- Human-centered technique uses artificial intelligence (AI) and mixed reality to enhance inspections of civil infrastructures
- Saves inspection time and lowers labor cost
- Enables inspectors to perform a more objective assessment
Researchers at the University of Central Florida have developed a method for reducing the time and cost of inspecting civil structures like bridges and buildings. Conventional methods for visually assessing civil infrastructures use subjectivity and may require long inspection time and high labor costs. Although some technologies (such as robotic techniques, augmented reality, and mixed reality interfaces) can collect objective, quantified data, an inspector's expertise is still critical in many instances. Such technologies, however, are not designed to work interactively with a human inspector.
In contrast, the UCF Collective Intelligence Framework uniquely blends AI with mixed reality (MR) and can be integrated into an MR-supported see-through headset or a hand-held device. An inspector can analyze damage in real time and calculate or assess its condition without performing any manual measurements. At any step of the analysis/assessment, the inspector can intervene and correct the operations of the AI. Another advantage of the system is that the inspector can analyze defects in a remote location safely while reducing the need for access equipment. For example, if the defect location is hard to reach and poses a safety concern (like under a bridge or atop high-rise buildings), the headset can zoom in and still perform an assessment without needing equipment like a ladder. Consequently, the methods can reduce inspection time and labor costs while ensuring a quantified, reliable, and objective infrastructure evaluation with human-verified results. The methodology is expandable for many types of structures.
The UCF invention integrates unique AI detection and segmentation algorithms into an MR framework that performs automatic detection and segmentation of defect regions using real-time deep learning operations. In infrastructure assessment, creating a large image dataset for machine learning is essential but can be challenging. As a solution, the invention offers an advanced data augmentation technique that enables the framework to generate a synthetically sufficient number of images (like cracks and spalls) from available image data.
In an example application, an inspector assesses a concrete pier using an MR headset integrated with the framework. While the inspector performs routine tasks, the AI system in the headset continuously guides the inspector and shows possible defects in real time. Once the inspector confirms a defect location, the AI system starts analyzing it by first executing defect segmentation, then characterization to determine the specific type of defect. If the defect boundaries need corrections or the segmentation requires fine-tuning, the inspector can make the necessary adjustments. The system uses the alterations made by the inspector to retrain the AI model so that the AI's accuracy improves over time.
The research team is looking for partners to develop the technology further for commercialization.
Stage of Development
- Ensures reliable and objective infrastructure evaluation
- Reduces the need for access equipment by enabling inspectors to zoom in on far locations
- Allows effective use of data in infrastructure management systems
- AI continues learning and improves its accuracy over time
- Transportation agencies
- Building and construction inspection