基于缺失数据的误差生成策略及其在故障检测中的应用

误差生成是基于机理模型故障检测方法的核心本质,但鲜有应用于统计过程监测方法中.为此,提出一种基于缺失数据的误差生成策略,将能反映出采样数据对统计模型拟合程度的误差作为新的被监测对象实施故障检测.所提出的基于缺失数据的主元分析(MD-PCA)方法通过逐一假设各变量测量数据缺失后,利用缺失数据处理方法推测出相应缺失数据的估计值,并对缺失数据的实际值与估计值之间的误差实施基于PCA模型的故障检测.利用误差实施故障检测的优势在于,生成的误差能在一定程度上降低原测量变量的非高斯性程度,而且误差体现的是对应缺失变量中与其他测量变量不相关的成分信息,更能揭示各测量变量的本质.通过在TE过程上的实验充分验证了所提出方法的优势,以及MD-PCA方法用于故障检测的可行性与优越性.

Missing data based method for residual generation and its application for fault detection

Residual generation is the essental step in the model-based fault detection methods, but it has not been applied in the statistical process monitoring approaches. Therefore, a missing data based residual generation strategy is proposed, the generated residual which can indicate the fittness of sampled data to the developed statistical model is utilzied for fault detection. The proposed missing data based principal component analsyis (MD-PCA) method first assumes the measured data of individual variables is missing one by one, and the technique handling missing data is then employed for calculating the estimation of the corresponding missig variable. Ultimately, the resdiual between the actual and estimated data is modeled and monitored using the PCA-based fault detection approach. The advantages of utilizing residual for fault detection lie in that the generated residual can reduce the non-Gaussianity of the origianl measured variable to some extent, and that the residual reflects the uncorrelated information from other measured variables in the corresponding missing variable, and more essential characteristic of indicidual variables can be recovered. The case study in the TE process sufficiently demonstrates these advantages of the proposed method, and the feasibility and superiority of the MD-PCA method are validated as well.