Network Theory and Global Sensitivity Analysis Framework for Navigating Insights From Complex Multidisciplinary Models

Abstract

In conventional vehicle design approaches, there is typically little understanding of the consequences of early stage design choices. This may be attributed to the conventional approach’s limitations in capturing complex interactions, which leads to increased design iterations. To overcome this, holistic multidisciplinary models have been developed. However, these models introduce the burden of complexity and costs due to their intricate nature. Additionally, it is challenging to gain meaningful insights without a deeper understanding of the model’s structure and behavior. In this article, an alternative form of model representation is proposed to address these shortcomings. This was achieved by integrating two concepts: network theory and sensitivity analysis. A detailed and robust framework is provided, which represents complex multidisciplinary models as network models, reducing their complexity and allowing insights to be extracted from them. This approach was demonstrated through a case study of a rail vehicle traction system, including a traction motor and an inverter, coupled with an operational drive cycle. Among the 246 factors identified in the traction system network model, the three most influential inputs were determined for the selected output factor of interest. Subsequently, the knock-on effects of these inputs were assessed. The results indicate a significant reduction in the network graph size compared to the complete network graph of the traction system model, highlighting a substantial decrease in the number of factors considered in the analysis. This demonstrates the capability of the proposed framework to simplify the analysis while retaining the ability to examine intricate interaction effects.

Keywords

• Change propagation analysis
• global sensitivity analysis
• knock-on effects
• metamodels
• multidisciplinary models
• network models