基于KPCA和MKL-SVM的非线性过程监控与故障诊断

Nonlinear process monitoring and fault diagnosis based on KPCA and MKL-SVM

利用核主元分析非线性过程监控的优势,结合多重核学习支持向量机在故障诊断方面的准确性,提出了基于核主元分析和多重核学习支持向量机的非线性过程监控与故障诊断方法。该方法运用核主元法对数据进行处理,在特征空间构建T2和SPE来检测故障的发生,若有故障发生,则计算样本的非线性主元得分向量,将其作为MKL-SVM的输入值,通过MKL-SVM的分类进行故障类型识别。将上述方法应用到Tennessee Eastman(TE)化工过程,多种故障模式的仿真结果表明该方法不但能有效地辨识故障,而且提高了故障检测和故障诊断的速度。

Taking the advantage of KPCA for nonlinear process monitoring and the accuracy of multiple kernel learning support vector machines （MKL-SVM） for fault diagnosis, a new method for nonlinear process monitoring and fault diagnosis based on kernel principal analysis and multiple kernel learning support vector machines is proposed. First, the data is analyzed using KPCA. T2and SPE are constructed in the feature space for detecting the fault occurrence. If T2 and SPE exceed the predefined control limits, a fault may have occurred. Then the nonlinear score vectors are calculated and fed into MKL-SVM to identify the faults through MKL-SVM classification. To demonstrate the performance, the proposed method was applied to the Tennessee Eastman （TE） chemical industry process. Simulation results of multiple fault modes show that the proposed method could not only effectively identify various types of fault sources, but also improve the speed of fault detection and diagnosis.