Data-Driven Sewer Pipe Data Random Generation and Validation

Abstract

Sewer pipe systems are of great importance to modern cities in various ways, making preventive maintenance a necessary activity to ensure an acceptable level of service at all times. In this respect, closed-circuit television (CCTV) inspection data for sewer pipe systems serve as the basis for preventive maintenance in the context of sewer pipe condition ratings, maintenance schedule planning, and other similar ideas. Defects (i.e., those classified as either cracks, fractures, roots, deposits, broken, or holes) and construction features (i.e., taps) are the targets of the CCTV inspection process, which is used to mark and record the defects and features in the inspection database for the purpose of developing maintenance strategies. In considering sewer pipe maintenance operations in practical terms, the following CCTV inspection data for sewer pipes are of particular interest to this research: length of the pipes, defect interval, and defect sequence for different types of defects (and taps). However, the data collection process using CCTV inspections is typically expensive and time-consuming from the perspective of the municipal department. In this context, an input modeling technique that aims to exploit the potential value of historical data is proposed by combining the Markov chain model with distribution fitting techniques and other relevant methods. The generated dataset goes through a rigorous validation process that includes statistical analysis and comparison, cluster analysis and comparison, and distance-based similarity comparison. The whole process proves that the randomly generated dataset is reasonable since it expresses similar characteristics to the original dataset in many aspects. Overall, the research proposes an input modeling process that could generate human-made sewer pipe inspection data that inherent the major characteristic of the real-life data. The generated data could benefit the real-life practice in various ways, especially in the context of data deficiency.