Comparative Study on Comprehensive Properties Prediction of DLC Thin Films Based on Different Neural Networks

Diamond-like carbon (DLC) films are widely used in mechanical, aerospace and other fields because of their excellent properties, such as high hardness and high wear resistance. However, in order to meet the needs of different industries, DLC thin films often adopt different preparation methods and process parameters to obtain different characteristics, and it is time-consuming and laborious to test the characterization performance of samples prepared by different process flows. In order to solve this problem, this paper explores the feasibility and applicability of different artificial neural networks in the field of film performance prediction. BP, ELM and KELM neural networks were trained by using the data of DLC films prepared by experiments as training samples, taking different process parameters as inputs and various properties of DLC films prepared by corresponding processes as outputs. Verified the three models by using the verification samples and compared the results of predictive values and true values of the three neural networks, relative error, determination coefficient and mean square error. The sensitivity analysis of the properties of the film was carried out with the best performance network model as a sample. The results show that the prediction accuracy and stability of KELM are better than BP and ELM neural networks, and KELM is more suitable for predicting the comprehensive properties of DLC films. And the influence of each process parameter on each performance index is obtained.