AI-Guided System Proposed for Robotic Inspection of Roads, Buildings, and Bridges

Our modern society faces a crisis of crumbling infrastructure since many of our buildings, roads, bridges, and tunnels show signs of damage and decay. These structures are at risk of failing or collapsing unless they are inspected and repaired in time. However, the current infrastructure inspection methods are costly, slow, labor-intensive, and cannot cover the vast scale of the problem.

(Photo: Wikimedia Commons/ Greg Henshall)

The Problem of Aging Structures

Inspection is important to maintain the integrity of infrastructures. Without regular inspection, failures can occur, leading to disruptions with far-reaching consequences.

Traditional methods of infrastructure inspection involve physical examination by trained personnel. It usually includes visual inspection, ultrasonic testing, and thermal imaging to detect anomalies that can indicate potential issues.

These methods put inspectors at risk of accidents, especially when they encounter electrical environments with high voltage. Since most equipment is located in hard-to-reach areas, inspection becomes challenging and may sometimes require specialized tools or techniques.

Aside from this, traditional inspection methods also rely heavily on the expertise and judgment of the personnel. Although their skill and experience are invaluable, there could still be human error that cannot be eliminated. This is usually the case when inspections are done under challenging conditions or when the personnel is pressured to complete the task within tight schedules.

Preventing Structural Failure

To address this concern, experts from Drexel University tried to give robotic assistants the tools to help inspectors with the job. The paper describes Their research as "A multi-scale robotic approach for precise crack measurement in concrete structures."

The researchers augmented visual inspection technologies with a new machine-learning approach. The multi-scale system combines computer vision with a deep-learning algorithm to detect problem areas of failure before directing a series of laser scans of the regions. This is followed by creating a "digital twin" computer model to assess and monitor the damage.

Instead of using physical measurement interpreted subjectively by human eyes, the system uses a high-resolution stereo-depth camera feed of the structure into a deep-learning program called a convolutional neural network. These programs are currently used in facial recognition, deepfake detection, and drug development. However, they have gained attention for their potential to spot the finest discrepancies in massive volumes of data.

Traditionally, human inspectors would make the final call on when and how the damages will be repaired. In the new system, the robotic assistants can significantly reduce their workload as the automated inspection process would lessen oversights and subjective judgment errors that can occur when overworked personnel take the first look. Furthermore, this approach can reduce unnecessary data collection from areas in good structural condition while offering comprehensive and reliable data for condition assessment.

According to the research team, the system represents a strategy that can significantly reduce the overall inspection workload and allow the focused consideration and care needed to avoid structural failures. They plan to integrate this innovation with an uncrewed ground vehicle to enhance the system's ability to autonomously detect, analyze, and monitor cracks.