Abstract
Sewer condition prediction is a fundamental element of proactive maintenance programs. The prediction relies mostly on the assessed condition of inspected segments, generally based on CCTV reports. However, several sources of uncertainty affect the condition assessment and may lead to inefficient maintenance. The present article focuses on three main questions. 1. What is the impact of uncertainty in assessed condition on the prediction model? 2. Considering uncertainties in the assessed condition, is it necessary to collect data on the characteristics of many segments, or are a small number of influential variables enough to build the condition prediction model? 3. Is it better to overestimate (false positive) or underestimate (false negative) the deterioration of a segment? These questions were evaluated on a semi-virtual asset stock and the results confirm that uncertainties affect the inspection efficiency negatively. Results also show that errors leading to the overestimation of the deterioration have less negative impact. The study suggests that data from a small number of influential segments is adequate to inform the prediction model.
