Dynamic statistical process monitoring based on online dynamic discriminative feature analysis

Abstract A novel online discriminative dynamic feature analysis (ODDFA) algorithm is formulated and then employed for dynamic process monitoring. Different from traditional multivariate analytical algorithms which derive representative signature inherited in a dataset given from the normal operating condition, the proposed ODDFA algorithm only seeks for a pair of projecting directions, which could be discriminative to the deviation between the online sampled data and the normal operating dataset. From the standpoint of its formulation, the ODDFA algorithm is activated online with the availability of stacking online time-serial samples into a matrix form, and the latent feature resulted from a two-dimensional projection could be extremely discriminative to the inconsistency inherited in the online time-serial samples. Therefore, the utilization of the ODDFA algorithm in dynamic statistical process monitoring is always a preferable choice in contrast to other counterparts, as demonstrated through comparisons in monitoring two classical dynamic processes.