Abstract Scope |
Dealing with numerous Welding Procedure Specification (WPS) / Welding Procedure Qualification Record (PQR) and selecting the most suitable one is frequently a not straightforward task for welding engineers. Moreover, selecting the welding parameters for intermediated conditions of thicknesses, or other boundary conditions, or even extrapolating such parameters are common responsibilities for the welding engineers. In this scenario, handy tools that help them are welcome. Thus, this work aims to investigate different algorithms in order to produce optimized responses with regard to the parametric relationships of the welding processes, i.e., a theme that is fully integrated into the current scenario of Industry 4.0. For the development of this research, an artificial neural network of the MLP type of two layers was developed with a database generated from different qualified PQR. The employed database refers to naval steel ASTM A131 welded by Flux Core Arc Welding (FCAW) for all root, fill and cap passes with different plate thicknesses. The obtained results show good assertiveness rate of the neural network used (less than 15% of error) and the convergence with the values obtained through the PQR. At the end, a mobile application was developed and it is based on the web-queue-worker architecture in the cloud with the focus on scalability. It is expected that the use of the developed application could possibly reduce/eliminate reworks and could lead to reduction in the cadence time and, consequently, increase in welding productivity. |