Bayesian probabilistic prediction of precipitation behavior in Ni-Ti shape memory alloys
Bayesian probabilistic prediction of precipitation behavior in Ni-Ti shape memory alloys
复制标题
DOI：
10.1016/j.commatsci.2019.109334
复制
发表时间：
2020-02
影响因子：
3.3
通讯作者：
P. Honarmandi;L. Johnson;R. Arróyave
中科院分区：
文献类型：
--
作者：
P. Honarmandi;L. Johnson;R. Arróyave
镍钛合金是在不同工业应用中最受欢迎的形状记忆合金，因为它们的特殊性能，至少在某些变体中，是由二次强化相的沉淀提供的。因此，沉淀建模和相关的模型校准以及对其性能的不确定性估计似乎对这些合金至关重要，特别是当重点放在设计导致最佳性能的沉淀方案时。在这项工作中，使用基于马尔可夫链Monte Carlo-Metropolis Hastings算法的贝叶斯方法，对个体和所有可用的实验数据进行了有影响的模型参数（即对模型响应有强烈影响的参数）的校准和不确定度量化。然后，通过正演模型分析，将所得参数的不确定性传递到模型结果的不确定性中。根据模型标定得到的界面能值，分别用给定的实验数据，建立了基体/析出物界面能与时效温度和标称成分之间的经验关系。利用该模型中的界面能方程，将所有实验数据一起用于标定模型中的其他影响参数。在对模型进行概率校正后，模型结果与实验数据的差异是采用co-Kriging代理模型的主要原因，该模型利用基于误差相关性的模型结果与实验数据的融合来更精确地预测该系统的降水行为。
Ni-Ti alloys are the most popular shape memory alloys in different industrial applications due to their especial properties provided, for at least some variants, by the precipitation of secondary strengthening phases. Therefore, precipitation modeling and associated model calibration coupled with uncertainty estimations of their performance seem to be crucial for these alloys, particularly when the focus is on the design of precipitation schemes leading to optimal performance. In this work, the calibration and uncertainty quantification of influential model parameters (i.e. the parameters with strong effects on the response of the model) have been performed against individual and all available experimental data using a Bayesian approach based on the Markov chain Monte Carlo-Metropolis Hastings algorithm. Then, the resulting parameters’ uncertainties have been propagated to the uncertainty in model results through a forward model analysis. An empirical relationship for matrix/precipitate interfacial energy in terms of aging temperature and nominal composition has been introduced according to the values of interfacial energy obtained from the model calibration with each given experimental data individually. Using this equation for interfacial energy in the model, all experimental data has been used together to calibrate the other influential parameters in the model. After the probabilistic calibration of the model, the discrepancies between the model outcomes and the experimental data were the main reasons to apply co-Kriging surrogate modeling that takes advantage of an error correlation-based fusion of the model results and the experimental data for more precise prediction of precipitation behavior in this system.