摘要：
现阶段由于智能装备的结构和功能不断完善,机械故障预兆和故障特征也不断复杂化,导致了故障诊断难度大大增加.由于机器学习和数据挖掘技术的不断革新,基于数据挖掘的故障诊断系统快速发展,提高了故障诊断效率,减少了因诊断延迟造成的损失.对此,提出一种基于辨识约简矩阵的决策树故障诊断方法,实现了故障样本决策表的高效生成并保证诊断的正确性.首先采用粗糙集的决策树方法建立故障诊断决策表,然后离散化处理特征数据;接着采用可辨别矩阵约简算法进行属性约简,删除冗余信息,形成精简的决策表;最后使用C4.5算法构造出最终决策树,并用该方法与直接使用C4.5算法所生成决策树进行对比分析.实验结果表明,该方案有一定的容错能力,并且是一种快速、可靠的故障诊断方法.%Due to the continuous improvement of the structure and function of intelligent equipment at present,the mechanical failure warning and fault features are also becoming increasingly complicated,resulting in a great increase in fault diagnosis.With the continuous innovation of machine learning and data mining technology,the fault diagnosis system based on data mining is developing rapidly,which improves the efficiency of fault diagnosis and reduces the loss caused by the delay of diagnosis.For this,we put forward a decision tree fault diagnosis scheme based on identification reduction matrix,which can effectively extract the fault samples and ensure the correctness of the diagnosis. Firstly,the decision tree method based on rough set is used to establish the fault diagnosis decision table,and then the characteristic data are discretized.Then the algorithm of the discernible matrix reduction is used for attribute reduction,deleting redundant information,and forming a simplified decision table.Finally,the final decision tree is constructed by C4.5 algorithm and is compared with the decision tree generated by C4.5 algorithm.The experiments show that the scheme,with a certain fault tolerance,is a fast and reliable fault diagnosis method.
