Abstract:
Nonlinearities in sensors, such as distortion, saturation, and dead-zone, can cause severe
performance deterioration of control systems. In this work, nonlinear sensors are modeled using a
block oriented nonlinear model, called the Wiener model, which consists of a linear dynamic system
followed by a static nonlinear system. The nonlinearities are then compensated either by using an
inverse static model or as the output of the linear dynamic part of the identified Wiener model. For
simplicity, we propose the estimation of the static nonlinear of the Wiener model as a piecewise
linear function. The selection of the optimal structure is based on cross validation. Three model
selective criteria, i.e. the predicted residual error sum of squares (PRESS), the Mallow's CP (CP),
and the final prediction error (FPE) have been applied and compared. Based on the simulation
results, we conclude that the most leading criterion that performs well in selecting the structure with
the smallest number of parameters is the FPE. We also suggest reconstructing the intermediate
variable by filtering the input through the identified linear dynamic system as it gives better
prediction results than the ones obtained from the inverse approach.