A Novel Incipient Fault Detection Method for Nonlinear Dynamic Industrial Process Based on CVD-SVDD

To address the problem that it is difficult to detect incipient faults in nonlinear dynamic industrial processes due to their low magnitude and inconspicuous fault characteristics, this paper proposes an incipient fault detection method based on canonical variate dissimilarity (CVD)-support vector data description (SVDD). Firstly, the correlation between past and future data is maximized by canonical variate analysis (CVA) to establish a state space model based on the training data under normal conditions, and the degree of prediction of past canonical variates (CVs) to future CVs, i.e., CVD, is used to discover small data deviations. Then, a minimal hypersphere is built for CVD using SVDD to complete the distinction between faulty and normal data. Finally, experimental simulations are conducted in the continuous stirred tank reactor system, and the experimental results show that CVD-SVDD has a better effect of incipient fault detection compared with the traditional method.