摘要：AbstractThis paper revisits the monitoring of traditional batch processes and introduces a different concept to diagnose incipient fault conditions. Traditional data-driven methods utilize a 3D arrangement of recorded data from a number of reference batches and unfold this arrangement to construct conventional non-negative squared monitoring statistics. To address the presence of non-stationary and non-Gaussian distributed process variables, the proposed concept relies on the statistical local approach. Following from the central limit theorem, the statistics derived from the statistical local approach have an asymptotic multivariate normal distribution. Moreover, the moving window approach that underlies the statistical local approach is applied over the entire batch cycle. This, in turn, allows describing the typically nonstationary trends in temperature and pressure variables as well as concentrations within a batch cycle. Through an application to a simulated penicillin fermentation process, the paper shows that monitoring statistics derived from the statistical local approach are more sensitive than traditional multivariate statistical methods.
