Data-Driven Framework for Modeling Productivity of Closed-Circuit Television Recording Process for Sewer Pipes

Abstract

Closed-circuit television (CCTV) is widely used in North America for sewer pipe inspection due to several benefits, such as easy operation and lower upfront costs. To be useful, video footage needs to be collected according to specific standards, which makes the video recording process a time-consuming operation, especially when pipes have operational issues like debris or tree roots. As a result, because city managers are usually limited by the available budget, a good understanding of the overall requirements for CCTV sewer pipe inspection is necessary for efficient resource planning. In this respect, a framework is proposed to model the productivity of the CCTV video recording process by predicting the duration of the recording process based on selected variables. In order to predict the CCTV recording duration, a type of machine learning algorithm and a linear regression model are developed. To be more specific, the random sample consensus (RANSAC) algorithm has been used to extract the benchmark for the CCTV recording process. This algorithm is adopted to screen the data automatically, arriving at a function of the CCTV recording time with two variables (i.e., the total length of the pipe segment and the number of taps in the pipe). As a result, the original dataset that records the CCTV collection process is segmented into three parts: benchmark dataset and two types of outlier datasets. Subsequently, two linear regression models are developed on the outliers to predict the recording duration. Finally, all the developed models are integrated into a simulation model to mimic the recording duration components. The framework is validated by historical data. For the convenience of implementation of the model, the parameters within the model are adjustable to adapt to different situations (such as different seasons, regions, and countries). The contribution of the research lies in two-folds: (1) the CCTV recording process is thoroughly investigated and well-understood, which provides a decision-making basis for the future CCTV collection process; and (2) the proposed simulation model development procedure can be applied to other studies that require data segmentation operation to improve the performance of the simulation model.