Knowledge Graph for CAE-based Development in Vehicle Safety
This dissertation presents a comprehensive exploration of Knowledge Graphs (KGs) to transform the domain of vehicle development, with a primary focus on improving crashworthiness. The proposed KG leverages insights from diverse structured and unstructured data sources, encompassing critical concepts within the automotive domain. This research serves the industrial goal of capturing and preserving knowledge derived from Computer-Aided Engineering (CAE) workflows, enabling
data reuse, improving guidelines, and facilitating Machine Learning (ML) applications. Creating a domain-specific KG is challenging, especially when dealing with complex CAE data with intricate 3D deformations. This complexity is rarely encountered in most existing text-based KGs. This thesis
presents a comprehensive path from data modelling to practical ML applications based on a concrete use case to address this challenge. This approach provides a cyclical feedback loop between data modelling and feature extraction, enhancing the utility of both processes.The research encompasses several key aspects, including data modelling and ontology development, feature engineering, data visualisation for knowledge discovery, and ML implementations to predict relationships between simulations. It innovatively transforms crashworthiness analysis into graph representations, uses SimRank to analyse weighted bipartite graphs, and abstracts vehicle structures to rank simulations based on load-path similarity. Additionally, a user-friendly web application for this project is developed
using Django and hosted on GitHub1, ensuring accessibility and ease of use for industry professionals and researchers alike. This work highlights the immense potential of KGs in the automotive sector, facilitating a data-driven approach to vehicle development and significantly improving crashworthiness analysis through the integration of CAE data.
