Prediction of Weld Bead Geometry Using Long Short-Term Memory

Accurate modeling of weld bead geometry in the gas metal arc welding (GMAW) process is crucial for optimizing weld quality and enhancing the productivity of this process. In this study, a recurrent neural network (RNN) based on long short-term memory (LSTM) architecture was employed to predict weld bead geometry (width and height) with high accuracy. Experimental tests were conducted using a full factorial design, considering input parameters such as voltage, wire feed speed, and welding speed. The data collected from these tests were used to train and evaluate the LSTM network. Results demonstrated that the model effectively captured the complex nonlinear relationships between process parameters and weld bead geometric characteristics. The high correlation coefficients of ۰.۹۶۴ for weld bead width and ۰.۹۷۶ for weld bead height underscore the model's predictive accuracy. Compared to traditional methods like regression analysis or practical experience, this approach not only reduces costs and time but also facilitates the adoption of advanced techniques. Machine learning offers powerful tools to optimize industrial processes, as evidenced by this research. It highlights the potential of deep neural networks for industrial applications, particularly in fields such as additive manufacturing.

کلیدواژه ها:

Gas metal arc welding ، Long short-term memory ، Additive manufacturing ، Modeling