Sound Source Identification in Vehicles Using a Combined Linear-Evolutionary Product Unit Neural Network Model

Abstract

The successful development of new products relies on the capacity to assess the performance of conceptual design alternatives in an early phase. In recent years, major progress has been made hereto, based on the extensive use of prediction models, particularly in the automotive industry. The proposal of this paper is to develop a novel vehicle noise prediction model based on the combination of a Linear Regression model and Evolutionary Product Unit Neural Networks (EPUNNs). Several methods were compared depending on the frequency of the noise since the system has a linear behavior at low frequencies and a more random one at high frequencies. The accuracy of all models has been evaluated in terms of the Mean Squared Error (MSE) and the Standard Error of Prediction (SEP) obtaining the smallest value for both measures when using the combined model (Linear Regression at low frequencies and EPUNN at high frequencies).