| Abstract |
Pipes are essential to sustain contemporary infrastructure like transporting fluids and allowing for organizing electrical wires. That said, they are subject to wear after prolonged usage, necessitating the need for conduit inspection methods. Current widespread conduit inspection methods include non-invasive ground-penetrating radar and ultrasonic technology, and invasive inspection robots, pigs, and acoustic sensors. Both have limitations, however, invasive techniques have proven to detect interior defects compared to their non-invasive counterparts more accurately. There are currently many adaptations of the pipe-inspection robot, however, they still need to be improved. AI and simplistic trajectory algorithms have yet to be widely implemented or tested on inspection robots. Pipe inspection modeling has most frequently been done with LiDAR (Light Detection and Ranging), however, reflective environments render it obsolete, which is why it should be paired with an additional type of ranging like ultrasonic or infrared ranging.
A well-established implementation of the pipe-inspection robot is the MRINSPECT series, which is a wall-press type robot. This series of robots deals with turning using a differential drive system. The robot has not yet been tested to detect defects using neural networks nor has it been tested to produce three-dimensional models of pipes. Automation using SLAM and LiDAR has been extensively reviewed and tested, however, it may be improved upon by using simpler algorithms for real-time inspection. Due to the size limitations of a pipe, it may be more effective to use a 1-dimensional ranging sensor to measure the inner pipe walls radially instead of using a bigger 2-dimensional LiDAR to measure the relative lateral distance of the pipe. SLAM may also have difficulty with dealing with low-light conditions and repetitive patterns, both characteristics of pipe environments. Dead reckoning, reliant on wheel odometry and accelerometer data, would be a more feasible method given the size constraints of a conduit. A combination of dead reckoning and LiDAR-ultrasonic ranging would provide the same data as 3-dimensional SLAM within a pipe interior.
Integrating multiple technologies with the pipe inspection robot may provide more accurate and reliable data on pipe interiors, which can aid in identifying potentially hazardous deformities. The purpose of my project is to develop a robotic system that fuses ultrasonic and LiDAR data and generates 3D point clouds of conduit interiors for inspection. My project also focuses on integrating a corrosion detection neural network with this system.
|
