Noise and vibration engineers make many frequency domain measurements during the development of an automobile. These measurements are used to develop prototype hardware or, in some cases, to increase the “degree of belief” of a computer model. In the case of hardware development, the engineer frequently must evaluate competing designs while, in the case of computer modeling, the engineer often must investigate the fidelity of his/her assumptions. In either instance, the engineer will perform some type of experiment to answer the question(s) of interest.
Many experiments, however, may be compromised by undesired variability associated with the data. This variability may arise from sampling uncertainty, transducer noise and digital signal processing bias to list a few. While most engineers consider this variability, they often do not account for other sources of variability such as experimental set-up, car to car differences and environmental conditions.
This paper's aim is to illustrate how to improve system understanding through the analysis of frequency domain data from designed experiments. A brief overview of experiment design and a method to estimate spectral design factor effects are presented. A technique to estimate experimental “error” from unreplicated experiments is also presented as well as a simple graphical display that facilitates easy data interpretation. The methods, techniques and graphical displays are presented using vehicle seat track autospectrum vibration measurements from a designed experiment.