Characterisation of Nonlinear Structural Dynamic Systems in Conceptual Design
Abstract: The engine and driveline systems of passenger cars generates and distributes the necessary driving power and are major contributors to vehicle emissions, noise and vibrations, etc. More environmental friendly technologies under development are expected to intensify and add new comfort related problems, since most of them affect vibration sources or system damping. A successful balancing of fundamental system qualities requires a better use of simulation in early design phase. This work focus on virtual tools for analysis of low-frequency structural dynamic vibrations. In conceptual driveline design, many possible system solutions are studied in parallel and their often nonlinear behaviour requires robustness evaluation across full operating and design parameter ranges. This situation calls for virtual methods that are generally valid and meet the demand for rapid prototyping. Thus, models need to be as simple as possible and as accurate as required for capturing phenomena that occur in real drivelines. Further, analysis tools must efficiently process data sets from extensive parameter variations and extract fundamental system characteristics that can be used to reliably rate competing proposals. For this, a complementing design analysis methodology is proposed that improves current automotive development tools and workflow. A general and over-parameterised multi-body system model is constructed from detailed linear structural and schematic nonlinear parts. State-space reduction methods are then applied to modal components to balance prediction accuracy and evaluation speed of resulting conceptual design models. Parameter variations in fully known system models are simulated under ideal periodic loading and low noise conditions. A feature based frequency analysis approach is used to extract precise system characteristics and sort responses into qualitative classes. To efficiently process large amounts of generated data, statistical learning methods are used to automate the response classification.
CLICK HERE TO DOWNLOAD THE WHOLE DISSERTATION. (in PDF format)