Artificial neural network approach for mechanical properties prediction of TC4 titanium alloy treated by laser shock processing. (November 2021)

• Record Type: Journal Article
• Authors: Wu, Jiajun; Huang, Zheng; Qiao, Hongchao; Zhao, Yongjie; Li, Jingfeng; Zhao, Jibin
• Is Part Of: Optics & laser technology. Volume 143(2022)
• Publication Date: 2021-11
• Subjects: Laser shock processing -- TC4 titanium alloy -- Residual stress -- Micro-hardness -- Artificial neural network; Optics -- Periodicals; Lasers -- Periodicals; Electronic journals; 621.366
• DOI: 10.1016/j.optlastec.2021.107385
• Languages: English
• ISSNs: 0030-3992

Abstract

Highlights: Artificial neural network is used to predict the mechanical properties of TC4 titanium alloy treated by LSP. The distribution law of residual stress is similar to the distribution law of micro-hardness. The strengthening by LSP treatment is depended on the basic properties of material. These input parameters are independent, while correlation among the two output parameters is strong. Abstract: Laser shock processing (LSP), which utilizes the stress effect induced by high-energy nanosecond pulse lasers to improve the mechanical properties and fatigue performance of metallic materials or alloys, is known as one of the most advanced surface modification techniques. In this work, a novel method based on artificial neural network (ANN) is applied to predict the residual stress and micro-hardness of TC4 titanium alloy treated by LSP. The experiment samples were treated by LSP with laser pulse energy of 3, 5, and 7 J and overlap rate of 10%, 30%, and 50%. Residual stress and micro-hardness were characterized by X-ray residual stress tester and Vickers micro-hardness tester. The ANN structure with four layers was employed, laser pulse energy, overlap rate and depth were set as the input parameters, while the residual stress and micro-hardness were set as the output parameters. The developed ANN model with the network configuration of 3 × 10 × 10 × 2 form a good correlation to predict residual stress and micro-hardness. The coefficient of determination R 2, mean absolute error (MAE) and root mean squared error (RMSE) of testing data sets for residual stress and micro-hardness are 0.997 and 0.987, 7.226 and 2.632, and 9.956 and 3.321, respectively. It can be concluded that the ANN is a suitable method to predict the mechanical properties of materials treated by LSP when with limited experimental data.