Methods for Analyzing Complex and Holistic Interactions in Early-stage design
A framework integrating Network Theory, Sensitivity Analysis, and Structural Equation Modelling for analyzing interactions among subsystems in early-stage design
Time: Thu 2025-10-23 10.00
Location: Kollegiesalen, Brinellvägen 8, Stockholm
Language: English
Subject area: Vehicle and Maritime Engineering
Doctoral student: Sai Kausik Abburu , VinnExcellence Center for ECO2 Vehicle design, Fordonsteknik och akustik, Conceptual Vehicle Design
Supervisor: Carlos Casanueva, VinnExcellence Center for ECO2 Vehicle design, Järnvägsgruppen, JVG, Fordonsteknik och akustik; Associate Professor Ciarán J. O'Reilly, VinnExcellence Center for ECO2 Vehicle design, Fordonsteknik och akustik
QC 250922
Abstract
The conventional approach to vehicle design is inherently restrictive in analysing complex interactions and the resulting knock-on effects. This often leads to disharmonious solutions and increased design iterations. To overcome this, the focus needs to shift to early-stage design where there is freedom to explore design alternatives. But, this phase typically has limited information about system behaviour, entailing the need for holistic multidisciplinary models and robust methods. However, the methods and models available often focus on single aspect in isolation, which necessitates the need for an integrated framework.
This thesis proposes a framework that analyses complex and holistic interactions within multidisciplinary systems to investigate how the variation in input propagates as knock-on effects and ultimately influences the system outputs. To achieve this, the framework is structured into five distinct phases. Each phase aims to address the research questions formulated and fulfil the functions of the framework. The framework integrates multidisciplinary modelling techniques with network theory to capture complex interactions. Global Sensitivity Analysis (GSA) methods are combined with tailored network algorithms to identify the interactions relevant to a specific input and output. These identified interactions are then quantified using curve fitting and two directional sensitivity measures, Average Local Sensitivity Coefficient (ALSC) and Average Combined Sensitivity Coefficient (ACSC), which were proposed in this thesis. Finally, Structural Equation Modelling (SEM) is utilised to investigate how input variations propagate as knock-on effects through intermediate variables, ultimately influencing system outputs.
The framework's capabilities were demonstrated through two case studies. An intra-subsystem interaction analysis with traction motor and an inter-subsystem interaction analysis involving a traction motor and a passive cooling model of an inverter. In the intra-subsystem case study, the framework successfully identified three influential inputs (voltage (U), rated power (Prated), and frequency (f)) for the chosen output of interest (rotor resistance (R'r)), and reduced the number of factors to consider in the analysis by 92.51% for U, 89.42% for Prated, and 93.83% for f. The results obtained from the framework was compared to the results from GSA methods, which showed a maximum error of 3%, thereby validating the proposed framework's ability to calculate the knock-on effects and the total impact while preserving the input-output relationship. Furthermore, the framework revealed insights into nuances of knock-on effects and total impact, such as the finding that the rated power induced more design changes despite having similar influence on the output as frequency.
In the inter-subsystem interaction analysis, the framework was similarly able to identify the influential input (Prated) for the chosen output of interest (Tbase,max), and reduced the number of factors to consider by 57%. The validation step revealed an error of 6.77%, which was acceptable considering the complexity of the network that involves 3568 paths between the input and the output. Furthermore, while calculating the direct ALSC value using curve fitting, it was observed that Tbase,max had distinct clusters across specific ranges of motor power values which was attributed to the presence of incorporated design margins and complex interactions in the traction motor model.
Thus, this thesis delivers a framework that is capable of systematically capturing, quantifying, and analysing complex and holistic interactions in a multidisciplinary system to investigate and quantify the knock-on effects and the total impact of varying an input on an output. It serves as a valuable comprehensive guide in early-stage design for designers to identify an effective input to significantly influence a specific output and understand the consequence of modifying that chosen input.