Analytic Model based Design of Experiments for Nonlinear Dynamic Systems with Constraints

Markus Stadlbauer, Christoph Hametner, and Stefan Jakubek


design of experiments, Fisher information matrix, nonlinear dynamic systems, neural network, model predictive control, constrained optimization


In this paper a novel analytic method for optimal model based design of experiments for nonlinear dynamic systems with incorporation of input, input rate and output constraints is presented. The purpose of experiment design is to excite an underlying process in a way that all the relevant information about the system becomes visible from measured input and output data. For complex processes, especially if the number of inputs is high the experimental effort can become tremendous, in terms of measurement time and costs involved with the execution of the experiment. Model based design of experiments (DoE) has proven to be a powerful method in order to make experiments as informative as possible concerning the reduction of experimental effort with simultaneous gathering of all relevant information of the process.In this context the compliance to system limits on inputs and outputs is decisive so that secure and stable operation conditions of the real system are guaranteed during the experiment. The analytic constrained optimization of the method proposed in this paper is based on multilayer perceptron (MLP) networks and its effectiveness is demonstrated for a dynamic nonlinear system.

Important Links:

Go Back