Abstract

Current asset management practices of municipal infrastructure focus on modeling water, sewer and road networks into isolated layers and hence overlook the spatial and functional interdependencies among these assets. For instance, the performance of an asset can be impacted spatially and/or functionally by its neighboring assets. Vulnerability assessment, in this study, measures the asset’s degree of susceptibility for structural and/or functional failures triggered by failure of these functions in neighboring assets. The objective of this research is to develop a computational framework for optimizing intervention policies for likely vulnerable civil infrastructure networks considering spatial and functional interdependencies. The developed framework integrates three models; 1) interdependency assessment model, 2) vulnerability assessment model and 3) system dynamics model.

The interdependency assessment model captures spatially and functionally interdependent assets utilizing two developed modules: spatial interdependency module and functional interdependency module. The spatial module utilizes ArcGIS geoprocessing tools in determining geographically interdependent assets. It encapsulates interdependent assets in a set of new layers and a newly developed database containing characteristics of such interdependencies. On the other hand, the functional module employs graph theory principles in determining an asset's degree of connectivity with its neighboring assets. The functional module will aid in recognizing the likely influence of an asset failure on its neighboring assets' performance using betweenness centrality. The output of the assessment model is in the form of bundles of spatially and functionally interdependent assets.
For vulnerability assessment, three computational models are developed and experimented with to rate vulnerability of civil infrastructure systems considering their spatial and functional interdependencies with neighboring assets. These models are; 1) multi-attribute utility theory (MAUT), 2) artificial neural network (ANN) and 3) fuzzy c-mean clustering (FCM). Operation and maintenance reports obtained from two Canadian municipalities (the Cities of London and Hamilton, Ontario) were used to select factors influencing the vulnerability of water, sewer and road assets. For the MAUT model and based on the identified factors from operation and maintenance reports, surveys were sent to 65 experts and their feedback was elicited to construct utility functions to rate the degree of vulnerability of interdependent assets. The response rate of the survey was 75%. On the other hand, the ANN model utilizes self-organized mapping algorithm (SOM) to rate vulnerability of these assets based on recognized patterns in each dataset. The ANN model is a data driven model requiring sufficient amount of observed patterns and extensive effort in modeling with less involvement from experts. On the other hand, the FCM model is capable of accounting for ambiguity and imprecision associated with experts’ input in rating vulnerability of interdependent assets.

Subsequently, the system dynamics (SD) model is developed to help identify possible least cost intervention policies for interdependent infrastructure assets that meet customers' expectations and decrease assets’ vulnerability. The developed SD model consists of 23 variables and 8 causal feedback loops. These causal loops are developed based on the reviewed literature and four unstructured interviews with three experts in the domain of municipal asset management; one from the City of London and two from the City of Hamilton. The SD is augmented with two optimization algorithms to find optimal intervention policies at bundle and network levels; 1) dynamic programing algorithm, 2) single objective genetic algorithm.

Two case studies were analyzed and presented to demonstrate the application of the proposed framework and its expected contributions using data obtained from the Cities of London and Hamilton, Ontario. The interdependency model constructed 10,500 bundles for the City of London and 12,350 bundles for the City of Hamilton. For the vulnerability model, the developed FCM model showed better performance than ANN in mimicking experts’ judgement. The mean square error (MSE) for the FCM model was 42% less than that of the ANN model. Also, there was a linear correlation between the number of breaks for water assets and their vulnerability ratings (R2=0.79). When the SD model was supplemented by the modified genetic algorithm, the computational time for finding near optimal solutions at network level was decreased by 50% for the City of London and by 47.2 % for the City of Hamilton when compared to traditional genetic algorithm.

The results of the developed vulnerability and SD models were shared with the experts. The developed vulnerability models will be useful for staff to justify increases to intervention budget to each City Council. In spite of the relatively complicated nature of ANN and FCM models, the experts were relatively comfortable using these models. However, the experts commented that this might not be the case with other municipalities that are still starting their asset management programs. For the SD model, the experts agreed that the model is beneficial for identifying possible least cost intervention policies at bundle and network level. They however pointed out that the SD model can be enhanced by accounting for factors related to social and economic characteristics of
their customers. The modified genetic algorithm can be enhanced more by the deployment of parallel computing techniques to decrease its computational time.