Sensor Data Interpretation with Clustering for Interactive Asset-Management of Urban Systems

Abstract

In responsive cities, user feedback and information provided by sensors are combined to improve urban design and to support asset managers in performing decision making. Optimal management of infrastructure networks requires accurate knowledge of current asset conditions to avoid unnecessary replacement and expensive interventions when cheaper and more sustainable alternatives are available. Structural model updating is a discipline that focuses on improving behavior-model accuracy by means of measurements taken from the built environment. Error-domain model falsification (EDMF) is a simple and practice-oriented methodology that uses measurements at sensor locations to identify plausible models among an initial population generated according to engineering judgment. However, many plausible models are often identified, making result interpretations difficult for practicing engineers. In this paper, a clustering methodology based on bipartite-modularity optimization (BMO) is used to clarify identification outputs. Compared with classical clustering methods such as K-means, BMO clustering provides more accurate interpretations and better visualization of the results. Moreover, engineers can actively interact with the clustering framework to obtain the knowledge that is needed at several stages of the decision-making process.