The Potential of Data-Driven Engineering Models: An Analysis Across Domains in the Automotive Development Process

Abstract

<div class="section abstract"><div class="htmlview paragraph">Modern automotive development evolves beyond artificial intelligence for highly automated driving, and toward an interconnected manifold of data-driven development processes. Widely used analytical system modelling struggles with rising system complexity, invoking approaches through data-driven system models. We consider these as key enablers for further improvements in accuracy and development efficiency. However, literature and industry have yet to thoroughly discuss the relevance and methods along the vehicle development cycle. We emphasize the importance of data-driven system models in their distinct types and applications along the developing process, from pre-development to fleet operation. Data-driven models have proven in other works to be fast approximators, of high accuracy and adaptive, in contrast to physics-based analytical approaches across domains. In consequence, we show the necessities and benefits of adopting such models by analyzing the current methods used in industry. We derive commonalities in approaches and applications across domains to subsequently provide detailed case studies along the development cycle. Here, we highlight essential data acquisition concepts and suggest promising approaches for four different engineering use-cases, while pointing out limitations and pitfalls in application. Conclusively, we present our perspective on further challenges and opportunities in the evolution of the automotive industry in terms of data-driven system models for technical use-cases.</div></div>