Research On Surface Mirror Fitting Method For Liquid Level Detection Of Single Crystal Furnace

Monocrystalline silicon is the basic material of electronic components and photovoltaic panels,which has a huge role on modern information industry and photovoltaic industry development,especially in the semiconductor industry.Monocrystalline silicon purity form a large regulatory role on else industries,so that the extraction of high-purity silicon is particularly important.In order to guarantee higher crystal pulling quality.in the process of crystal pulling,with the crystal growth,the fused silica liquid level in the crucible descends.the fused silica liquid level must be controlled to a constant position.Therefore the liquid level need to be detected with high accuracy so that it call be regulated to the expected position timely.Curved mirror method is a method of measuring the molten silicon single crystal silicon furnace level,the principle is the use of curved mirrors to change the optical path of the reflected light level to achieve the small change in the level of amplification detector points of light displacement,liquid level error thus obtained will be smaller.In the curved mirror method throughout the measurement process,as the most important piece of hardware-curved mirror,it it good or bad have a significant direct impact on the entire measurement results,by curve fitting method to solve curved mirror function formula.Therefore,in order to obtain good meaurement results,the subject will be curved mirror fitting method to study.to fit a curved mirror as much as possible close to the real smooth curvature of the analytic function formula.To solve the problem that polynomial functions order and parameters of the curved mirror function analytic formula are unknown,a new fitting method is presented,which is based on cross-dimensional simulated annealing algorithm(TDSA),it fits with the conventional method except that:a conventional method for determining the order the model is to calculate the value of each possible standard model with maximum likelihood criterion,however the method in this paper used to solve decision model and parameters are not deterministic,which not only to optimize the parameters,and can simultaneously achieve order to select the best model,in order to achieve the model parameters in the reasoning process joint and several optimization parameters for people to solve optimization under uncertainty model fitting case provides a new idea.In order to sampling model parameter and order,we propose a mixed the MCMC sampling method which combinate the M-H sampling and Gibbs sampling.By using the fast convergence property of M-H to improve the convergence rate of the algorithm and the sampling accuracy of Gibbs small range to improve the estimation accuracy of algorithm to estimate the parameters and the order of the model.The experiments show that the cross-dimensional simulated annealing algorithm efficiency and superiority,and the calculated curved mirror letter Analytic number.