Abstract: An integrated framework consisting of sub-stage models and improved multi-way principal component analysis (MPCA) is developed to monitor the multistage batch process. After batch data time unfolding, lattice degrees of nearness (LDN) of correspondence statistics of adjacent time slices that are the center of variables of loading matrixes are calculated based on the theory of fuzzy pattern recognition.According to the criterion of minimum close-degree, sorted LDN is analyzed to realize operation sub-stage particularly. To overcome alarms' shortcomings of conventional MPCA, an improved MPCA method is adopted. Sub-stage PCA model is established by performing improved MPCA in each sub-stage. The case studies from a simulated fed-batch penicillin cultivation process indicate that it gives better monitoring results in terms of sensitivity and time to fault detection than MPCA and ATMPCA methods.