Research On Nonlinearity Detection Algorithm Of Industrial Processes

The rapid development of modern industrial process brings more and more challenges to automation and control technology,where the performance of control loop has a major impact on the production cost,economic benefit and operation safety of industrial manufacture.The performance evaluation and fault diagnosis technology need prompt development.As one of the significant characteristics of the poor performance of control loops,the oscillation will cause equipment aging and result in the fluctuations of product quality by means of the feedback between single loop and the propagation among multiple loops.Researches have shown that the cause of oscillation is mainly composed of linear causes and nonlinear causes and the nonlinear characteristic caused by the valve stiction is the most commonly found problem.Most of the existing nonlinearity detection methods are based on the framework of hypothesis testing.It can be divided into parametric and nonparametric methods according to the selection of statistical index,where the nonparametric test independent of prior knowledge is commonly used.However,in actual industrial process,the above-mentioned nonlinearity detection methods are limited by the length of the signal in most situations,whose performance on nonstationary sequences and weakly nonlinear sequences is not good.This paper focuses on the nonlinearity detection problems of the control loops of industrial processes and the follow:ing issues are addressed in this paper:1.A nonlinearity detection method combining high-order spectrum and surrogate data method under the framework of hypothesis testing is proposed.A new high-order spectral statistic called Bihocerence is introduced,which can effectively eliminate the influence of signal amplitude on the estimation variance of high-order spectrum.A nonlinearity detection statistical index is designed based on the new high-order statistical tool.Furthermore,the traditional surrogate data method is combined with de-trending and re-trending to determine the confidence level of the nonlinearity detection.The numerical simulation verifies the superior performance of the proposed algorithm to the Choudhury's algorithm based on Bicoherence on weakly nonlinear sequences and short-term sequences2.Propose a method to apply the data fusion algorithms based on hypothesis testing theory to nonlinearity detection.This fusion method integrates nine single statistical indexes of nonlinearity detection and four different data fusion models to complete the nonlinearity detection.On this basis,a set of nonlinearity detection toolbox based on MATLAB is developed which can realize both function of single nonlinearity detection method and fusion method.The simulation parts provide the robustness estimation of each single nonlinearity detection method,and verifies that the fusion algorithm can make up for the deficiency of the nonlinearity detection performance of single nonlinearity detection method.3.A nonlinearity detection method based on nonstationary excitation signal to detect nonlinear distortions in the output of a dynamic system is proposed.The Local polynomial method is used to remove the leakage error and the periodic property of the input data is used to distinguish the stochastic nonlinear distortion and measurement noise in the output noise.The simulations confirm that the nonstationary excitation method is more sensitive to the nonlinear distortions in the process than the fast method proposed by Pintelon and Schoukens based on stationary excitation,and the former one can achieve higher detection accuracy on low-energy excitation level.