Probabilistic fault prediction of incipient fault

In this work, a probabilistic fault prediction approach is presented for prediction of incipient fault in an uncertain way. The approach has two stages. In the first stage, normal data is analyzed by principle component analysis (PCA) to get control limits of the statistics of T² and SPE. In the second stage, fault data starts by PCA so as to derive the statistics of T² and SPE. Then, the samplings of these two statistics obeying some certain prediction distribution are obtained using Bayesian AR model on the basis of the Winbugs software. At last, one-step prediction fault probabilities are estimated by kernel density estimation method according to the statistics´ corresponding control limits. The prediction performance of this approach is illustrated using the data from the simulator of the Tennessee Eastman process.