Study On Change Point Identification Based On Data Density

Usually, control charts are used to detect the occurrence of assignable causes in industrial processes. If any assignable causes have occurred, the chart will give an out-of-control signal, and then people will do their best to identify the assignable causes and correct the problem to make the process return to the in-control state. there are two phases in the implementation of control charts. In Phase , conducted off-line, based on the baseline data, which reflects successful production, a statistical model is built. In Phaseē, conducted on-line, new observations will be compared to the baseline model. If new observations do not match closely enough with the baseline model, we conclude that the process needs adjustment.In Phase , it is commonly assumed that the baseline observations represent a single distribution. However, it is possible that the baseline may represent more than one. Therefore, it is necessary to identify different distributions implied in the history data. On the basis of historical data from filtering, extracting the baseline from different distribution and building the control chart, then we can ensure that the on-line monitoring with the a better efficiency.The identifying of change-point is the key of the stage of phase I. Most of the traditional method is based on the assumption of some known parameters distribution, and the method is very highly depended on the distribution. Based on this, it proposed the study on change-point identification based on cluster analysis of K-means. The proposed method is based on the K-means clustering techniques, a moving window theory, and the probability density profile. The proposed method is compared with the traditional method (LRT, MMA), it is proved better to identify the change-point in the baseline period.Finally, this paper gives a summary about the proposed method, and then draws a prospected of the future research direction。...