Fault Detection Method Based on Improved Isomap and SVM in Noise-Containing Nonlinear Process

In order to solve the problem of high dimension and nonlinearity of monitoring data in chemical process, a fault detection method based on the combination of improved isometric mapping (Isomap) and Support Vector Machines (SVM) is proposed. First of all, a new method of Isomap improvement is proposed in this paper, called Standardized Residuals-Isomap (SR-Isomap), to solve the problem that Isomap algorithm is easily affected by noise. Then based on the statistic-proximity ratio r, the residuals are analyzed and the noise is separated within the confidence intervals [-2, 2] to accurately extract the low-dimensional principal components in the high-dimensional and Nonlinear manifold under the noisy environment, the robustness of Isomap algorithm to noise is enhanced. Finally, based on the feature of minimizing the structural risk of support vector machines, an SR-Isomap-SVM fault detection model is constructed and the radial basis function suitable for process monitoring signal is chosen to train and learn the low-dimensional clustering data to realize the fault detection of nonlinear monitoring data with noise. The simulation results of Tennessee Eastman(TE) Process show that this method can effectively realize the fault detection of non-linear chemical process with noise.