Controlling the in-service welding parameters for T-shape steel pipes using neural network

Abstract

One of the most common practices in petrochemical and power generation industries is “in-service welding,” and the possibility of a catastrophic event known as “burn-through” in this process may lead to financial or even human losses. To reduce the risk of burn-through, it is necessary to simulate the process and study the effects of controlling parameters. In this paper; firstly, a finite element based numerical model is developed using a model updating method and experimental data. The model is employed to simulate the in-service welding of a T-shape steel pipe connection. Then, the effects of the main parameters on this process such as heat input, welding speed, pipe thickness, fluid flow and especially material properties are investigated. Finally, the experimental data together with a large set of results produced by the numerical simulation are used to compose a user-friendly computer code based on the neural network algorithms to predict the temperature levels in the critical points for different welding conditions. The output of this code can be used in the industrial environment to prevent burn-through accidents during the in-service welding.