Validation of vibro-acoustic numerical models

Arthur Blanc, Purdue University

Abstract

Models have historically been used to better understand and to optimize engineering designs. The development and refinement of numerical methods allow the modeling of increasingly complex systems. However, to gain full benefit of these increasingly powerful modeling methods requires that modeling approaches be validated. Validation increases confidence in the capability of the models and allows the rejection of flawed modeling strategies. In this investigation, the validation power matrix is introduced as a framework to objectively evaluate the predictive capability of a model. This framework assumes the system to be modeled is non-deterministic, typically because of manufacturing variations. The validation power matrix is composed of two power scales. The physical power scale is used to evaluate the ability of physical measures to describe the modeled phenomena. The statistical power scale is used to evaluate the hypothesis tests used to compare the model predictions to a comparison ( e.g. measurement) data set. High confidence in the model’s validity results from the combination of high physical and statistical powers. The validation power matrix is a general method that is adaptable to specific physical representations by defining appropriate physical power scales. Three physical power scales are defined for vibro-acoustic problems: modal decomposition, statistical energy analysis and wave decomposition. An efficient method is also introduced to compute the model response probability density function. This method uses design of experiments, metamodels and the change of variable formula from statistics. The method is shown to be accurate and more efficient than Monte-Carlo simulations. Finally, the validation power matrix approach is illustrated for the validation of the model of a compressor housing by first considering components of the compressor and then the assembled system. The physical power scale for modal decomposition is used. Multiple frequency response measurements are made on the components of fifteen manufactured compressor housings before and after the system assembly. Finite element models of the components and the system are validated using the validation power matrix at various validation powers. Validation using the validation power matrix improves modeling practices and confidence in the numerical models.

Degree

Ph.D.

Advisors

Bernhard, Purdue University.

Subject Area

Mechanical engineering

COinS