A Data-Driven Learning Approach for Nonlinear Process Monitoring Based on Available Sensing Measurements

A data-driven learning approach is proposed to monitor nonlinear processes based only on the available sensing measurements in this paper. To achieve this aim, locally weighted projection regression (LWPR) is used to establish the model of the underlying nonlinear process with local linear models, in which the modified principal component analysis (MPCA) could be further applied for process monitoring. Moreover, the normalized weighted mean of all the proposed test statistics is employed to detect the possible abnormalities. A detailed discussion is made on principal-component-analysis-based approaches as well as on the reason of selecting MPCA under LWPR framework. The Tennessee Eastman process is first employed to demonstrate the superiority of MPCA. A numerical simulation example and an industrial benchmark of an autosuspension system are finally utilized to validate the effectiveness of the proposed scheme.