Method for correcting frequency response dynamic model of a few-sample structure based on transfer learning

Abstract

The invention discloses a method for correcting a frequency response dynamic model of a few-sample structure based on transfer learning. In order to improve the correction precision of a finite element model under a small amount of structural frequency response value data, vibration data in other fields are used as supplementary data, source domain data and target domain data are mapped to similarfeature spaces by using a domain adaptive method, and auxiliary training under a small amount of frequency response data is realized by using a transfer learning method. According to the method, theadvantages of feature analysis of transfer learning under a small number of samples are combined, and the deep convolutional neural network is utilized to perform feature analysis on sufficient vibration data in other fields. Furthermore, the feature distribution and solving tasks of the target domain data and the source domain data are different; the invention also adopts a domain self-adaptive method. The vibration features extracted by the deep convolutional neural network are mapped to a similar feature space; the method reduces the feature difference between different fields, finally achieves the sample expansion of a small amount of source domain data in the feature dimension, improves the accuracy of model correction, and prevents the overfitting caused by the insufficient data andthe reduction of the model correction precision caused by the simplified calculation of a conventional method.

Description

technical field [0001] The invention belongs to the field of structure dynamic frequency response dynamic model correction, relates to structure frequency response dynamic model correction and transfer learning theory under a small number of samples, and specifically relates to a small sample structure frequency response dynamic model correction method based on transfer learning. Background technique [0002] technical background: [0003] Among the dynamic model correction methods, the model correction method based on the frequency response function has been widely used in dynamic model correction in recent years. [1-3] . The invention uses the acceleration frequency response function as the basis for model correction to correct the structural dynamics model. Frequency Response Model Correction One of the major challenges faced by the model correction method in the process of algorithm implementation is the large demand for sample data. In order to reduce the computation...

AI Technical Summary

Problems solved by technology

[0005] The technical problem to be solved by the present invention is: to overcome the limitations of the traditional method for model correction under a small number of samples and the deficiency of the artificial extraction feature method, and to shake other fields The data is used for auxiliary learning, using the deep convolutional neural network and transfer learning principles to extract features from different vibration data in the source domain and the target domain and map their features into a similar feature space, so as to utilize existing knowledge in other fields. Enrich data and complete supplementary training under a small amount of data

Images