||Batch processes have been applied in many industries to manufacture high-value-added products and meet the requirements of today’s rapidly changing market. In order to ensure the product quality and the operation safety, batch process online monitoring is necessary. Due to the limited product-to-market time as well as process high dimensionality and complexity, it is difficult to build first principle models for batch processes. The data-based multivariate statistical methods, such as principal component analysis (PCA) and partial least squares (PLS), have been extended to batch process monitoring. Most of the PCA or PLS based monitoring methods have been originally developed for concerning with static rather than dynamic relationships among process variables. When they are applied in batch process monitoring, a statistical assumption of batch independence is implicitly required. Dynamics are inherent characteristics of batch processes, which may exist not only within a batch run, but also from batch to batch. It is desirable to develop a monitoring scheme to capture both within-batch and batch-to-batch dynamics simultaneously. With such motivations, a series of modeling, monitoring and fault diagnosis methods have been developed in this thesis, focusing on multivariate 2D dynamic batch processes. Two-dimensional dynamic PCA (2D-DPCA) method integrates time-series model structure with PCA algorithm to extract both cross-correlations and 2D autocorrelations in batch processes. 2D dynamics are described with the correlations between the current measurements and the lagged measurements in a lagged region called region of support (ROS). Two kinds of ROS automatic determination methods have been designed. With this scheme, the fault detection and diagnosis of 2D dynamic batch processes can be well performed in both PCA residual and score spaces, and the problems of multiphase and multiple time scales have been solved. Subspace identification method has also been combined with 2D-DPCA for model simplification.