Industrial batch processes are often characterized by durations of uneven lengths and large fluctuation of initial conditions. In this paper, to effectively detect weak faults caused by large fluctuation of initial conditions and capture inherent dynamics and nonlinearities and, meanwhile, to deal with the problem of batch trajectories of unequal duration and reduce computational complexion, a FP-based multimodel single dynamic kernel principle component analysis (M-SDKPCA) modeling and online monitoring method is proposed. Since the method integrates kernel PCA and autoregressive moving average exogenous time series model and refines models, M-SDKPCA improves the models on weak faults detection ability. In order to solve the problem of large amount of calculation brought by kernel operation, a three-step feature points (FP) extraction method is used to reduce the number of samples and equalize all batch length. The FP-based M-SDKPCA monitoring method was applied to fault detection for benchmark of fed-batch penicillin production and compared with M-SDKPCA. The monitoring results demonstrated that the proposed approach showed similar monitoring results with M-SDKPCA and the computational complexion and storage requirements are significantly reduced.