Research Based On Partial Least Squares For Industrial Nonlinear Processes Monitoring

The proposed process monitoring technology indicates that the industrial processes are becoming automatic and intelligent.As the key to ensure the safety and stability of industrial systems,the process monitoring technology is increasingly significant.Formerly the process monitoring methods are mainly based on the mechanism models of industrial systems.It is beyond doubt,however,that the accurate mechanism models based on prior knowledge are difficult to establish to a great extent,even if the advanced means of model identification is involved in,on account of the increasing complexity of modern industrial systems.Therefore,more attention has been paid to the massive process history data,whose record,storage and processing are enabled by the development of sensors and computer communication technology.It is evident that the process data contains the correlation between variables,which proves to facilitate the process monitoring.Currently,the data-driven process monitoring methods are mostly applied to linear static processes so that the methods cannot be directly transferred to the prevalent nonlinear processes.Considering these facts,a scheme of industrial process monitoring will be discussed in this thesis,including the monitoring of linear,nonlinear and dynamic processes.The introduction of standard partial least squares(PLS)is firstly given in detail.Given the drawbacks of standard PLS,a modified version of PLS whose core idea is to achieve complete decomposition of the data space is introduced here.The application of the modified PLS in fault diagnosis is subsequently discussed.The application of kernel partial least squares(KPLS)in nonlinear processes is then discussed.A online fault diagnosis scheme of nonlinear processes is further proposed and applied in the wastewater treatment system.Considering that the data generated from real industrial processes is relatively complex,wavelet transform is also introduced to process the data.It turns out to be an effective method for monitoring nonlinear processes with immense measurement data.An exploratory study on the monitoring of dynamic processes is finally carried out.Based on the idea of decomposing the process data and the process models,multi-subphases model is combined with KPLS,whose basic idea is to set up the accurate model of the dynamic processes before monitoring it.This method serves as a promising solution of the monitoring of nonlinear dynamic processes.