Author:
Bobzin K,Wietheger W,Johann L M
Abstract
Abstract
In the designing of iron-based metallic glasses the prediction of the glass transition temperature Tg, crystallization temperature Tx and liquidus temperature Tl is of special interest. The determination of these temperatures allows not only conclusions about the glass forming ability by calculating glass formation criterions but also about the thermal stability of the alloy in the amorphous state. In the course of alloy development, Tg, Tx and Tl are usually determined by means of differential scanning calorimetry (DSC) on an amorphous sample produced, for instance, by melt spinning or copper casting techniques. The test procedures are time-consuming and cost-intensive. In the present work Tg, Tx and Tl of iron-based metallic glasses are predicted. For that purpose, shallow neural networks with a varying number of neurons are trained using Bayes regularization. The data set for training and testing consists of corresponding literature data. The chemical compositions of iron-based metallic glasses are used as inputs and Tg, Tx and Tl are the outputs. A threshold method is used for data balancing and limiting the number of inputs. Low Root Mean Squared Error (RMSE) and correspondingly high prediction accuracies were achieved during the testing.
Cited by
1 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献