Maxime Deregnaucourt, M. Stadlbauer, C. Hametner, S. Jakubek, J. Wurzenberger:

"Nonlinear conjugate gradient algorithm for model based design of experiments";

Vortrag: 6th European Congress on Computational Methods in Applied Sciences and Engineering (ECCOMAS 2012), University of Vienna; 10.09.2012 - 14.09.2012; in: "Proceedings of 6th European Congress on Computational Methods in Applied Sciences and Engineering (ECCOMAS 2012)", (2012).

This paper shows how to apply the methodology of model based design of experiments (DoE) for the identification of nonlinear dynamic systems on a combustion engine. This relies on an algorithm developed to solve the problem of generating informative data from the process with low experimentation effort using a reference model of the same process. The algorithm optimizes the input signal by maximizing the determinant of the Fisher information matrix computed with the reference model. At the same time the computed signal complies with constraints on the system inputs, input rates and output. The algorithm is an optimized iterative constrained gradient algorithm that relies on the nonlinear conjugate gradient method with inexact line search. It solves the constrained nonlinear programming problem using a penalty or barrier function, or using the Karush-Kuhn-Tucker conditions. The effects of the conjugate direction on the efficiency of the algorithm are discussed. The proposed model based DoE method is applied to a complex nonlinear dynamic engine simulation model. The simulations show significant improvements of the presented model based DoE compared to conventional DoE.

design of experiments, nonlinear system identification, nonlinear conjugate gradient method, barrier function, inexact line search

Erstellt aus der Publikationsdatenbank der Technischen Universität Wien.