Interval Methods for Model Qualification: Methodology and Advanced Application

Abstract

An actual model in simulation (e.g. in chemistry) or control (e.g. in robotics) is often too complex to use, and sometimes impossible to obtain. To handle a system in practice, a simplification of the real model is often necessary. This simplification goes through some hypotheses made on the system or the modeling approach. These hypotheses are rarely verified whereas they could lead to an inadmissible model, over approximated for its use. In this paper, we propose a method that qualifies the simplification validity for all models that can be expressed by real-valued variables involved in closed-form relations and depending on parameters. We based our approach on a verification of a quality threshold on the hypothesis relevance. This method, based on interval analysis, checks the acceptance of the hypothesis in a full range of the whole model space, and gives bounds on the quality threshold and on the model parameters. Our approach is experimentally validated on a robotic application.