Convolutional Neural Network to Predict the Penetration Coefficient of Metamaterials Based on Their Structure and Composition

Abstract

The work is devoted to the development of a technique for predicting the coefficient passage of metamaterials based on topological structure and chemical composition with the use of machine learning techniques, namely artificial neural networks using convolution. In modern scientific and technical research, the methods of machine learning, namely: convolutional neural networks, occupy the most rapidly researched method in the design tasks of metamaterials and their properties. The advantage of this technique is the ease of implementation, the availability of data for this approach, the speed of calculations compared to the exact methods of predicting properties and topological structure based on physical laws. Of course, artificial neural networks are a rather complex process that has its own drawbacks – the need for a large amount of data, the relative complexity of optimization, and the complexity of problem formulation. With the development of machine learning technologies, these disadvantages are more and more eliminated, and therefore their use becomes more accessible. A large amount of information about metamaterials from relevant sources was used, namely the topology, chemical composition and measurement conditions of metamaterials. Software environments were used for writing digital code and building 3D objects of metamaterials with defined properties. An algorithm for predicting the transmission coefficient based on the structure, chemical composition of metamaterials based on a convolutional neural network using experimental data of laboratory metamaterials has been developed. An algorithm for saving information about the chemical composition of metamaterials has been developed. It is shown that using information about the electromagnetic properties of chemical elements, it is possible to predict the transmission coefficient of metamaterials. The process of presenting the coefficient of passage of metamaterials in a form convenient for training a convolutional neural network is described. Two methods were used to compare the effectiveness of both methods. It is shown that the method of representing experimental characteristics in the form of polynomial coefficients is faster, but not suitable for solving problems of predicting the characteristics of metamaterials. Data augmentation is shown to be the most effective method for improving forecasting results. Nevertheless, performance improvement methods based on architecture changes and hyperparameter changes should be continually evaluated and used whenever possible.