Traditional bridge and road inspection methods are
time-consuming and expensive, requiring a lot of coordination, such as traffic
control, and they put personnel in danger. They involve labor-intensive
techniques like visual inspection, thermal imaging, ground penetrating radar,
and acoustic inspection, all of which often necessitate complex traffic
lane-closure management, additional labor hours, expensive equipment, and can
potentially place workers in unsafe environments.
The United States alone has close to 600,000 bridges
dispersed across the roadway network. Federal regulations mandate that each
bridge that has a length of 20 feet or greater must be inspected at least once
every 24 months.
Drones offer a faster, safer, and more cost-effective
approach. They can access areas that are difficult or dangerous for humans,
such as under bridges or along train tracks. Studies indicate significant cost
reductions. Drones can collect data from a variety of angles, aiding in
identifying problems that would not be visible from a single vantage point.
They capture significantly more thorough inspection data and provide high-definition
photos from limited and inaccessible areas, such as beneath bridges and along
beams and girders. Drone inspections can reduce the environmental impact of
bridge and road inspections by reducing the need for traffic control and other
measures that can disrupt traffic and pollute the air.
Drone inspection of infrastructure systems allows for data
collection beyond capturing images using camera alone. Drones enable
integration of high-resolution optical camera with infrared (IR) camera. The IR
camera is crucial because it can provide more details to the interior
structural damages of a bridge or a road surface than an optical camera, which
is more suitable for inspecting damages on the surface of a bridge or a road.
Drones can be
equipped with a minicomputer that runs Machine Learning algorithms enabling autonomous
drone navigation, image capture of the bridge or road structure, and analysis
of the images. When damage is detected, the location coordinates are saved. Machine
learning models are used to detect defects from subscribed image topics. These
features enable the drone to self-operate and carry out the inspection process
using AI algorithms, reducing the need for human operation.
The machine learning capabilities that can be used for
defect detection include deep learning models. Deep Convolutional Neural
Network (DCNN) models can be employed for image classification. Two main types
of models are utilized:
· Classification Deep Neural Network: Can be used on the drone itself for quickly analyzing photos and classifying the photos into one of any number of groups, such as faulty roadways or okay roadways. These models perform a preliminary analysis.
· Region-Based Convolutional Neural Network (RCNN): Can be used for ground station analysis due to higher processing power requirements. These models can localize, identify, classify, and bound objects inside the photo itself. They prioritize images flagged as faulty roadways by the classification models.
Sample images of potholes and manholes should be collected
using both visible light and thermal cameras to train the neural networks. The
images undergo feature extraction through multiple convolution and pooling
layers.
The regulatory considerations relevant to AI-enabled drone
inspection of infrastructure systems encompass drone licensing, airspace
restrictions, and bridge inspection regulations.
