Physical system modeling: Algorithms for assessing model quality based on design specifications

Physical system models, with physically meaningful states and parameters, have become increasingly important, parts of the design, control and even procurement processes. However, while algorithms have been developed to help formulate and integrate physical system models, and while algorithms have been developed to generate minimum complexity physical system models subject to a modeling metric constraints, missing from the literature are the algorithms to assess the “quality” of dynamic system models. The objective of this research is to develop a methodology to quantify the accuracy of the predicted system variables and to determine the validity of a model of given complexity with respect to a set of criteria derived from engineering design specifications or measurement noise. Another objective of this thesis to integrate the methodology with an activity based MORA (Louca et al., 1998) to quantify the accuracy of the different models as a function of activity model complexity.; A&barbelow;ccuracy & V&barbelow;alidation A&barbelow;lgorithm for Simulation (AVASIM), proposed in this dissertation, is a time-domain perspective comparing the model's time based output trajectories to targets at user-defined points. In addition, the model accuracy over the total simulation horizon is also used. The tolerances on specific time point and overall response are determined by the user's domain knowledge (design specifications) or the measurement noise associated with empirical validation studies. The performance of the models is formulated as a performance index that is determined from measures such as relative error, residuals sum, etc. to quantify the accuracy with respect to target points and overall response. The validity of model is then determined by comparing the performance indices to that of a “threshold model” derived from the design specifications or measurement noise on a data set.; To illustrate the AVASIM, case studies based on automotive applications are presented. Results from the case studies show that accuracy and validity of the reduced models can be systematically assessed using the proposed AVASIM. Integration of AVASIM with MORA seems to be a very powerful design tool to generate the proper design model with desired accuracy.