and more. We have gathered a strong team of SRF and vacuum scientists, machine learning physicists, control and mechanical engineers to design and build the power system. In Phase I, we will be primarily focusing on developing a framework and workflow for communicating with a commercially available robotic arm to identify, locate and move components to designated places. Our deliverables include: A machine vision algorithm that identifies a vacuum system component and removes the background (image matting); A machine learning model that can be trained to identify conflat flanges with the radial grooves for helium leak checking; An HD camera group, each mounted and calibrated at a certain location, to independently calculate the position and orientation of the flanges with the above algorithm; An actuator-driven gripper for securely holding and moving a component; A robot control algorithm that communicates with the robotic arm through the robodk API and controls the gripper to grab or release the target component; A bench setup to test the system, including the robotic arm. Because of the ever-increasing demand and interest in SRF development and cleanroom assemblies, especially from high energy physics (HEP) experiments, our Phase I work has received support from many SRF and accelerator researchers. We believe that the proposed R&D on this topic will be of significant benefit to many experiments that operate with SRF or UHV systems, where stringent cleanroom operations are a necessity.