Machine Learning for Retrieving Optimal Process Parameters Toward Artificial Diamond Synthesis

Synthetic diamonds offer an increasingly advantageous alternative to natural ones, which exhibit exorbitant prices and often originate from political and social crisis areas. This contribution highlights and compares two approaches based on Deep Neural Networks (DNNs), respectively Random Forests (RFs), contrived in order to retrieve optimal process parameters for lab-grown diamond films. Additionally, the importance of the parameters substrate temperature, gas pressure, and methane concentration on the diamond film growth rate is scored and ranked by a RF regressor. The evaluation section reveals that the proposed machine learning techniques yield more accurate results than those of statistical methods used in the past. Moreover, the DNN with feature normalization leads to a test set based mean absolute error lower by 29% than the RF technique.