Quality-relevant monitoring for multiphase batch processes is necessary. Between-phase transitions carry significant quality information and need particular attentions. In this paper, a Just-in-time-learning (JITL) based method is introduced to identify transitions and update modeling dataset of transitions. Due to the non-Gaussian distribution of the samples in the local model, a PLS-SVDD based method is proposed for modeling and monitoring. Fed-batch penicillin fermentation process is tested for performance evaluation of the proposed method.