Research On Dynamic Load Identification Method And Application Based On Regularization

In practical engineer problems, external dynamic load information plays important roles in many areas, such as structural dynamic analysis, health monitoring, parameter identification, strength check. In many cases, due to economic cost or complex environments, the dynamic load acting on engineer structure is very difficult or even impossible to be measured directly. However, structure responses are often relatively easy to obtain, so identifying dynamic load based on structure response and structure feature is becoming important research contents. Therefore, the dynamic load identification method is put forward to be applied to dynamic load identification problem. Dynamic load identification method is widely used in the load identification problems, among which the ill-conditioned problem of dynamic load identification problem becomes the focus. Regularization method can overcome the ill-conditioned problem of dynamic load identification problem, so dynamic load identification problem based on regularization method is applied in general circumstance. This paper will concentrate on applicability research for dynamic load identification method based on regularization method in the time domain. The current work includes the following specific aspects:Firstly, the state-of-the-art of dynamic load identification method is reviewed. Based on the discussions of current dynamic load identification methods, one finds that there still have some limitations in the current dynamic load identification methods. So the research thesis is the dynamic load identification method research based on regularization. The highlights of research includes three points:(1) the establishment of dynamic load identification system model,(2) the selection of optimal regularization parameter,(3) the establishment of noise immunity for system model method and the regularization method which is adapted for non-Gaussian noise.Secondly, an improved least square fitting shape function method which is used for identification of structure external load is proposed. The method reduces reasonable assumption at the extreme, and provides more accurate structure mode of optimal approximate load and shape function response matrix. The accuracy of the results of load identification is improved effectively. Moreover, the quotient function method for regularization parameter determination is proposed. The quotient function which regard regularization parameter as independent variable is defined by the least squares solution of optimal problem in the context of Tikhonov regularization method. Based on quadratic programming theory, one can get quotient function values corresponding to different regularization parameters. Then according to the characteristics of different quotient function values, the determination of the optimal regularization parameter of Tikhonov regularization method can be determined effectively. Quotient function method can overcome the limitations of common used GCV(Generalized Cross-Validation) method and L-curve method, and have stability for measured noise and model errors.Thirdly, the structural response function which is obtained by the measurement will inevitably be affected by noise, thus a weighted variable limit integral moving average method is proposed for establishing the system model. Weight function is constructed by moving average combination coefficients. And the weighted variable limit integral moving average function model of response function of the structure is constructed based on the integral moving average method. Weighted integral moving average has a filtering effect on the noise, so the weighted variable limit integral moving average function model is the optimal approximate response function one of the real response function in the least-squares sense. Further, by increasing the times of integral moving average, one can achieve a better inhibition effect on the noise. Thus a better approximate response function model of the real response function can be obtained. However, too many times of the integral moving average will cause error accumulation. The paper presents the suggested times of integral moving average through in-depth analysis. Weighted variable limit integral moving average function method has strong noise resistance, in the presence of high level measured noise it can get load identification results with high accuracy and very good smooth property.Finally, the L∞ norm fitting regularization method is introduced. Research on load identification problem with two kinds of typical non-Gaussian noise is performed. In the optimization problem derived from L∞ norm fitting regularization method, the monotonicity validation function is constructed by using the monotony of the regularization and fitting term on regularization parameter, and the monotonic validation method to choose the optimal regularization parameter is proposed. High levels of Gaussian white noise after appropriate times of integral moving average remains non-Gaussian noise with smaller amplitude. Compared with the traditional L2 norm regularization method(Tikhonov regularization method), L∞ norm fitting regularization method for such non-Gaussian noise can get load identification results with better smoothing property and higher precision. In addition, the noise in the telemetry data is mainly caused by the low resolution of the data acquisition system. The system noise is a kind of non-Gaussian noise with high level and approximately satisfies the uniform distribution. Compared with the L2 norm regularization method, load identification results indicates that L∞ norm fitting regularization method is more suitable for telemetry data. The monotonic validation method can effectively determine the optimal regularization parameter for L∞ norm fitting regularization method.