A dynamic nonlinear process fault diagnosis method using Canonical rotation forest

摘要

Conventional rotation forest does not consider process dynamics. In order to use process dynamics and nonlinearity for classification, a canonical rotation forest (Canonical RF) is proposed and used for process fault diagnosis. For each time instant, a past vector and a future vector are formed. Canonical variate analysis is used for features extraction to obtain dynamic and uncorrelated features at each node of rotation forest. The improved rotation forest is used as a classifier to diagnose process fault class. Simulations on Tennessee Eastman process show that the proposed Canonical RF outperforms the conventional rotation forest in terms of fault diagnosis accuracy.