The parametric design of mechatronic systems requires several detailed analyses of the system, thereby slowing down the design process significantly. In the recent past, there has been a lot of interest in using lower fidelity, but higher efficiency metamodels (also called surrogate models) instead of the actual detailed models to guide parametric design, particularly in the early stages of parametric design. One common approach to forming metamodels is to run the detailed model to obtain the system response at selected points in design space and fit a response surface to the results which becomes the metamodel. Since this method uses only zero order information at each design point, a large number of points are required to form a reasonably accurate metamodel. For example, in a single design variable problem, a two-point response surface can only be linear, whereas we can generate a cubic response surface if we also had derivative information at the two points. In this paper, we present a metamodeling approach for mechatronic systems that computes and utilizes first order derivative information at each point in the design space at which a detailed analysis is performed. The first order derivative information that is computed is the set of design sensitivity coefficients of the system state variables and performance functions. A unified modeling approach for the mechanical, electrical, and electronic aspects of the system is first developed. This approach generates a single set of governing equations for the entire system in the form of a system of differential-algebraic equations (DAE’s). Based on these DAE’s, a set of equations in the state design sensitivity coefficients is analytically derived using a direct differentiation approach. This set of equations also turns out to be a set of DAE’s which can be solved simultaneously in parallel with the system governing equations. We have successfully implemented this methodology for design sensitivity analysis of multidisciplinary systems in a computational platform called MIXEDMODELS (Multidisciplinary Integrated eXtensible Engine for Driving Metamodeling, Optimization, and DEsign of Large-scale Systems). Once we know the state design sensitivity coefficients, we can compute the design sensitivity coefficients of any system performance function. After we have obtained the necessary design sensitivity information, we can devise several schemes for generating a metamodel for the system based on the sensitivity information. Some examples of metamodels obtained using this approach are presented for selected mechatronic systems, along with the relevant accuracy measures.

This content is only available via PDF.
You do not currently have access to this content.