Online measuring method using an evolving model based test design for optimal process stimulation and modelling

For data driven modelling the information content of system input data and measured output data is decisive for the achievable model quality of the underlying process. Process stimulation is targeted to maximize the information content per data sample, in order to limit the measurement time. Especially for processes with an increasing number of system inputs the experimental effort is continuously rising. Therefore methods for an efficient process excitation combined with advanced modelling strategies are necessary. In this context online methods, where the design of experiments and the model training are in parallel to the ongoing experiment are a very promising approach for an efficient generation of process models. The compliance with constraints on the system input as well as on the system output is essential in order to provide secure and stable operational conditions during the experiment. In this paper a recursive algorithm is proposed, which uses an evolving local model network for the online generation of optimal dynamic experiments under constraints. The effectiveness of the proposed method is demonstrated on a nonlinear dynamic exhaust temperature model of an engine and a comparison with a standard excitation signal is given.