Keywords

Ground penetrating radar; semi-autonomous/autonomous GPR image interpretation; intelligent data analysis; water distribution network health asset assessment; water leakage

Start Date

7-7-2022 2:40 PM

End Date

7-7-2022 3:00 PM

Abstract

The provision of reliable, safe, affordable and continuously supplied drinking water is essential in the development of cities. To ensure sustainability, efficient use of water resources, and health assessment of such critical infrastructures as water distribution networks (WDNs), the improvement of monitoring/verification systems with the collection of reliable data at a reasonable cost is an impending priority. This paper seeks to facilitate the health assessment of WDNs assets, using the ground penetrating radar (GPR) as a reliable and powerful non-destructive method for capturing information (radargrams) from the subsoil, and through improved interpretation of those images. This paper uses a multi-agent-based GPR image pre-cleaning process and data pre-classification using a perceptron neural network to classify the objects of interest into various families of objects. Afterwards, the centroid of each group is treated in such a way that a shape can be obtained that can represent the potential area of influence of the GPR inspection cone. Each of the resulting shapes is then located at its corresponding centroid position. The relationships generated by the inclusion of these shapes are finally explored. Radargrams of leaking pipes, a variety components, and complex cases with more than one asset (which usually mask the target object making difficult its identification), are evaluated herein. The results suggest that it is possible to identify beyond the first reflection of the objects of interest, as well as to extract characteristics that allow delimiting the embedded target objects. The results are promising: the approach allows the selection of areas of interest within the images, which facilitates the subsequent classification process, while advancing in semiautomatic/ automatic interpretation of GPR images, thus reducing the dependency on human interpretability.

Stream and Session

false

Share

COinS
 
Jul 7th, 2:40 PM Jul 7th, 3:00 PM

Advancing GPR Interpretations by the Use of Multi-agent Systems and Intelligent Data Analysis for Water Distribution Health Asset Assessment

The provision of reliable, safe, affordable and continuously supplied drinking water is essential in the development of cities. To ensure sustainability, efficient use of water resources, and health assessment of such critical infrastructures as water distribution networks (WDNs), the improvement of monitoring/verification systems with the collection of reliable data at a reasonable cost is an impending priority. This paper seeks to facilitate the health assessment of WDNs assets, using the ground penetrating radar (GPR) as a reliable and powerful non-destructive method for capturing information (radargrams) from the subsoil, and through improved interpretation of those images. This paper uses a multi-agent-based GPR image pre-cleaning process and data pre-classification using a perceptron neural network to classify the objects of interest into various families of objects. Afterwards, the centroid of each group is treated in such a way that a shape can be obtained that can represent the potential area of influence of the GPR inspection cone. Each of the resulting shapes is then located at its corresponding centroid position. The relationships generated by the inclusion of these shapes are finally explored. Radargrams of leaking pipes, a variety components, and complex cases with more than one asset (which usually mask the target object making difficult its identification), are evaluated herein. The results suggest that it is possible to identify beyond the first reflection of the objects of interest, as well as to extract characteristics that allow delimiting the embedded target objects. The results are promising: the approach allows the selection of areas of interest within the images, which facilitates the subsequent classification process, while advancing in semiautomatic/ automatic interpretation of GPR images, thus reducing the dependency on human interpretability.