Inspiration

Traditional pipe inspections have inconsistent data quality, limited coverage, and delayed response, often leading to small fissures and cracks going unnoticed. This could potentially result in high emergency repair costs, significant risks to both personnel and the environment, and catastrophic failure. Manual methods expose workers to hazardous confined spaces and often yield subjective, incomplete results that make it difficult to accurately predict deterioration trends. Moreover, conventional inspection tools struggle to adapt to pipes with varying diameters, curves, and obstructions, frequently missing hidden defects that can escalate into major incidents. Recognizing these challenges, PipeX was developed as an agile, AI-powered pipe crawler that automatically adjusts to diverse pipe sizes, navigates obstacles and patch issues, and continuously captures high-resolution video and sensor data. By leveraging advanced computer vision, deep learning models, and sensor fusion, PipeX delivers real-time, actionable insights that enable proactive maintenance, enhance safety by eliminating the need for dangerous manual inspections, and significantly reduce overall lifecycle costs. This project is essential for transforming pipeline integrity monitoring, ensuring our critical infrastructure remains reliable and safe for the future.

Overview

PipeX revolutionizes pipe integrity monitoring through the seamless integration of AI, sensor fusion, and a custom‑designed robotic crawler. Engineered to adjust to a wide range of pipe diameters, PipeX navigates curves and obstacles, such as debris or repair patches, with ease. As it crawls through pipes, its onboard high‑resolution camera and array of sensors continuously capture data. Deep learning models process this data in real time to identify defects, predict potential failures, and pinpoint leak locations. The result is an automated inspection system that not only accelerates the evaluation process but also provides detailed, actionable insights for proactive maintenance. PipeX is designed to serve a wide range of inspection needs across various industries. In water utilities, it continuously monitors municipal water and sewer lines, rapidly detecting early signs of wear, leakage, or blockages—even in complex and narrow pipelines. For industrial maintenance, PipeX streamlines routine inspections of large-scale pipeline networks, reducing operational downtime and ensuring safe performance across diverse pipe geometries. In remote or hazardous environments, PipeX eliminates the need for personnel to enter dangerous confined spaces, providing precise, real-time defect analysis while skillfully navigating around obstacles and patch repairs.

Development Process

Mechanical

We engineered a state‑of‑the‑art, mechanically robust pipe crawler that can adjust to a wide range of pipe diameters and navigate through complex geometries. Designed to maneuver through tight curves, overcome obstacles such as debris and patch repairs, the crawler’s adaptive structure ensures comprehensive internal coverage. Its modular design and rugged construction enable it to operate reliably in harsh, corrosive, high‑moisture environments, significantly reducing the need for risky manual inspections. Our initial concept involved an X or cross-bar frame; however, we adopted a staggered bar layout to enhance stability during traversal. To accommodate pipes of varying diameters, we developed an expandable suspension system. While we initially explored a scissor linkage mechanism, however, it proved structurally unstable. We then transitioned to a rack-and-pinion system, which delivers both robust structural support and up to 20mm of frame expansion. This system was specifically engineered to preserve the crawler’s structural integrity throughout expansion. The base of the robot includes a camera mount and a two-axis robotic arm. Our demonstration model features a scaled-down version of this arm, which can accurately pinpoint defects within the pipe. To maximize coverage and compensate for limited visibility from the stationary camera, we integrated a servo-driven rotating base, allowing the crawler to scan the entire internal surface. For traction, we upgraded from standard wheels to Gecko Wheels, which offer superior grip and ensure stable, reliable traversal through various pipe environments.

