【摘要】 针对传统的基于贡献图的多元统计方法中变量贡献值不能真实反映故障原因的不足,提出一种基于符号有向图(SDG)模型和数据重构的故障诊断方法,该方法使用平方预测误差(SPE)和累积和(CUSUM)统计量进行故障检测,通过在故障发生时对SDG的所有相容路径方向的样本数据进行重构,重构后残差变化最大的方向被认为是故障的传播方向,该方向上的起始节点为导致故障的原因变量。通过在TEP模型上仿真表明:该方法能有效地诊断出引起故障的根本原因。更多还原

【Abstract】 Considering the fact that in traditional contribution plots-based multivariate statistics,variable contribution value has insufficiency in reflecting the failure cause,a new fault diagnosis method based on signed directed graph( SDG) and data reconstruction was proposed to have square prediction error( SPE) and cumulative sum( CUSUM) used to detect a fault; when a fault happens,all sample data on SDG’s consistent branches can be reconstructed; and the branches where maximal residual change exists after to be reconstructed is the fault propagation branch,the initial node is thought to be the causal variable which resulting in the fault. Simulation in TEP process shows that this method can find the fault cause accurately.更多还原