基于数据变化率和重构贡献图的微小故障诊断方法

针对复杂工业过程的微小故障诊断问题，提出一种数据预处理与重构贡献图相结合的故障诊断方法；为了克服非高斯分布数据对故障检测准确性的影响，通过基于数据变化率的方法对样本原始数据进行预处理后，可以有效地检测过程变量的微小故障，以此建立故障诊断主元分析模型；检测出系统故障后，为了提高故障辨识准确度，采用一种平均残差差值重构贡献图的方法对故障进行辨识；通过正常样本数据和故障数据在残差子空间中的投影，获取两个数值的残差差值向量，计算重构贡献值来确定故障变量；以田纳西-伊斯曼(TE)过程为对象进行了故障诊断仿真实验，并与传统贡献图和重构贡献图方法的辨识准确率相比较，结果表明所提方法具有良好的故障诊断性能。

Incipient Fault Diagnosis Using Data Change Rate and Reconstruction-based Contribution Plot

For the incipient fault diagnosis in complex industrial processes, a fault diagnosis method is proposed based on data preprocessing and reconstructed contribution plot. In order to overcome the influence of non-Gaussian distribution data on the accuracy of fault detection, the original sample data is preprocessed based on the data change rate, and the principal component analysis model for fault diagnosis is established. In order to improve the accuracy of fault identification, an average residual difference reconstructed contribution plot is used to identify the fault. Through the projection of normal sample data and fault data in the residual subspace, the residual difference vector is obtained and the reconstruction contribution value is calculated. The simulation results of fault diagnosis on Tennessee-Eastman (TE) process show that the proposed method has good diagnostic performance.