Abstract: The fault detection time of many methods based on kernel independent component analysis(KICA)is easily affected by the order of independent components(ICs)and the number of dominant ICs chosen empirically.Aiming at this problem,a method based on KICA and Gaussian mixture model(GMM)is proposed.KICA is used to extract ICs from the dataset measured under the normal condition,GMM is adopted to fit the probability density function of each IC,and then a monitoring statistic and corresponding control limit are constructed based on the GMM.The average value of each IC’s monitoring statistic is calculated and applied to judge its non-Gaussian degree.Each strong non-Gaussian IC is monitored independently by the built monitoring statistic,and the weak non-Gaussian part is monitored by principal component analysis.The simulation results on the Tennessee Eastman process illustrate that,in contrast to the fault detection methods based on KICA,the proposed method needn’t sort the ICs and select the number of dominant ICs avoiding their effects on fault detection time,make effective use of process information,and shorten the fault detection latency.

Key words: fault detection, kernel independent component analysis, Gaussian mixture model, order of independent components, principal component analysis

摘要： 核独立元分析(kernel independent component analysis,KICA)故障检测方法的故障检测时间易受独立元顺序和主导独立元数目经验选取的影响,针对这个问题,提出基于KICA和高斯混合模型(Gaussian mixture model,GMM)的故障检测方法。采用KICA从正常工况测量数据中提取独立元,用GMM拟合各独立元的概率密度函数,建立基于GMM的监控量及其控制限;计算各独立元的监控量均值,以此判断其非高斯性强弱,对每个强非高斯独立元进行单独监控,对弱非高斯部分采用主元分析法进行监控。在Tennessee Eastman过程上的仿真结果说明,相比于KICA故障检测方法,所提方法不需要排序独立元和选取主导独立元数目,避免了其对故障检测时间的影响,能够有效利用过程信息,缩短故障检测的延迟时间。

关键词: 故障检测, 核独立元分析, 高斯混合模型, 独立元顺序, 主元分析法