(Dr. Jerry Ginsberg, advisor)
"A Modified Approach to Improve the Robustness of the Algorithm of Mode Isolation"
Several problems limit the capabilities of current modal analysis techniques. Typically, modes with high damping ratio can be missed or poorly described. Modes whose natural frequencies are close require a high frequency resolution. Even then, existing methods generally fail if damping is sufficiently high, such that the modal bandwidths are comparable to the frequency separation, and the modes are said to be coupled. Modes with very low drive point mobility generally require multiple excitations and the presence of substantial noise further complicates the identification process. Furthermore, a fundamental requirement of most current techniques is the presence of an estimate of the number of modes in the system, or model order. Typically, the order of the model is overestimated then adjusted, and in doing so, false modes and instability may be encountered. Although several methods have been developed to evaluate the accuracy of the model order, current algorithms still face difficulties.
The Algorithm of Mode Isolation (AMI), referred to as the Mode Isolation Algorithm (MIA) in previous literature, is a frequency domain technique that was developed in 2000 and has proven to be superior to the current methods in addressing these difficulties. It provides a self-adjusting mechanism of identifying the order of the model, as well as higher robustness in the presence of noise. The algorithm was also extended to the state space to perform damped modal analysis. However, in its original form, the capabilities of the algorithm in identifying the normal modes of the system were limited. The present work uses synthetic data from the analytical solution of two relatively complicated systems to further assess AMI. The parameters of both systems are tuned such that they exhibit several of the problematic issues previously mentioned. The scope of the current work is twofold. First, to introduce a strategic change in the way the data is processed and analyzed, leading to a more robust algorithm that can identify all the elements of the normal mode matrix to a very good level of accuracy. Second, to assess the effect of noise on the accuracy of the results of the modified algorithm.