基于加权统计局部核主元分析的非线性化工过程微小故障诊断方法

传统统计局部核主元分析（statistical local kernel principal component analysis, SLKPCA）在构造改进残差时未考虑样本的差异性，使得故障样本信息易于被其他样本所掩盖，针对该问题，提出一种基于加权统计局部核主元分析（weighted statistical local kernel principal component analysis, WSLKPCA）的非线性化工过程微小故障诊断方法。该方法首先利用KPCA获取过程的得分向量和特征值并构建初始残差。然后设计了一种基于测试样本与训练样本之间距离的加权策略构建加权改进残差，对含有较强微小故障信息的样本赋予较大权值，以增强故障样本的影响。最后，采用基于测量变量与监控统计量之间的加权互信息构建贡献图以识别故障源变量。在连续搅拌反应釜和田纳西伊斯曼（Tennessee Eastman, TE）化工过程上的仿真结果表明,所提方法具有良好的微小故障检测与识别性能。

Incipient fault diagnosis method of nonlinear chemical process based on weighted statistical local KPCA

The traditional local kernel principal component analysis (SLKPCA) does not consider the difference of samples when constructing the improved residual, so that the fault sample information is easily covered by other samples. This paper proposes a new fault diagnosis method of nonlinear chemical process based on weighted statistical local kernel principal component analysis (WSLKPCA). Firstly, the score vectors and the eigenvalues are obtained using KPCA and the residual function is constructed. Then, a weighting strategy based on the distance between the test sample and the training sample is designed to construct the weighting improved residual, which assigns larger weights to samples with strong incipient fault information to enhance the impact of fault samples. Finally, the contribution graph is constructed based on the weighted mutual information between the measured variables and monitoring statistics to identify the fault source variables. Simulation results on continuous stirred tank reactor and TE process show that the proposed method can effectively detect incipient faults, and has better fault recognition performance.