Reduction of Experimental Data Points in the Base Calibration by Estimation of Engine Maps Using Regularized Basis Function Neural Networks

The estimation of calibration maps for engine control systems is not a trivial problem. Approximating these maps depends on experimental data obtained on an engine dynamometer, which may require a great number of test points. There are also some working regions in which steady state measurements cannot be taken. Thus, the map surface must be estimated from a finite set of data that does not cover the whole working conditions. High order polynomial models tend to produce oscillating functions, and low order ones do not present an accurate model. Therefore, this paper presents a method for the approximation of engine calibration maps with a Neural Network model, using a Regularized Radial Basis Function.