Autonomous Aerial Robots for Comprehensive Inspection of Maritime Vessel Tanks: Field Experiences and Public Dataset Release

מושגי ליבה

This paper presents the field deployment and lessons learned from using an autonomous aerial robot system for comprehensive inspection of maritime vessel tanks, including ballast tanks and cargo holds. The system leverages advanced autonomy modules for robust localization, mapping, exploration, and visual inspection, and has been evaluated across multiple vessels.

תקציר

The paper presents the field deployment and lessons learned from using an autonomous aerial robot system for comprehensive inspection of maritime vessel tanks, including ballast tanks and cargo holds. The system utilizes the RMF-Owl collision-tolerant aerial robot equipped with advanced autonomy modules for robust localization, mapping, exploration, and visual inspection.
The field deployments were conducted across 3 different vessels (FPSO1, FPSO2, and an Oil Tanker) and covered a diverse set of tank environments, including side tank sections, double bottom sections, and bilge sections. The authors conducted a total of 15 missions using the RMF-Owl robot, including autonomous exploration and inspection missions as well as manual multi-level flights. Additionally, data was collected using a handheld sensor setup called Mjolnir.
The key lessons learned from these field deployments include:

Resilient Autonomy: The need for a collision-tolerant robot design, robust and high-precision mapping, and scalable autonomy modules that can handle diverse tank environments and mission objectives.
Semantic Reasoning: The importance of detecting and representing inspection-critical structures and objects to improve the efficiency and effectiveness of the inspection process.
Image Quality for Defect Detection: The challenges of maintaining good image quality for reliable detection of defects like corrosion and cracks, and the need for active lighting adjustment behaviors.

The dataset collected from these field deployments, including data from the RMF-Owl robot and the Mjolnir handheld setup, has been made publicly available to support further research and development in this domain.

סטטיסטיקה

The ship ballast tanks and cargo holds present dark, dusty environments with narrow openings and wide open spaces, posing several challenges for autonomous navigation and inspection.
The RMF-Owl aerial robot weighs 1.45 kg and has a flight time of 10 minutes.
The authors conducted a total of 15 missions across 3 different vessels, including 12 missions with the RMF-Owl robot and 4 missions with the Mjolnir handheld sensor setup.

ציטוטים

"Vessel tanks including ballast tanks and cargo holds present dark, dusty environments having simultaneously very narrow openings and wide open spaces that create several challenges for autonomous navigation and inspection operations."
"Motivated by the above, our prior work [12] presents a method for fully autonomous exploration and general visual inspection of multiple compartments of bal-last tanks that is demonstrated in three field experiments."
"Precise localization of defects across multiple inspection missions is critical for tracking the health of the tanks. Hence, robust and accurate localization and mapping is key for tank inspections."

תובנות מפתח מזוקקות מ:

Maritime Vessel Tank Inspection using Aerial Robots: Experience from the field and dataset release

by Mihir Dharma... ב- arxiv.org 05-01-2024

https://arxiv.org/pdf/2404.19045.pdf

שאלות מעמיקות

How can the autonomy system be further enhanced to handle more complex and dynamic tank environments, such as those with moving obstacles or changing lighting conditions?

To enhance the autonomy system for handling dynamic tank environments, several improvements can be implemented:

Dynamic Obstacle Detection: Incorporating real-time obstacle detection algorithms using sensors like LiDAR and cameras can help the aerial robot navigate around moving obstacles within the tank environment. Machine learning techniques can be utilized to predict obstacle movements and adjust the robot's path accordingly.

Adaptive Path Planning: Implementing adaptive path planning algorithms that can dynamically adjust the robot's trajectory based on changing environmental conditions or obstacles can improve navigation in complex tank environments. This can involve real-time re-planning based on sensor feedback.

Lighting Adjustment Mechanisms: Developing mechanisms for adjusting onboard lighting based on the changing lighting conditions inside the tank can improve image quality for defect detection. This can involve integrating light sensors and algorithms that optimize lighting angles for better visibility of defects.

Multi-Modal Sensor Fusion: Enhancing the autonomy system with multi-modal sensor fusion techniques can provide a more comprehensive understanding of the environment. Combining data from LiDAR, cameras, IMUs, and other sensors in real-time can improve situational awareness and decision-making in dynamic environments.

What are the potential limitations or drawbacks of using aerial robots for tank inspection compared to other robotic platforms, and how can these be addressed?

Aerial robots for tank inspection have certain limitations compared to other robotic platforms:

Limited Payload Capacity: Aerial robots have restrictions on payload capacity, which can limit the types of sensors and equipment that can be carried for inspection. This limitation can be addressed by optimizing the design for lightweight components and utilizing advanced miniaturized sensors.

Restricted Maneuverability in Confined Spaces: Aerial robots may face challenges maneuvering in tight and confined spaces within tanks, especially with narrow openings like manholes. This limitation can be mitigated by designing robots with collision-tolerant features and advanced control algorithms for precise navigation.

Limited Flight Time: Aerial robots typically have limited flight times, which can constrain the duration of inspection missions. This limitation can be addressed by optimizing energy efficiency, using swappable batteries for extended operation, or implementing autonomous docking and recharging stations.

Susceptibility to Environmental Factors: Aerial robots are susceptible to environmental factors such as air currents and turbulence, which can affect stability and control during inspection. This drawback can be addressed by implementing robust control algorithms and sensor fusion techniques for better stability and adaptability to changing conditions.

Given the importance of detecting structural deformations, how can the autonomy system be integrated with advanced computer vision and structural analysis techniques to provide more comprehensive and reliable defect detection capabilities?

Integrating advanced computer vision and structural analysis techniques into the autonomy system can significantly enhance defect detection capabilities:

Defect Recognition Algorithms: Implementing deep learning algorithms for defect recognition in captured images can enable the autonomy system to automatically identify and classify structural deformations. Training neural networks on a diverse dataset of defects can improve the system's accuracy and reliability in detecting anomalies.

3D Reconstruction and Analysis: Utilizing 3D reconstruction techniques from LiDAR and camera data can provide a detailed spatial representation of the tank environment. Structural analysis algorithms can then analyze this 3D model to detect deformations, cracks, corrosion, or other defects with high precision.

Semantic Segmentation: Applying semantic segmentation algorithms to images can help differentiate between inspection-important areas and irrelevant surfaces. By focusing on inspection semantics, the autonomy system can prioritize defect detection on critical structural components, enhancing the efficiency of the inspection process.

Integration with Structural Health Monitoring Systems: Integrating the autonomy system with structural health monitoring systems can enable real-time analysis of structural integrity. By combining sensor data from the robot with structural analysis techniques, the system can provide continuous monitoring and early detection of deformations or damages, ensuring timely maintenance and repair actions.