This is one of the first projects that I got involved in while at CMU. It took a while before I could really dive in, but especially with the addition of a new member on the project, we’ve picked up steam. I’m primarily responsible for hardware and got my feet wet (pun-intended) with waterproofing.
Overview
Just after I arrived CMU, my would-be advisor won a contract to develop a system that would inspect the concrete walls of water basins storing spent nuclear fuel rods at the Department of Energy’s (DOE’s) Savannah River Site. These water basins act to cool the fuel rods. The concrete walls of these basins have shown some signs of degradation, small cracks. At certain times of the year, workers have noticed some water leakage on the outside of the walls. Current inspection methods involve using an underwater camera on a stick that often auto-focuses on the wrong thing. CMU’s goal is to develop a system that will map the basin using Site Localization and Mapping (SLAM) techniques and precisely locate where cracks are and leaks occur.
The Hardware
The Pod
The sensors and electronics that we are using include a stereo camera pair, an IMU, a laser pattern projector, an LED, a water sensor (for leaks), and a pressure sensor. This data is all fused together to create a model of the environment. Before we could get this data, however, I needed to build a waterproof sensor pod. It took a bit of tweaking because of leaking, but the CAD of the design is shown below.
The back plate is sealed with a large O-ring. The front has a face seal, which is a made of a cut-out of rubber. The two red bolts have built in o-rings, and the barbed fitting uses a hose clamp over the tube that fits over it. We’ve tested this housing to be waterproof up to 10 feet.
The Electronics
In addition to making the sensor pod, I also worked with electronics. I contributed to the wiring diagram, breadboard testing, and created the protoboard myself. This is the most “permanent” electronic assembly that I’ve personally put together (aka no breadboards), and I’m currently working on a PCB using Altium.
Testing
This was the fun part. The first tests we ran were in air. Another student, Larry Papincak, made an x-y gantry in a small tank to be used for early testing. This pod was mounted on the gantry, and was placed to look at scenes created in the tank.
Once we confirmed that this pod was waterproof (we left it inside a 10 foot deep water tank with no electronics), we assembled the pod with electronics and tested in water.
The laser pattern projector created a cool sci-fi-looking beam. This whole process reminded me of this scene from The Incredibles where Buddy scans a cave to look for Mr. Incredible (skip to 3:08).
Early data was compiled using VisualSFM. Textureless objects aren’t easy to recreate, this is why the pattern projector was added to the sensor pod. The reconstruction below, however, only used stereo camera data without the use of the pattern projector.
Summary
There’s much left to do on this project. As mentioned before, I’m working on creating a PCB that will consolidate the current protoboard, DC/DC converters, the laser pattern projector PCB, and the IMU breakout board. After that, I will be designing a smaller version of the sensor pod (less empty space means less buoyancy), and design it for easier assembly disassembly. (The tolerance on the current 8″ tube is +- 0.050″, which means the o-ring needed to be very large, which means taking the tube on and off is difficult. Having a precisely machined aluminum tube, which we could also heat sink to, allows us to use smaller o-rings, which should make it easier to assemble and disassemble.)
Apart from the hardware, there’s plenty of software work to be done. We haven’t gotten through working with the data we’ve collected, and we have yet to fully integrate the IMU.
A paper that will appear in the Symposium on Waste Management: “Localized Imaging and Mapping for Underwater Fuel Storage Basins” that details this work.
P.S.
For all those machinists out there, check out the really nice finish I got when turning the back plate of the sensor pod, just from the lathe. It’s nice to see what a new bit and the automatic feed can do.
Andrew Tallaksen 