Parameter Identification Of Composite Materials Based On Deep Learning

Because of its excellent mechanical and physical properties,composite materials have been widely used in engineering field.It is very important to analyze the mechanical properties of composite structures in structural design.In order to obtain the accurate structure model of composite materials,the first task is to identify the parameters of composite materials.Based on the modal data of the structure,the parameters of the composite laminates were identified,and nine parameters of the laminates were identified.A parameter identification method based on LSTM and GRU neural network was proposed to avoid the local optimal problem in the sensitivity method.A composite material parameter identification method based on small sample learning is also proposed to realize a small number of samples to train the model to improve the computational efficiency,and to solve the overfitting problems caused by the small number of training sample data in the deep learning method.The main work of this paper is as follows:In order to solve the modeling problem of composite laminates due to the difficulty in determining many parameters,a sensitivity based method of composite parameters identification was studied.Taking composite laminates as the research object,four modal experiments were carried out on composite laminates by hammer method,and the finite element model of composite structures was modified.The nine parameters with high structural sensitivity were modified based on four sets of modal frequency data.The correction results show that the correction parameters and the final modal frequency of the four groups of data converge,and the parameter changes are within the set range.The feasibility of using the model correction method based on modal frequency to modify multiple parameters of composite laminates provides a more accurate finite element model for the study of composite laminates.In order to improve the accuracy and efficiency of parameter identification,a novel parameter identification method for composite materials based on deep learning was proposed.The inverse problem analysis of traditional design parameter model modification method is transformed into positive problem analysis.With structural mode as input and frequency and structural parameters as output,parameter identification of simple composite plate is carried out by LSTM neural network,which verifies the feasibility of parameter identification method based on neural network.Then LSTM and GRU neural networks were used to identify the parameters of the composite laminates.The results show that the modal frequency error of the neural network recognition method is smaller than that of the traditional recognition method.The variation range of recognition parameters based on neural network is also much smaller than that of traditional recognition methods,which proves that the recognition accuracy of this method is higher than that of traditional methods.The method of using neural network to identify parameters of composite structures not only avoids the problem of traditional parameter identification methods falling into the local optimal solution during correction,but also greatly improves the accuracy of parameter identification structure of composite materials,which lays a foundation for the subsequent research on mechanical properties of composite structures.Aiming at various problems that may be encountered in the parameter identification method of composite lamination structure based on neural network.For example,limited by test conditions and environment,it is impossible to obtain enough test data to train the network,resulting in insufficient accuracy of the network and so on.In this paper,the parameter identification of composite materials based on few samples is proposed.By using the thinking method of small sample learning,the machine learning method to solve practical problems is obtained by learning from a small number of sample data to realize the parameter identification of structure.It uses VAE algorithm for data expansion to optimize the small sample learning model,uses Bayesian estimation to estimate the original data distribution,and introduces KL divergence to measure the difference between the two distributions to improve the accuracy of the extended data.