This paper deals with detecting a process change time and identifying a changed process model. It is assumed that the underlying process be described by Auro)-Regressive (AR) models and piecewise changing. Two different approaches are integrated : one is for detecting model change time by a sequential on-line testing procedure and the other is for identifying a change process model which is based on Ba yesian classification. Experimental results verify the proposed approach could be effcientlv utilized for an on-line monitoring system.