Research On Data-Driven Oscillations Detection And Diagnosis Of Industrial Control Loop

In modern process industries,control loop oscillations lead to a decrease in control loop performance,which in turn lead to equipment loss,increased energy consumption and product quality decline.Control loop oscillations seriously affect the economic benefits of enterprises.With the enlargement,large-scale and integrated of devices,the scale of the control loops are getting larger and larger.The plant-wide oscillation caused by the propagation of oscillation between control loops seriously affects production and is not easily traceable.Aiming at the detection and diagnosis of control loop oscillations,this thesis proposes a single loop oscillations detection method based on improved complete ensemble empirical mode decomposition adaptive noise and a plant-wide oscillations detection method based on improved noise assisted fast multivariate empirical mode decomposition,and puts forward the method based on dynamic slow feature analysis and improved surrogate data to the oscillation of diagnostic framework.These methods have achieved good results in the oscillations detection and diagnosis of process with noise disturbance,external disturbance and multiple oscillations.The specific research content of the thesis includes the following aspects.Aiming at the problem that the CEEMDAN has disadvantages such as endpoint effect and noise redundancy of decomposition results when processing time-varying and non-stationary signals.By carrying out end point extension,adding special auxiliary noise and residual signal mean processing method,a single loop oscillation detection method based on ICEEMDAN is proposed.This method uses ICEEMDAN to decompose the one-dimensional signal to be detected into intrinsic mode functions of different frequencies to realize the identification of multiple oscillation source signals,and then calculates the single-loop oscillations detection indexes for each intrinsic mode function to determine whether the signal has oscillation characteristics.This thesis verifies the effectiveness of this method in dealing with the control loop oscillations detection of nonlinear and non-stationary processes with noise disturbances,external disturbances and multiple oscillations through five classic practical industrial cases.Aiming at the shortcomings of the decomposition results of fast multivariate empirical mode decomposition(FMEMD)algorithm in processing complex multidimensional signals,such as mode mixing effect,the methods of endpoint extension,selection of appropriate sampling sequence and increasing of auxiliary noise are adopted.An improved FMEMD(IFMEMD)method was proposed to detect plant-wide oscillations.This method can decompose multi-channel signal into different frequency of multi-dimensional intrinsic mode function,and then calculates the plant-wide oscillation detection indexes for the multi-dimensional intrinsic mode function of each channel,so as to detect with oscillation characteristics of the channel signal and for grouping similar frequency oscillation signals.In this thesis,two simulation cases are presented to verify the effectiveness of the method in plant-wide oscillation detection of nonlinear non-stationary process control loop with noise disturbance,external disturbance and multiple oscillations.Aiming at the problem that it is difficult to realize the positioning and non-linear discrimination of multiple oscillation sources in industrial process control loops,an oscillation diagnosis method based on dynamic slow feature analysis(DSFA)and improved surrogate data method is studied.According to the slow characteristics of process interference,this method realizes the positioning of the oscillation source signal at the loop level through projection analysis and signal energy characteristics based on the effective recovery of the oscillation source signals,and further realizes the oscillation source signals based on surrogate data method.It solves the method that is easily affected by the non-stationary process faults.This thesis verifies the effectiveness of the method in dealing with the oscillation diagnosis problems caused by multiple oscillation sources through three simulation industrial cases.