Abstract: To solve the problems that multi-way independent component analysis (MICA) need to assume process variables conforming non-Gaussian distribution and the monitoring statistics based on the Mahalanobis distance of MICA will cause reduction in fault detection rate, a new monitoring method based on Multi-way Independent Component Analysis and One-Class Support Vector Machines (MICA-OCSVM) was researched. Firstly, the independent components (ICs) from all batches of batch process were extracted by MICA. Secondly, OCSVM was used to model for all batches' ICs at each time, separately. Meanwhile, decision hyper-plane of the OCSVM model was chosen to construct monitoring statistics. Finally, the confidence limits were determined using kernel density estimation by the monitoring statistics calculated from all modeling data. The method was applied to fed-batch penicillin fermentation process. The experiment results show that in contrast to the fault detection methods based on traditional MICA, the proposed method can make full use of ICs' structure information regardless of the distribution of process variables and can reduce the rate of misinformation and omission effectively.