Towards Data-driven Manufacturing of Diamond-like Carbon Coatings – Hardness Prediction with Machine Learning and Deep Learning Techniques

Diamond-like carbon (DLC) coatings with high hardness contribute to superior mechanical and tribological performance. Therefore, these coatings are widely used in a range of industrial sections such as automotive, aerospace, machine elements, cutting tools, etc. A range of experimental factors such as thermal conditions, vacuum pressures, plasma kinetics, electric and magnetic potentials, and post-treatment strategies are extensively investigated to identify those parameters that deliver higher hardness. This study uses our experimental data on substrate bias voltage and post-heat treatment to articulate DLC hardness through a data-driven approach. Machine learning and deep learning algorithms are used to predict the hardness of DLC coatings. Since this work is first of its nature for DLC coating, therefore, suitability and precision of algorithmic models are studied using linear regression model, LASSO regression, Ridge regression, Elastic net regression, LARS regression, LASSO-LARS regression, support vector regression, XG boost regression, CAT boost regression, Light GBM regression, extreme learning machines, artificial neural network, convolutional neural network, and recurrent neural network. The algorithm's performance is assessed using mean squared error, root mean squared error, coefficient of determination, mean absolute error, mean absolute percentage error, and correlation coefficient. The data-driven modeling suggests that the application of deep learning algorithms to anticipate DLC coating hardness has shown outstanding accuracy when compared to other methods. In addition, the significance of each explanatory variable is also indicated using feature importance analysis. In summary, this research presents the details of major explanatory factors of the DLC coating hardness; provides the platform to select the parameters to produce optimal DLC coating hardness; and will act as a significant milestone towards data-driven manufacturing of DLC coatings.