Prediction Of Amorphous Forming Ability Based On Machine Learnin

Amorphous alloys have attracted more and more focused attention from researchers because of their unique physical and mechanical properties.However,one of the biggest challenges faced in the research and development of amorphous alloys is the problem of amorphous forming ability.The generally poor amorphous forming ability of the most amorphous alloys compared to many commercially available oxide glasses can lead to greater difficulties in the development of new amorphous alloys,to the extent that their large-scale application is largely limited.However,the amorphous forming ability of amorphous alloys needs to be obtained experimentally by tedious and expensive X-ray diffraction tests or thermal analysis calculations,which is a rather complicated,time-consuming and consumable process and wastes a lot of manpower and resources in the process.Therefore,how to design a method to accurately obtain the amorphous forming ability of amorphous alloys quickly,efficiently and inexpensively has become a topic of great interest to researchers.Subsequently,some researchers have also used various methods to investigate the amorphous forming ability of amorphous alloys.However,these methods are only for special amorphous alloys,and the prediction accuracy of these methods is not high enough to accurately identify new amorphous alloys.Therefore,there is an urgent need to devise an alternative method to predict the amorphous forming ability of amorphous alloys.In recent years,the rapid development of machine learning,with its powerful data analysis capability,has led to its wide application in various aspects of research and good results.Based on this paper,machine learning is also used to study the problem of amorphous forming ability of amorphous alloys.It is concluded from the study that the prediction accuracy obtained by the models developed in this paper are all more than 19% higher than those obtained by other previous models.And the more the input features,the higher the prediction accuracy obtained by the model its gets 87%.The results illustrate that all three models developed in this paper can more accurately predict the amorphous forming ability of a variety of amorphous alloys and provide theoretical guidance for the development and preparation of amorphous alloys.The main research of the thesis is as follows.(1)The knowledge of artificial neural network structure,the activation function and optimizer selection are introduced,and the principle of artificial neural network training process is derived.The artificial neural network structure is also explored and improved according to the characteristics after analyzing the dataset,and a prediction model applicable to amorphous forming ability is established.(2)The knowledge and characteristics of convolutional neural networks are introduced,the influence of relevant parameters in convolutional neural networks on the prediction performance is investigated,a prediction model for the amorphous forming ability of various amorphous alloys is established,and the predicted values of this model are compared with the results of 13 prediction standards for amorphous forming ability.(3)The structural features of the LightGBM algorithm are introduced in depth,and its training procedure is derived.A grid search is used to obtain the optimal values of the LightGBM algorithm parameters as a way to build a prediction model applicable to the amorphous forming ability.Also,the prediction results of the LightGBM algorithm model is compared with those of the multilayer artificial neural network and convolutional neural network models.