Utilizing Neural Networks for Condition Assessment of Sanitary Sewer Infrastructure

Abstract

This paper will describe the development of an improved methodology for accurately analyzing and interpreting data regarding the condition of sanitary sewer systems. The proposed methodology enables fast and accurate assessment, which is significant in building a sewer condition database for asset management. The inspection system obtains optical data from the Sewer Scanner and Evaluation Technology (SSET). Multiple neural networks are developed to classify the pipe defect features and a fuzzy logic system suggested to filter and fusion the multiple neural network outputs.

A Deep Dive into Computer Vision Aided Sewer Inspections

“…The Canny edge detector is utilized in different ways such as only considering horizontal or vertical edges in a sliding window approach [115], introducing an extra filtering step based on size and shape [116], and using multiple threshold settings to adaptively find the "core edges" [132]. In some cases, an undefined operator is used [33,37,39,55,92].…”

“…The NF approach was also compared with, and outperformed, a fuzzy K-NN classifier, and a normal K-NN classifier. Chae et al [16,[92][93][94] used an ensemble of MLPs to determine attributes of cracks, joints, and laterals in an image and applied a set of fuzzy rules to consolidate these rules into a condition rating for the pipe segment.…”