Residual stresses prediction in transition metal nitrides sputtered coatings using artificial neural network and experimental evaluation of surface morphology

In the production of transition metal nitride hard coatings under specific deposition conditions, the resulting residual stresses play a critical role in determining the mechanical properties, adhesion, and overall performance of the coatings. In this work, an artificial neural network (ANN) that utilizes feed-forward architecture is used to predict the average residual stress in titanium nitride (TiN) coatings with different coating thickness. These coatings are manufactured under varying bias voltages and temperature on substrate. The impact of variations in coating thickness, bias voltage, and temperature on development of internal residual stresses and changes in microstructure within sputtered coatings have been gathered from existing literature. The data collected is within the context of fibrous (zone I) and mix columnar and fibrous structure (combined zone) as defined by an existing structure zone model. The model is trained with experimental data from literature. The model is trained with experimental data from literature. The validation of the ANN model is carried out by comparing the predictions with experimental results. X-ray diffraction is used to analyze the residual stresses, preferred orientation, and the structure (zone I and combined zone) of produced coatings under different sputtering parameters. The regression and performance curves are used to assess the efficacy of ANN model. The coefficient of determination of trained model and mean square error are 0.9918 and 0.35, respectively. The predicted results show good agreement with experimental and subsequently AFM grain morphology analysis captures the growth mechanisms involved in the residual stresses evolution.