针对变转速、变负载条件下的电机故障诊断问题,提出了一种基于自相关矩阵奇异值分解(Singular Value Decomposition,SVD)的特征提取和迁移学习分类器相结合的诊断方法。对于Hankel矩阵提取的奇异值向量,设计了平均曲率区分度指标来描述...针对变转速、变负载条件下的电机故障诊断问题,提出了一种基于自相关矩阵奇异值分解(Singular Value Decomposition,SVD)的特征提取和迁移学习分类器相结合的诊断方法。对于Hankel矩阵提取的奇异值向量,设计了平均曲率区分度指标来描述特征差异性,迁移学习TrAdaBoost算法通过迭代过程中调节辅助振动数据的权重来帮助目标数据学习,提升了分类正确率,同时利用向量夹角余弦进行可迁移度检测从而避免负迁移。试验结果表明,SVD无需利用故障先验知识,具有通用性,且迁移学习相比传统机器学习在目标振动数据较少条件下性能得到显著提升。展开更多
According to the aggregation method of experts＇ evaluation information in group decision-making,the existing methods of determining experts＇ weights based on cluster analysis take into account the expert＇s preferen...According to the aggregation method of experts＇ evaluation information in group decision-making,the existing methods of determining experts＇ weights based on cluster analysis take into account the expert＇s preferences and the consistency of expert＇s collating vectors,but they lack of the measure of information similarity.So it may occur that although the collating vector is similar to the group consensus,information uncertainty is great of a certain expert.However,it is clustered to a larger group and given a high weight.For this,a new aggregation method based on entropy and cluster analysis in group decision-making process is provided,in which the collating vectors are classified with information similarity coefficient,and the experts＇ weights are determined according to the result of classification,the entropy of collating vectors and the judgment matrix consistency.Finally,a numerical example shows that the method is feasible and effective.展开更多
文摘针对变转速、变负载条件下的电机故障诊断问题,提出了一种基于自相关矩阵奇异值分解(Singular Value Decomposition,SVD)的特征提取和迁移学习分类器相结合的诊断方法。对于Hankel矩阵提取的奇异值向量,设计了平均曲率区分度指标来描述特征差异性,迁移学习TrAdaBoost算法通过迭代过程中调节辅助振动数据的权重来帮助目标数据学习,提升了分类正确率,同时利用向量夹角余弦进行可迁移度检测从而避免负迁移。试验结果表明,SVD无需利用故障先验知识,具有通用性,且迁移学习相比传统机器学习在目标振动数据较少条件下性能得到显著提升。
文摘According to the aggregation method of experts＇ evaluation information in group decision-making,the existing methods of determining experts＇ weights based on cluster analysis take into account the expert＇s preferences and the consistency of expert＇s collating vectors,but they lack of the measure of information similarity.So it may occur that although the collating vector is similar to the group consensus,information uncertainty is great of a certain expert.However,it is clustered to a larger group and given a high weight.For this,a new aggregation method based on entropy and cluster analysis in group decision-making process is provided,in which the collating vectors are classified with information similarity coefficient,and the experts＇ weights are determined according to the result of classification,the entropy of collating vectors and the judgment matrix consistency.Finally,a numerical example shows that the method is feasible and effective.
