Analyzing Welding Performance of Metal Using Artificial Neural Network

Abstract

In recent years, the speed of modernization construction in China has been exponentially growing. The trend of high parameters, large capacity, and large-scale development of the welding structure has been promoted. It needs higher requirements on the type and quality of the welding materials. Most of the welding materials are imported to China. The main reason is that China still follows the traditional design method. The quality of the welding materials is also low. The design of metal welding materials involves many factors and properties. There is no fixed-function relationship between the properties and components of the welding materials. This makes it difficult to design metal welding materials. The emergence of neural network algorithms provides a new way to analyze the weldability of metal materials. In this paper, BP (backpropagation) network is used to analyze the welding performance of metals. The tensile test of welded joints is carried out through training test samples. The results show that the tensile strength and yield strength of metal materials are about 500 MPa (megapascals) and 400 MPa, respectively. For further analysis of the influence of welding current, electrode pressure, and power-on time on the tensile and shear strength of metal materials, a shear test and tension test were used. With the increase of welding current, the shear strength of spot welding continuously increased. When the welding current was 10,000A (Ampere), the shear strength decreased rapidly from 24.25 MPa to 18.84 MPa. After prolonging the welding time, at first, both tensile strength and shear strength increase and then decrease. When the welding pressure increases from 32psi to 48psi, the tensile strength increases from 16.47 MPa to 24.52 MPa and then decreases continuously to 17.26 MPa, whereas the shear strength decreases first and then increases.

Advanced ANN Techniques for Precise Detection and Classification of Welding Defects

“…The optimization process applied to the model used the L-BFGS-B solver that was allowed to face an option of 200 maximum iterations (Livieris, 2019). This is because different works have ascertained it to be working correctly in the handling of complex classification tasks (Gundewar et al, 2022;Wu, 2022). A data set was made prepared to train the model.…”