Application of nonlinear feature extraction and least square support vector machines for fault diagnosis of chemical process