Research On Detection And Processing For Several Key Parameters In A Class Of Complex Industrial Environment

With the development of science and technology and the continuous improvement of industrial production automation,the requirements for detection accuracy and stability to parameters of the control system are more and more demanding,therefore,how to improve detection accuracy,resolution,real-time detection and stability become important issues in the detection field.Three research questions in a class of complex industrial environment with high-temperature and airtight are studied in this dissertation,namely the question of image segmentation and ellipse fitting in the cylinder diameter detection using CCD sensor,the question of low frequency interference suppression in the temperature detection using infrared temperature sensors that are often interfered by the low frequency narrow band,and the question of laser spot coordinate calculation,importance distribution construction and the loss of diversity of the particles in the liquid level detection using a laser triangulation method.Firstly,two main problems of image segmentation and ellipse fitting in the cylinder diameter detection have been studied.A multi-objective artificial fish swarm algorithm(MAFSA)is designed to search best thresholds of 2D histogram oblique segmentation.Simulation results on a standard test function set demonstrate the effectiveness of MAFSA.Then a 2D histogram multi-threshold oblique segmentation method based on MAFSA is proposed,which employs both the maximum between-class variance and maximum entropy as objective functions of MAFSA to search the best thresholds for the 2D histogram oblique segmentation.Image segmentation experiments conducted on practical monocrystalline silicon diameter detection show the validity of the proposed method.At last,an ellipse detection method based on artificial fish swarm algorithm-Hough transform is presented.The artificial fish swarm algorithm(AFSA)is used to search the optimum ellipse parameters in the parameter space that maximize the accumulator after the Hough transform.Simulation experiments also show the validity of the proposed method.Secondly,the problem of multi low frequency narrowband interfering in the temperature signal in the temperature measurement using infrared sensors has been studied.By introducing the iterative adaptive approach(IAA)to the adaptive cancellation,an adaptive cancellation without reference signal is presented.This method directly estimates the frequency,amplitude and phase of interferences with IAA from the measurement signals in the context of the number of interferences known,so as to eliminate interferences from the measurement signals.The validity is verified by simulation experiments.Furthermore,a trans-dimensional simulated annealing(TDSA)based-method to estimate a slow variable is proposed,which can accurately estimate the number,frequency,amplitude and phase of the interference when they are completely unknown.Simulation results show that this method can effectively estimates a slow variable with unknown number,frequency,amplitude and phase of the interference.Thirdly,two main problems of laser spot coordinate calculation and liquid level measurement data filtering have been studied.A laser spot coordinate calculation method based on the voting mechanism is presented.This method estimates actual coordinates of the laser spot by voting the small spot coordinates in the level measurement image.The validity of this method is demonstrated by simulation on silicon liquid level measurement videos.Then a particle filter method based on the set-membership estimation theory is proposed,which utilizes the set-membership estimation theory to construct importance distribution of the particle filter and make it possible to solve the problem of non-Gaussian and unknown distribution noise.The acceptance and rejection mechanism of Metropolis Hastings is introduced into the particle filter resampling to suppress the lack of particle diversity.Simulation experiments demonstrate the effectiveness of this method,and show that it can effectively track the target in the problems of Gaussian distribution,uniform distribution and mixing Gaussian and uniform distribution.Finally,the proposed algorithms,namely the AFSA-based Hough transform,the adapt cancellation without reference signal,the trans-dimensional simulated annealing(TDSA)based-method and the set-membership estimation theory-based particle filter method,were respectively applied to the measurement data of diameter,heat zone temperature and silicon liquid level in a class like Czochralski monocrystalline silicon growth process of complex industrial environment with high-temperature and airtight.Experimental results demonstrate the validity of the proposed methods to solve practical engineering problems.