Improved multiple-input/multiple-output modeling procedures with consideration of statistical information
Abstract
A new generalized Multiple-Input/Multiple-Output (MIMO) modeling procedure is developed for application to noise source identification and modeling of structural-acoustic systems. The new procedure extends the scope of the traditional procedure and is designed to derive a MIMO model which will accurately predict system responses for various system operating conditions. New source characterization techniques are developed for computing the multiple coherence function and for determining the number of (1) incoherent source processes, (2) independent operating conditions, and (3) input transducers to be used for developing an accurate MIMO model. The new techniques are computationally robust and are well suited for complex systems with a large number of source processes correlated in a complicated manner. A new frequency response function (FRF) estimation technique is developed. The new technique computes FRFs which are accurate for several system operating conditions. The new technique is particularly helpful for systems with completely coherent or harmonic processes. For distributed noise processes, the technique finds a best least squares FRF vector to fit all the system operating conditions. The new modeling procedure is successfully applied to a simple acoustic system. Two existing statistical methods, the method of generation of moments and the Monte Carlo simulation technique, are used for developing statistical MISO models of structural-acoustic systems. These two methods together provide a flexible, feasible and mathematically simple means of obtaining accurate statistical information of the system response. The method of moments is used to monitor which component random variables are the dominant contributors to the mean and the variance of the output random variable. The Monte Carlo simulation technique estimates complete statistical information for the output random variable. Both methods help monitor quality control. The methods of generation of moments and Monte Carlo simulation are applied to Single-Input/Single-Output data measured on samples of two different types of automotive vehicles. The results indicate the feasibility of developing useful statistical models by using the methods.
Degree
Ph.D.
Advisors
Bernhard, Purdue University.
Subject Area
Mechanical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.