Prediction of Austenite Formation Temperatures Using Artificial Neural Networks
Prediction of Austenite Formation Temperatures Using Artificial Neural Networks
复制标题
使用人工神经网络预测奥氏体形成温度
DOI：
--
复制
发表时间：
2016
期刊：
影响因子：
--
通讯作者：
T. Lampke
中科院分区：
文献类型：
--
作者：
P. Schulze;E. Schmidl;T. Grund;T. Lampke
对于热处理的建模和设计，考虑到微观结构的发展/转变，根据化学成分、相应的微观结构/相和温度需要不同的材料数据。材料数据包括导热系数、热容、热膨胀和相变数据等。热模拟的质量在很大程度上取决于材料数据的准确性。对于许多材料，所需的数据-特别是对于不同的微观结构和温度-在文献中是罕见的。此外，无法预测所考虑的钢合金的允许范围内的不同化学成分。通过使用人工神经网络（ANN）计算材料数据，提供了解决这个问题的方法。在本研究中，使用人工神经网络计算亚共析钢从体心立方晶格结构转变为面心立方晶格结构的开始和结束温度。一个适当的数据库，包含不同的转变温度（奥氏体形成温度）来训练人工神经网络是从文献中选择的。为了找到一个合适的前馈网络，网络拓扑结构以及隐藏层的激活函数是不同的，随后在预测精度方面进行评估。通过人工神经网络计算的相变温度表现出很好的顺应性相比，实验数据。结果表明，预测性能甚至高于经典的经验方程，如安德鲁斯或布兰迪斯。因此，可以认为，提出的人工神经网络是一个方便的工具来区分bcc和fcc相在亚共析钢。
For the modeling and design of heat treatments, in consideration of the development/ transformation of the microstructure, different material data depending on the chemical composition, the respective microstructure/phases and the temperature are necessary. Material data are, e.g. the thermal conductivity, heat capacity, thermal expansion and transformation data etc. The quality of thermal simulations strongly depends on the accuracy of the material data. For many materials, the required data - in particular for different microstructures and temperatures - are rare in the literature. In addition, a different chemical composition within the permitted limits of the considered steel alloy cannot be predicted. A solution for this problem is provided by the calculation of material data using Artificial Neural Networks (ANN). In the present study, the start and finish temperatures of the transformation from the bcc lattice to the fcc lattice structure of hypoeutectoid steels are calculated using an Artificial Neural Network. An appropriate database containing different transformation temperatures (austenite formation temperatures) to train the ANN is selected from the literature. In order to find a suitable feedforward network, the network topologies as well as the activation functions of the hidden layers are varied and subsequently evaluated in terms of the prediction accuracy. The transformation temperatures calculated by the ANN exhibit a very good compliance compared to the experimental data. The results show that the prediction performance is even higher compared to classical empirical equations such as Andrews or Brandis. Therefore, it can be assumed that the presented ANN is a convenient tool to distinguish between bcc and fcc phases in hypoeutectoid steels.