Left-inversion of nonlinear fading memory systems from data

A method for the left-inversion of nonlinear fading memory systems from data is proposed. The method is based on the identification of a model of the system to invert, and the computation of the left-inverse directly from this model. It is not required to identify an inverse system. Such an identification is in general more difficult than the identification of the ¿direct¿ system. The invertibility of the regression function defining the system is also not required. The inversion error, defined as the difference between the desired output and the actual system output, is shown to be bounded by the identification error, measured by the L _¿ norm of the difference between the system and the model. The nonlinear set membership identification approach is used for the identification of the model. This approach provides models with minimal identification error. A simulation example on the inversion of a nonlinear dynamic semi-active suspension shows the effectiveness of the method.