To solve the problem of uncertainties in real processes, a probabilistic fault prediction algorithm is proposed
based on principle component analysis and Bayesian autoregression (AR) model. The statistics of ??2 and SPE are derived through principle component analysis. Then, the statistics of ??2 and SPE are transformed to approximately obey a normal distribution by using Box-Cox transformations. The predictions of next step can be deduced by Bayesian AR models for these transformed statistics. The predictions are retransformed to the original distribution, and probabilistic fault prediction is realized by calculating probability in the next step through kernel density estimation according to the corresponding control bounds. Finally, simulation results show the effectiveness of the algorithm.