基于动态主元分析的统计过程监视

STATISTICAL PROCESS MONITORING BASED ON DYNAMIC PRINCIPAL COMPONENT ANALYSIS

针对时序相关观测数据 ,提出了一种动态主元分析过程 .仿真计算结果表明 ,过程观测数据的动态主元分析可看作是扰动“驱动”信号的提取过程 ,分析得到的主元变量实际上是驱动扰动的估计 .对CSTR过程的仿真监视研究 ,验证了基于动态主元分析的统计过程监视的有效性 .

Multivariate statistical process control (MSPC) forms the basis of process performance monitoring and detection of process malfunctions. The cornerstones of MSPC are the projection methods of principle component analysis (PCA) and projection to latent structures (PLS). These methods assume that process measurements are serially independent, which unfortunately is often invalid in real situations. For coping with serially correlated observations, A dynamic PCA (DPCA) based monitoring procedure is presented in this paper.By implementing PCA to observed disturbances other than process measurements, the new procedure eliminates the effects of process dynamics and meanwhile retains the ones of driving forces. Simulations carride out to a 6-dimesional linear system indicate that the DPCA can be viewed as a procedure of extracting driving forces ,and the dynamic principle components obtained are essentially an estimation of them . A CSTR process model is finally utilized to verify the effectiveness of the DPCA-based statistical process monitoring.