Affiliations: College of Computer Science and Technology, HuaQiao University, Xiamen, Fujian, China
Correspondence:
[*]
Corresponding author: Cheng Wang, College of Computer Science and Technology, HuaQiao University, No. 668, Jimei Road, Xiamen 361021, Fujian, China. E-mail:wangcheng@hqu.edu.cn
Abstract: Due to ill-conditioned inverse characteristics of linear
time invariant dynamic system uncorrelated multi-sources load identification
in frequency domain, it have large condition number and errors for classic
least-squares of generalization method. In order to overcome these
shortcomings, an improved Tikhonov regularization method is put forward in
this paper. This method uses minimization maximum relative error of
identification multi-sources load as criterion to determine the optimal
regularization parameter. At the same time, combination of improved Tikhonov
regularization and least square of generalized matrix inverse method can
reduce the time overhead of determining the best value of regularization
parameter. Only when condition number in a frequency is larger than a
threshold value, must the improved Tikhonov regularization is used.
Uncorrelated multi-sources vibration load identification in frequency domain
results on cylindrical shell simulation datasets shows that this new method
works much better than classic least-squares of generalized matrix inverse
when measurement noise exists and could basically meet the engineering
precision requirement of ± 3 dB.
Keywords: Uncorrelated multi-sources load identification, frequency domain, Tikhonov regularization, least-square of generalized matrix inverse, minimization maximum relative error
