Experimental study is performed on the probabilistic models for the long fatigue crack growth rates (da/dN) of LZ50 axle steel. An equation for crack growth rate was derived to consider the trend of stress intensity f...Experimental study is performed on the probabilistic models for the long fatigue crack growth rates (da/dN) of LZ50 axle steel. An equation for crack growth rate was derived to consider the trend of stress intensity factor range going down to the threshold and the average stress effect. The probabilistic models were presented on the equation. They consist of the probabilistic da/dN-ΔK relations, the confidence-based da/dN-ΔK relations, and the probabilistic- and confidence-based da/dN-ΔK relations. Efforts were made respectively to characterize the effects of probabilistic assessments due to the scattering regularity of test data, the number of sampling, and both of them. These relations can provide wide selections for practice. Analysis on the test data of LZ50 steel indicates that the present models are available and feasible.展开更多
In order to describe and control the stress distribution and total deformation of bladed disk assemblies used in the aeroengine, a highly efficient and precise method of probabilistic analysis which is called extremum...In order to describe and control the stress distribution and total deformation of bladed disk assemblies used in the aeroengine, a highly efficient and precise method of probabilistic analysis which is called extremum response surface method(ERSM) is produced based on the previous deterministic analysis results with the finite element model(FEM). In this work, many key nonlinear factors, such as the dynamic feature of the temperature load, the centrifugal force and the boundary conditions, are taken into consideration for the model. The changing patterns with time of bladed disk assemblies about stress distribution and total deformation are obtained during the deterministic analysis, and at the same time, the largest deformation and stress nodes of bladed disk assemblies are found and taken as input target of probabilistic analysis in a scientific and reasonable way. Not only their reliability, historical sample, extreme response surface(ERS) and the cumulative probability distribution function but also their sensitivity and effect probability are obtained. Main factors affecting stress distribution and total deformation of bladed disk assemblies are investigated through the sensitivity analysis of the model. Finally, compared with the response surface method(RSM) and the Monte Carlo simulation(MCS), the results show that this new approach is effective.展开更多
社交网络中,节点间存在多种关系类型,节点数量会随着时间的推移而变化,这种异质性和动态性给链路预测任务带来极大的挑战。因此,本文提出一种基于增量学习的社交网络链路预测方法(incremental learning social networks link prediction...社交网络中,节点间存在多种关系类型,节点数量会随着时间的推移而变化,这种异质性和动态性给链路预测任务带来极大的挑战。因此,本文提出一种基于增量学习的社交网络链路预测方法(incremental learning social networks link prediction,IL-SNLP)。通过对网络进行分层,使每一层网络只包含一种关系类型,以更好地获取节点在每种关系类型下的语义信息;针对网络的动态性,利用时序随机游走捕获社交网络中的局部结构信息和时序信息;针对增量数据,采用增量式更新随机游走策略对历史随机游走序列进行更新。通过增量式skip-gram模型从随机游走序列中提取新出现节点的特征,并进一步更新历史节点的特征;针对网络的异质性,采用概率模型提取不同关系类型之间的因果关系关联程度,并将其作用于每一层的节点特征,以改善不同关系层下节点特征表现能力;利用多层感知机构建节点相互感知器,挖掘节点间建立连接时的相互贡献,实现更高的链路预测准确率。实验结果表明,在3个真实的社交网络数据集上,IL-SNLP方法的ROC曲线下的面积(AUC)和F1分数比基线方法分别提高了10.08%~67.60%和1.76%~64.67%,提升了预测性能;对于增量数据,只需要少次迭代就能保持预测模型的性能,提高了模型训练的速度;与未采用增量学习技术的IL-SNLP−方法相比,IL-SNLP方法在时间效率上提升了30.78%~257.58%,显著缩短了模型的运行时长。展开更多
基金Project supported by the National Natural Science Foundation of China (Nos.50375130and50323003), the Special Foundation of National Excellent Ph.D.Thesis (No.200234) and thePlanned Itemforthe Outstanding Young Teachers ofMinistry ofEducationofChina (No.2101)
文摘Experimental study is performed on the probabilistic models for the long fatigue crack growth rates (da/dN) of LZ50 axle steel. An equation for crack growth rate was derived to consider the trend of stress intensity factor range going down to the threshold and the average stress effect. The probabilistic models were presented on the equation. They consist of the probabilistic da/dN-ΔK relations, the confidence-based da/dN-ΔK relations, and the probabilistic- and confidence-based da/dN-ΔK relations. Efforts were made respectively to characterize the effects of probabilistic assessments due to the scattering regularity of test data, the number of sampling, and both of them. These relations can provide wide selections for practice. Analysis on the test data of LZ50 steel indicates that the present models are available and feasible.
基金Projects(51375032,51175017,51245027)supported by the National Natural Science Foundation of China
文摘In order to describe and control the stress distribution and total deformation of bladed disk assemblies used in the aeroengine, a highly efficient and precise method of probabilistic analysis which is called extremum response surface method(ERSM) is produced based on the previous deterministic analysis results with the finite element model(FEM). In this work, many key nonlinear factors, such as the dynamic feature of the temperature load, the centrifugal force and the boundary conditions, are taken into consideration for the model. The changing patterns with time of bladed disk assemblies about stress distribution and total deformation are obtained during the deterministic analysis, and at the same time, the largest deformation and stress nodes of bladed disk assemblies are found and taken as input target of probabilistic analysis in a scientific and reasonable way. Not only their reliability, historical sample, extreme response surface(ERS) and the cumulative probability distribution function but also their sensitivity and effect probability are obtained. Main factors affecting stress distribution and total deformation of bladed disk assemblies are investigated through the sensitivity analysis of the model. Finally, compared with the response surface method(RSM) and the Monte Carlo simulation(MCS), the results show that this new approach is effective.
文摘社交网络中,节点间存在多种关系类型,节点数量会随着时间的推移而变化,这种异质性和动态性给链路预测任务带来极大的挑战。因此,本文提出一种基于增量学习的社交网络链路预测方法(incremental learning social networks link prediction,IL-SNLP)。通过对网络进行分层,使每一层网络只包含一种关系类型,以更好地获取节点在每种关系类型下的语义信息;针对网络的动态性,利用时序随机游走捕获社交网络中的局部结构信息和时序信息;针对增量数据,采用增量式更新随机游走策略对历史随机游走序列进行更新。通过增量式skip-gram模型从随机游走序列中提取新出现节点的特征,并进一步更新历史节点的特征;针对网络的异质性,采用概率模型提取不同关系类型之间的因果关系关联程度,并将其作用于每一层的节点特征,以改善不同关系层下节点特征表现能力;利用多层感知机构建节点相互感知器,挖掘节点间建立连接时的相互贡献,实现更高的链路预测准确率。实验结果表明,在3个真实的社交网络数据集上,IL-SNLP方法的ROC曲线下的面积(AUC)和F1分数比基线方法分别提高了10.08%~67.60%和1.76%~64.67%,提升了预测性能;对于增量数据,只需要少次迭代就能保持预测模型的性能,提高了模型训练的速度;与未采用增量学习技术的IL-SNLP−方法相比,IL-SNLP方法在时间效率上提升了30.78%~257.58%,显著缩短了模型的运行时长。
