Application of neural networks in the development of nonlinear error modeling and test point prediction

This paper explores a neural network approach for empirical nonlinear error modeling for systems that have a significant amount of nonlinearity, nonlinear error models require fewer parameters compared to linear models and require fewer test points to achieve the same prediction accuracy. A neural network with a five-layer structure is investigated. The test point error predictions from nonlinear modeling are compared with the results of linear modeling for an artificial nonlinear model, a circuit with nonlinearity, and an instrument with suspected nonlinearity. The nonlinear modeling shows move improvement when the data set contains more nonlinearity.