Bridgebase: A Knowledge Graph Framework for Monitoring and Analysis of Bridges

Abstract

AbstractInfrastructural budgets on bridge maintenance and monitoring always demands for figuring out optimised ways which could perform rehabilitation work efficiently. In this work, we try to create a pipeline for (1) transforming inventory data to a knowledge graph which supports more readability by defining a question answer system (2) converting bridge nodes from the knowledge graph into embeddings based on sparse knowledge about bridge ratings and, (3) clustering similar bridges based on parameters to perform maintenance work easily. In this work, a knowledge graph framework is presented, then the nodes of the knowledge graph get converted to latent-vectors using Deepwalk. The conversion of bridge nodes to a latent vector transformation leverages essential knowledge graph structure, which makes it easy to cluster (by selecting appropriate features based on the formed graph structure) using Density Based Clustering (DBScan). The main objective of this work is to cluster similar bridge vectors, with less knowledge about the dataset and bridge properties, hence making it appropriate for streaming consumption. We create a real time bridge monitoring system (visualiser) with any employed structure. The efficiency of bridge node embedding is evaluated based on National Bridge Inventory dataset and the F1-scores were obtained with varied concentrations of unlabeled data. The obtained results showed how Deepwalk can perform better bridge node embedding (higher F1 scores) than other algorithms which show lower accuracy with sparse knowledge. As the knowledge graph gets stored in graph database systems, it opens up an interface for easy integration with Internet of Things applications.