A Framework for Virtual Knowledge Graph Construction over Time Series Data

Abstract

Analyses of products during and after development are essential to improve and guarantee their quality. In the automotive industry, analyses are often based on time series data coming from numerous heterogeneous sensors equipped by vehicles. The analysis of time series and sensor data is challenging due to the high volume and high variety of the data, often leading to unstructured storage solutions. In this work, we show how these challenges can be addressed relying on data preprocessing and a Virtual Knowledge Graph (VKG) approach. The data preprocessing is used to create tables from the time series data that integrate efficiently with the VKG, and the VKG itself introduces semantics, virtualization, and a graph-representation of the data. To experiment with the setup, we preprocessed the data, developed an ontology for the virtual knowledge graph, and developed a set of mappings that connect the graph to the preprocessed data. The system is evaluated using a set of analysis questions derived from a real business use case that we have encoded as SPARQL queries. The results obtained are promising, showing that a virtual knowledge graph is a viable alternative to current analytical approaches for time series data.