This paper presents fault detection and diagnosis methodology for batch/semi-batch processes using a multi-way orthogonal nonlinear PCA approach. In this work, a sequential extracting process of linear and nonlinear correlations from process data is performed. The approach reduces the complexity of the nonlinear PCA model structure, which dramatically improves the model generalization. An orthogonal nonlinear PCA procedure is incorporated to capture the nonlinear characteristics with a minimum number of principal components. A trajectory-boundary-limit crossing point discriminant analysis is proposed to diagnose the process faults. A two-step discriminant analysis is also incorporated to improve the diagnostic performance in the case of isotropically distributed trajectories. The validity of the proposed strategy is demonstrated by the application to an emulsion copolymerization of styrene/MMA semi-batch process. © 2010 Berkeley Electronic Press. All rights reserved.