Information about the location of modeling erros is crucial for the efficient improvement of an invalid model. This article discusses how to pinpoint modeling errors through comparison of experimental data with data obtained through simulation of the invalid model. An observer-based approach is presented. By designing a dedicated observer for the system using the invalid model, a signal vector is generated, on which each modeling error imposes an easily identifiable feature. An algorithm to analyze the featured signal is then presented. With this algorithm, the features of each of the modeling errors are extracted. The approach is illustrated for a denitrification reactor model in which errors in the dimension of the state vector, in the structure of the biokinetic relationship and in the values of the parameters could be identified.

You do not currently have access to this content.