We investigate the impact of different numbers of positive and negative examples on machine learning for sapphire crystals defects prediction. We obtain the models of crystal growth parameters influence on the sapphir...We investigate the impact of different numbers of positive and negative examples on machine learning for sapphire crystals defects prediction. We obtain the models of crystal growth parameters influence on the sapphire crystal growth. For example, these models allow predicting the defects that occur due to local overcooling of crucible walls in the thermal node leading to the accelerated crystal growth. We also develop the prediction models for obtaining the crystal weight, blocks, cracks, bubbles formation, and total defect characteristics. The models were trained on all data sets and later tested for generalization on testing sets, which did not overlap the training set.During training and testing, we find the recall and precision of prediction, and analyze the correlation among the features. The results have shown that the precision of the neural network method for predicting defects formed by local overcooling of the crucible reached 0.94.展开更多
基金supported by the Russian Foundation for Basic Research Projects under Grant No.16-52-48016ИНД_оми(R.Kumar and A.V.Filimonov)。
文摘We investigate the impact of different numbers of positive and negative examples on machine learning for sapphire crystals defects prediction. We obtain the models of crystal growth parameters influence on the sapphire crystal growth. For example, these models allow predicting the defects that occur due to local overcooling of crucible walls in the thermal node leading to the accelerated crystal growth. We also develop the prediction models for obtaining the crystal weight, blocks, cracks, bubbles formation, and total defect characteristics. The models were trained on all data sets and later tested for generalization on testing sets, which did not overlap the training set.During training and testing, we find the recall and precision of prediction, and analyze the correlation among the features. The results have shown that the precision of the neural network method for predicting defects formed by local overcooling of the crucible reached 0.94.
