Propagation Analysis And Diagnosis Of Oscillation In Control Loops

Nowadays, the highly integrated and large scale industry processes are increasingly popular, and a typical chemical plant has hundreds or thousands of control loops. The loops will interact each other by material streams, heat integration, feedback and other means, by which the oscillation propagates throughout a plant and may lead to plant alarm, and thus endanger the operation stability and economic benefit of enterprises. The automation detection and propagation analysis of oscillation will helps to locate the root cause and remove disturbance, which will bring up the control performance and the efficiency of energy. The main contents contained in the thesis are as following:1. As for multi-frequency oscillation and noise disturbance of collected industrial variables,Propose a filtered Auto correlation function method, which calculate ACF by inverse Fourier transform of variable spectrum with spectral peak separated. This method successfully detect multi-frequency oscillation and at the same time, low frequency removed spectra enhance the accuracy of oscillation cluster method with spectra matrix. A comparison of all the methods above is give by a simulation example and TE Benchmark test.2. Overview the oscillation propagation analysis method based on cross-correlation function, transfer entropy, and propose the symbolic transfer entropy technique to estimate the causality of variables which reduces the calculated amount on the premise of acceptable accuracy. TE Benchmark test is followed to prove the validity.3. Compare the oscillation clustering methods based frequency feature, like principle component analysis of spectral matrix and get the similarity by score vectors, estimate the spectral envelope of all the variables and calculate oscillation index to get the main oscillation frequency, non-negative component analysis of spectral matrix and list the strength factor to cluster the variables. All the limitations and advantages are explained by TE Benchmark test.4. Applly Choudhury stiction model in TE process and analyze several non-linearity test methods that use the stiction data but do not require process knowledge or invasive tests.