There exists some discord or contradiction of information during the process of fault diagnosis for rotary machine. But the traditional methods used in fault diagnosis can not dispose of the information. A model of fa...There exists some discord or contradiction of information during the process of fault diagnosis for rotary machine. But the traditional methods used in fault diagnosis can not dispose of the information. A model of fault diagnosis for a rotary machine based on information entropy theory and rough set theory is presented in this paper. The model has clear mathematical definition and can dispose both complete unification information and complete inconsistent information of vibration faults. By using the model, decision rules of six typical vibration faults of a steam turbine and electric generating set are deduced from experiment samples. Finally, the decision rules are validated by selected samples and good identification results are acquired.展开更多
针对CART(classification and regression tree)分类决策树构建过程中由于小样本集特征维数高及噪声等造成的过拟合问题,在CART决策树算法训练过程中引入基于互信息的粗糙集(rough sets,RS)属性约简,考虑信息熵与基尼(GINI)系数刻画样本...针对CART(classification and regression tree)分类决策树构建过程中由于小样本集特征维数高及噪声等造成的过拟合问题,在CART决策树算法训练过程中引入基于互信息的粗糙集(rough sets,RS)属性约简,考虑信息熵与基尼(GINI)系数刻画样本集"纯净度"的相似关系,对历史故障数据进行属性约简,降低属性维度以优化训练集,在此基础上构建分类决策树,可视化输出规则。实验结果表明:将改进的CART决策树算法应用于某型航空发动机油液故障诊断,提取的规则可解释性强,能够减小冗余属性及噪声对决策的影响,与常用故障诊断算法相比,该模型的诊断准确率提升20%左右,AUC(area under curve)值高达92%,可以有效处理高维离散型航空发动机小样本故障问题。展开更多
基金The paper is supported by the 863 Program of China under Grant No 2006AA04A110
文摘There exists some discord or contradiction of information during the process of fault diagnosis for rotary machine. But the traditional methods used in fault diagnosis can not dispose of the information. A model of fault diagnosis for a rotary machine based on information entropy theory and rough set theory is presented in this paper. The model has clear mathematical definition and can dispose both complete unification information and complete inconsistent information of vibration faults. By using the model, decision rules of six typical vibration faults of a steam turbine and electric generating set are deduced from experiment samples. Finally, the decision rules are validated by selected samples and good identification results are acquired.
文摘针对CART(classification and regression tree)分类决策树构建过程中由于小样本集特征维数高及噪声等造成的过拟合问题,在CART决策树算法训练过程中引入基于互信息的粗糙集(rough sets,RS)属性约简,考虑信息熵与基尼(GINI)系数刻画样本集"纯净度"的相似关系,对历史故障数据进行属性约简,降低属性维度以优化训练集,在此基础上构建分类决策树,可视化输出规则。实验结果表明:将改进的CART决策树算法应用于某型航空发动机油液故障诊断,提取的规则可解释性强,能够减小冗余属性及噪声对决策的影响,与常用故障诊断算法相比,该模型的诊断准确率提升20%左右,AUC(area under curve)值高达92%,可以有效处理高维离散型航空发动机小样本故障问题。