Software

On the software side, PipeX employs a lightweight computer system to power the demonstrator and leverages cutting-edge AI and computer vision frameworks to automate defect detection. The demonstrator is controlled by a Raspberry Pi 3 Model B, which handles the brunt of the data processing and logic and controls an Arduino UNO. The UNO enables control over the servos and other electrical components. The Pi communicates with the UNO over a Serial connection, using the default protocol provided by the Python libraries. The control loop is written in Python and run on the Pi. This sends instructions over Serial to the Arduino, telling the Arduino what servos to run and whether or not a target is captured by the camera vision algorithm. The camera vision is run on OpenCV with a camera connected to the Pi. First, the video input is masked to only see the pixels that lie within a certain color range. Then, this masked input is converted into a binary mask. This binary mask defines the space where all the potential targets lie. The mask is eroded and dilated to smooth out any noise. This joins regions that lie close together while minimizing the impact of very small detections. Bounding boxes for the contours are then calculated, and the position of the largest one is found. Depending on the location of this on the screen, the Pi will instruct the Arduino to rotate the camera and arm assembly to center the target. From there, measures can be taken to repair or alleviate the issue. As the Raspberry Pi has limited computational resources, a more complicated computer vision algorithm cannot be run on it without major sacrifices in performance. Separately, but to be integrated in the future, PipeX has an automated pipe defect detection algorithm. This algorithm first converts the image to grayscale, then finds key points using ORB. These key points are filtered to remove all those without enough neighbors in their vicinity. This reduces the noise of the keypoints, allowing the algorithm to focus on the most interesting parts of the picture. These key points are then merged together in a mask, which is smoothed out to ensure that smaller regions are joined together to create large areas of interest. These areas of interest are captured by contours from which bounding boxes are calculated. These bounding boxes are then given a score based on their size and number of key points contained relative to the largest bounding box. Scores below a certain threshold are thrown out, leaving only the largest and most interesting regions for analysis. The image, annotated by the bounding boxes, is then given to the 4o model with prompt reinforcement, a growing system prompt that allows for quicker learning than traditional reinforcement. 4o then returns a JSON with the relevant classifications of defects, or lack thereof, in the pipe. This information can be used to define an actionable plan for repairs on the robot.

Challenges Overcome

Developing PipeX required overcoming a range of complex engineering and software challenges. One of the key hurdles was packaging all mechanical, electrical, and sensing components into a compact and modular frame. Ensuring that the camera, two-axis robotic arm, motor drivers, and computing units could coexist in a confined space required spatial optimization and modular design thinking. To handle tight curves and bends within pipes, we engineered a crawler with an adaptive chassis and Gecko wheels, which offer enhanced traction and flexibility during traversal. Detecting pipe defects in real-world conditions requires robust computer vision and machine learning. We trained an AI model to differentiate between cracks and discolorations using prompt reinforcement and real defect data. Establishing reliable communication between the Arduino and the Raspberry Pi was a key integration challenge. We implemented a lightweight serial protocol to synchronize commands and data between the two systems efficiently. This enables PipeX to identify defects and provide real-time feedback during inspections, allowing for faster and safer decision-making in the field.

Accomplishments

Mechanical

One of our key mechanical accomplishments was designing a unique X-style crossbar frame that enhances structural stability and maintains balance during traversal. We engineered and manufactured a fully custom chassis, carefully optimized for modularity and the compact integration of all core components. To ensure adaptability across various pipe diameters, we developed an expandable suspension system that maintains structural integrity even during dynamic movement. A major integration achievement was the implementation of a software array onboard the robot, which includes a camera, Arduino, Raspberry Pi, and LEDs to signal defect detection. This setup enables seamless communication between hardware and software, allowing for real-time processing and control during inspection runs.

Software

We had not had experience working with Raspberry Pi or Arduino before, so learning how these systems worked and enabling them to communicate across platforms seamlessly was a major hurdle. One of the main challenges was running a computer vision algorithm on the Raspberry Pi while simultaneously using its output to control the Arduino in real time. Coordinating commands over Serial communication while maintaining efficient performance within the OpenCV pipeline required careful balancing and optimization. Another major hurdle was fine-tuning the automated pipe defect detection algorithm. Specifically, identifying accurate bounding boxes around defects and integrating the 4o model into the classification process presented unique challenges. Unlike conventional bounding box detection, we had to develop a custom method to convert a set of key points into reliable bounding regions. Additionally, implementing prompt reinforcement for the 4o model proved difficult due to the limited availability of high-quality training data and the model’s tendency to deviate slightly from standard output formats. Overcoming these issues was critical to enabling accurate, real-time defect identification within the pipe environment.

Future Vision

Our vision for PipeX is to evolve into a fully autonomous, data‑driven pipeline monitoring network that is highly scalable and adaptable to diverse industry needs. In the future, we aim to enhance its capabilities by integrating specialized repair tools—such as welding modules and cutting instruments—that enable PipeX not only to detect a wide array of defects but also to perform immediate on-site remediation across various pipe materials, diameters, and configurations. This adaptability will allow PipeX to serve everything from municipal water and sewer systems to industrial oil and gas pipelines, providing real‑time, actionable insights while proactively addressing issues before they escalate. By continuously expanding its tooling and repair functionalities, PipeX is poised to transform how critical infrastructure is maintained, dramatically reducing downtime, cutting costs, and ensuring safer, more sustainable operations for tomorrow.