Online Detection Method Of Seeding Feature Image For Sapphire Crystal Growth Via The Kyropoulos Method

Sapphire has a wide range of applications in many fields due to its excellent physical properties and chemical stability.The artificial preparation of sapphire crystal is a process in which high-purity Al₂O₃ is used as the raw material,and the raw material is melted by heating and recrystallized by using the seed crystal.At present,the Kyropoulos method is widely used in the industry to prepare large-size sapphire crystals.High-quality inoculation is the guarantee for growing high-quality crystals by the Kyropoulos method.However,at this stage,domestic enterprises rely entirely on manual experience in the inoculation stage,and the judgment of inoculation timing is too subjective,lacks a systematic inoculation theory,and the melt in the inoculation stage is in an environment of high temperature and strong light,which is not conducive to human observation,the above subjective and objective factors limit the yield of finished products,resulting in low inoculation efficiency.Therefore,how to improve the theoretical system and form technical standards is the key technology facing the current sapphire preparation industry.In view of the problems existing in the inoculation stage,this paper makes theoretical calculation and numerical simulation on the inoculation stage of sapphire prepared by the Kyropoulos method.Combined with the results of numerical simulation and artificial experience,a new method to judge the inoculation time is proposed,and the visual speed model of crystal introduction by bubbling method is established.Firstly,this paper uses the full convolution neural network based on semantic segmentation to replace the traditional image processing technology,and adds the step-by-step structure to the network for optimization,which solves the problem that the existing theoretical model has low accuracy in image processing and does not meet the real-time requirements,and improves the detection efficiency.Then,based on the images recognized by the fully convolutional neural network,the motion of the melt-free surface stripes was used as a key factor to judge the timing of inoculation,and a speed detection model of the melt-free surface stripes was proposed.A key factor in quality,a feedback model for nucleation growth is proposed.The two models are respectively aimed at the pre-seeding operation and after the seeding operation,and combined to form a visual speed model of seeding by the Kyropoulos method.Finally,an experimental platform was built to verify the model,and compared with the existing the OCS（Observation Convergence Seeding）model and the MIVD（Multi-frame Intercomparison Visual Detection）model.The experimental results show that the accuracy of the visual speed model of bubble growth method in judging the inoculation time is 92.2%,and the accuracy of judging the growth state of crystal nucleus is 97.5%.Compared with manual inoculation,the average inoculation time is shortened by 31.21%.The visual speed model of crystal introduction by the Kyropoulos method effectively improves the inoculation efficiency and lays a foundation for further reducing energy consumption in sapphire manufacturing industry.