Sensitivity of Markov Model to Different Sampling Sizes of Condition Data

Abstract

After many years of service, constructed infrastructure facilities including drainage pipes, bridges, and roads show sign of deterioration. To ensure public safety and efficient management of these crucial assets, condition monitoring and mathematical predictive models have been widely used. Despite the advance in condition-monitoring techniques, closed-circuit television (CCTV) and expert-based inspection technique are still commonly used owing to their ease of use, productivity, and lower cost. Utilizing those condition data, Markov models have been widely used as a predictive tool for asset management of constructed infrastructure facilities. However, the sensitivity of the Markov model to different sampling sizes of condition data has not been investigated. This has a practical implication as more owners of infrastructure facilities start to collect condition data and are interested in understanding current and future deterioration of their infrastructure assets. This study addresses this knowledge gap with a case study of stormwater pipes. The results of the case study show that Markov models are sensitive to the sampling size of condition data. A sampling size between 600 and 700 data points is recommended for industry to collect condition data since it could provide a good starting view on deterioration patterns of stormwater pipe networks while suffering from a less than 10% error rate.