The Study Of Modal Parameter Identification Method Based On Neural Network

With the development of modern industry, structural dynamic characteristics of mechanical systems are increasingly demanding, therefore the structural dynamic analysis is becoming more and more important. Modal parameter identification is an important method for structural dynamic analysis with the purpose to provide theoretical basis for system dynamic response, fault diagnosis and prediction, structural dynamic modification and optimization of design parameters.An in-depth study on modal parameter identification methods has been carried out in this paper. A neutral network-based modal parameter identification method has been introduced to solve the anti-noise interference, noise-caused false modal identification and proposal, and model-order issues in the present time-domain method. Based on the free response data or impulse response data, this method has achieved structural response prediction through establishment of neural network model. Furthermore, modal parameters can be acquired by constructing characteristic equation using neural network weights matrix. In order to verify the correctness and superiority of this algorithm, both numerical simulation and experimental application have been carried out.Numerical simulation study:1) established structural dynamics model by Matlab/Simulink software, identified modal parameters using neural network approach with sinusoidal excitation, pulse excitation and excitation-free response respectively. High recognition accuracy has been proved through comparison between recognition results and theoretical values;2) identified modal parameters of response signal with varying signal-to-noise ratio by Ibrahim Time-Domain method (ITD), the Least Complex Exponential Squares method (LSCE) and proposed Linear Neural Network method (LNN), respectively. The simulation results show LNN method has superior anti-noise interference ability. These numerical simulation examples have proved feasibility and superiority of applying neural network method to modal parameter identification.Experimental study:1) built a set of modal test and analysis system based on the demand analysis, modal analysis theory, virtual instrument technology and information technology. The system integrates the Least Square Complex Exponential method and the Linear Neural Network method, with the function of real-time test data acquisition, display, storage, reading-back and modal analysis, etc. It can present the whole process of hammer exciting including structural model inputs and animated output. The system interference is user-friendly, cost-effective and easy to expand;2) tested and verified the proposed algorithm, hardware and software of system through cantilever beam and simply-supported beam experiments. Experimental results show the algorithms correct and effective, the system stable and reliable. High practical value is expected.