基于贝叶斯动态模型的自相关控制图

Auto-correlated charts based on dynamic Bayesian model

传统控制图应用的假设前提是从过程得到的观测值彼此独立.但许多过程存在自相关现象,并且这种自相关现象对控制图性能有严重影响.从工程实际出发,对AR(1)过程模型提出了基于贝叶斯动态模型的自相关控制图,阐明了此自相关控制图的原理和方法;并将该控制图与基于过程模型的自相关控制图的性能进行比较.此控制图借助贝叶斯先验估计,建模迅速,不以时间序列模型为基础,对强自相关模型具有很好的性能.对强自相关过程,并要求迅速建立过程控制时,可以使用此自相关控制图,进行过程监视.

In statistical process control it is usually assumed that the observations taken from the process are independent. However, for many processes the observations are autocorrelated, and this autocorrelation can have a significant effect on the performance of traditional control chart. The problem of monitoring the mean of process in which the observations can be modeled as an AR （ 1 ） process was considered. A one-step forecasting autocorrelated control charts based on dynamic Bayesian model was presented for engineering practices. The principle and methodology to build autocorrelated control charts was described in detail. This control chart＇s performance was compared to the performance of control chart based on process model, and the effect of process parameter estimation on the control chart was also investigated. When the process model is strong autocorrelation AR （ 1 ） , the autocorrelated control chart has good performances with few samples. In this case, autocorrelated control chart was suggested instead of traditional control charts.