Journal of the Korean Geo-Environmental Society ISSN:1598-0820(Print) 2714-1233(Online) 한국지반환경공학회 논문집

When applying a service-life-based regression model to underground infrastructure failure prediction, newly identified variables can be assigned weights and incorporated, based on failure information collected during maintenance. This study obtained data for four groups ofcandidate new variables of the district heating pipe failure probability model, namely failure-investigation information (damage cause, pipe condition, fault grade, and discovery method), monitoring type, pressure, and flow velocity, todiagnose statistical weight-error types (circular reference, confounding, and variable redundancy)inthe model. Failure-investigation informationshowed a statistically significant association with service life (e.g., damage cause: F=16.36, p<0.001), but since it cannot be observed under normal operating conditions and is generated only after a failure occurs, it corresponds to a circular reference error and could not be set as a weight. For monitoring type, assuming it as an independent new variable and applying residual adjustment revealed what appeared to be confounding with service life, as the category ranking of the simple-ratio weights (0.540-2.605) was completely reversed after adjustment (1.337-0.763); however, this category was found to coincide with the existing cable-status variable at a specific value in 94.5 % of cases, indicating that this apparent confounding was itself an artifact of variable redundancy (multicollinearity), and that monitoring type could not be used as a new variable. Pressure and flow velocity obtained from hydraulic network analysis are generated independently of failures and do not correspond to circular reference. For pressure, the simple-ratio weights (0.439-2.262) narrowed to 0.687-1.322 after residual adjustment for existing variables, but the increase in failure risk with pressure persisted when the weights were recalculated separately for supply and return pipes, so pressure was judged applicable as a new variable. For flow velocity, the weight increased in the 1.0-2.3 m/s range, but 75.9 % of the pipes in this range are large-diameter pipes (400 mm or larger) and the trend disappeared within diameter groups, so velocity was judged inappropriate as a new variable.