Chemical processes usually have many transitions, e.g., startup and shutdown, transferring between different modes. These processes are highly nonlinear with large variations and require continuous monitoring by experienced operators. Multivariate statistical methods (MSM) is a popular monitoring method for transition processes. The main advantage of MSM is the quick detection of abnormal events, but the contribution plot analysis cannot obtain better accuracy and reliability for fault diagnosis. A hybrid fault diagnosis strategy is proposed in this paper by combining principal component analysis (PCA) and on-line dynamic time warping. By means of a simulation experiment of the startup process of a laboratory scale distillation column case, the performance of the hybrid fault diagnosis strategy is demonstrated.