| This thesis presents a technique to quantitatively search for decoupling among dynamic elements of a model, and to partition models in which decoupling is found. The method can increase the efficiency and accuracy of simulation-based design by improving physical-domain model reduction and preventing the use of inappropriate simplifying assumptions based on decoupling.; A full model is first created using the bond graph formalism. The relative contributions of the terms of the generalized Kirchoff loop and node equations are computed by calculating an aggregate measure of their power flow. Negligible terms are replaced by a power variable source modulated by the quantity common to the loop or node. If bond graph submodels joined by modulating signals result, and certain connectivity criteria apply, then the model can be partitioned into driving and driven subsystems. The driving partitions excite, but are not significantly affected by, the dynamics of the driven partitions.; Upon identification of driving and driven partitions within a model, the user can eliminate partitions that do not contain outputs of interest. If the output of interest is associated with an element in a driving partition, then the driven partitions can be eliminated in their entirety. Both dynamic and "kinematic" (power-conserving) elements may be eliminated. The driven partition, if it contains outputs of interest, can be excited by driving partition outputs from a parallel simulation, or the driving partition can be replaced with a data file as long as design changes do not modify the partition boundaries. Existing model reduction methods can be applied to each partition separately.; Three partitioning case studies were carried out. In the first, a pitch-plane vehicle dynamics model was partitioned into longitudinal and pitch subsystems. In a second case study, the dynamics of an in-line six-cylinder engine were partitioned into reciprocating dynamics (driving) and motion of the block on its compliant mounts (driven). The final case study demonstrated the applicability of the method to modal decomposition models of continuous systems.; The partitioning algorithm contributes to the simulation-based design state-of-the art by identifying more comprehensive model reduction opportunities, allowing the user to quantitatively verify and track simplifying assumptions based on decoupling, and generating models of intermediate complexity as the system parameters and inputs change and decoupling is diminished. The case study results show improved computation time and greater physical insight into decoupling than that afforded by ad-hoc decoupling approaches that rely on intuition and assumption. |
