An online fault detection approach is developed based on the multivariate statistical process control in this paper. It integrates the time-lagged windows of process dynamic behavior with the multi-way principal component analysis(MPCA). Using the previous process variables during the process without expensive computations to anticipate the future measurements, the method emphasizes particularly for on-line process monitoring and exactly faults detecting which results in extraordinary behavior of processes. Like traditional MPCA approaches, the only information needed to set up the control chart is the historical data collected from the past successful fermentation processes. This leads to simple monitoring charts, easy tracking of the progress in each process and monitoring the occurrence of the observable upsets. Erythromycin fermentation process is used to investigate the potential application of the proposed method.