基于MAGMM-KECA的间歇过程故障诊断方法

Fault diagnosis method for batch process based on MAGMM-KECA

对具有多阶段操作特性的间歇过程进行建模时,往往忽略了不同阶段间的差异性,没有考虑阶段间的相关特性和过渡特性,影响了过程监控的精确性.针对此问题,提出了多方向自适应高斯混合模型-核熵成分分析(multiple adaptive Gaussian mixture model-Kernel entropy component analysis,MAGMM-KECA)算法,该算法无需先验知识,在自动获取不同阶段高斯模型信息的同时对多阶段进行自适应柔性划分,得到更加精确的高斯模型,解决阶段间的相关特性和过渡特性;再利用KECA处理非线性、高维数据的优越性,分别对不同阶段进行建模,引入CS(Cauchy-Schwarz)统计量对故障进行监控,检测到故障后用贡献图方法诊断出故障变量.通过青霉素发酵过程进行了验证,结果表明所提算法比MPCA、MKPCA算法具有更好的故障诊断精度.

When batch process with multiple operation phases characteristic is modeled, the difference be tween different phases is often neglected, and the interstage correlation and transition characteristics are not taken into account, so that the accuracy of the process monitoring is affected. Aimed at this problem, a multiple adaptive Gaussian mixture model kernel entropy component analysis （MAGMM KECA） algo rithm is proposed, with which the adaptive flexible division of the multiple phases is conducted at the same time for obtaining automatically Gaussian model information at different stages without knowledge of a priori, a Gaussian model with even more precision, and the inter phase correlation and transition charac teristics are resolved. Then by using the superiority of KECA for processing nonlinear and high dimension al data, the different phases are respectively modeled, the fault is monitored with introduced CS（Cauchy Schwarz） statistical data and the contribution plot method is used to diagnose fault variables after the faults being detected. The result of verification with penicillin fermentation process shows that the pro posed algorithm will have higher accuracy of fault diagnosis than the MPCA and MKPCA algorithms.