During the height of the Cold War, the United States focused much of is resources on the production of nuclear weapons. In pursuit of this, they constructed huge facilities dedicated to refining and enriching uranium for use in bombs. Now, as conflict has subsided and more modern techniques are favored, these larger-than-life facilities face decontamination and decommissioning. The process of vetting a building for demolition involves hundreds of millions of taxpayer dollars, hundreds of thousands of man hours working in harsh conditions, and is rife with opportunity for schedule setbacks and stalls. Containing hundreds of miles of large diameter process piping, each building undergoes an lengthy inspection process. A measurement is taken roughly every four inches of pipe to determine the content of U235 in any holdup deposit present in the pipe. This amounts to millions of manual measurements per building. If enough U235 is present, they are required to remove that section of piping and manually removed the deposit and dispose of it in a criticality-safe way.
Wielding the power of automation, a team lead by my adviser, Red Whittaker, and I developed a system capable of inspecting these pipes from within, revolutionizing the way the government decommissions these facilities. We developed a robot capable of measuring the volume of deposit within the pipes. This involved creating a new type of sensor, which was capable of sensing (1) the surface of the steel pipe, and (2) the surface of deposits. The metal of the pipe was sensed with an inductive proximity sensor, which ignored the presence of any uranium deposit, and the surface of the deposit was sensed with an laser triangulation sensor. Together, these sensors were mounted on a spinning disk, and as the robot traversed longitudinally down the pipe, a 3D point cloud of the two surfaces is sampled in a corkscrew pattern.
I served as co-lead of on-board robot software that implemented autonomy and robot safeguarding. This was done within the ROS framework, written mostly in C++. I also lead the software development for the various device drivers that were needed to interface with the motor drivers, cameras, data acquisition units, and various other sensors. Apart from my work in software, I also lead the characterization and calibration of the volumetric sensor. I automated the calibration procedure for the metal and deposit surface measuring sensors using a CNC machine and controlling its movements via external software.
This allowed me to run the 5+ hour calibrations necessary per sensor and give me the freedom to work on the robot software. Above is a time-lapse video of the calibration of a inductive proximity sensor. The sensor outputs a voltage that is non-linearly related to the offset distance. The procedure for calibration is to move the sensor from 0-25mm offset from the metal plate at 0.127mm increments. The sensor dwells and takes a thousands measurements to facilitate the creation of a probabilistic sensor model.
Above is a shot of to PipeDream robot, without the main sensor attached. The project culminated in a deployment inside the actual process piping with visible uranium deposit. Below is a shot of the robot sitting inside the actual process pipe, with the yellowish coating being the uranium deposit.