基于人工神经网络的高温合金蠕变曲线模拟

项目名称： 基于人工神经网络的高温合金蠕变曲线模拟

项目编号： No.50801060

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 无线电电子学、电信技术

项目作者： 侯介山

作者单位： 中国科学院金属研究所

项目金额： 20万元

中文摘要： 采用神经网络杂合模型来模拟和预测较宽温度和应力范围内合金的蠕变曲线。物理模型采用θ26041;程，由θ26041;程对蠕变曲线进行拟合，分出θ26041;程有效区和无效区。θ26041;程有效区内，方程中参数及截断应变点采用神经网络模型得出。θ26041;程无效区内，一方面，考察代表性蠕变试样组织演化情况，利用神经网络探索蠕变应变的本构方程，结合现有物理模型，对方程中各项物理意义进行解释。同时，利用该本构关系对 θ26041;程的弱化项作出修正。另一方面，不考虑具体本构关系，直接建立包括蠕变时间在内的输入参量和应变为输出参量的神经网络模型，实现对这一区域的模拟。由此建立具有明确物理意义的神经网络的杂合模型。移植和集成商用软件神经网络模块，完成模块的模拟功能。本项目的研究对于蠕变加速阶段蠕变行为和蠕变机理的理解具有重要的理论价值，同时，对高温材料的蠕变寿命和蠕变变形行为预测具有重要工程应用价值。

中文关键词： 高温合金；蠕变曲线；人工神经网络；本构关系；蠕变机理

英文摘要： Hybrid neural network model is developing to simulate and predict creep curves of the investigated superalloy in a wide temperature and stress range using. θquation is employed in this model, with which creep curves are fitted and it is divided into effective and ineffective region. In effective region, the parameters and intercept strain spot are determined with the neural network model. In ineffective region, with microstructural evolution of typical creep specimens investigated, the creep constitutive relations are developed with neural network, and physical signification of their parameters is clarified by combining with exiting physical models. On the other hand, without considering constitutive equations, a neural network model with creep time as one of input parameters and creep strain as output parameter can be directly developed to simulate creep curves at this region. So, a hybrid neural network model with definite physical signification is developed and simulation function of the model works after neural network module of commercial software is transplanted and integrated. This project is valuable not only on theoretical investigation of creep behavior and mechanisms in accelerating creep stage but also on engineering application of prediction of creep life and creep deformation behaviors of high temperature materials.

英文关键词： Superalloy; Creep curve; Artificial neural network; Constitutive relation; Creep mechanism