نوع مقاله : علمی و پژوهشی
عنوان مقاله English
نویسندگان English
Lead-based solder alloys are still widely used in many electronic systems and industrial joining applications, particularly where thermal stability, reliable mechanical performance, and suitable processability are required. Among these alloys, Sn 40Pb 2.5Sb is of particular interest due to the role of antimony in improving mechanical strength, microstructural stability, and creep resistance. However, the stress–strain behavior of this alloy under mechanical loading is inherently nonlinear, path-dependent, and strongly influenced by metallurgical features, making its accurate prediction by conventional constitutive models challenging. In the present study, a data-driven approach based on a Long Short-Term Memory (LSTM) neural network was employed to predict the stress–strain curve of the Sn 40Pb 2.5Sb alloy. Experimental mechanical test data were first preprocessed, normalized, and organized as sequential input data for the LSTM model. The network architecture was designed to capture the dependency between successive points along the stress–strain curve and to reconstruct the evolution of the alloy’s mechanical response. The model was trained using the Adam optimization algorithm. Its predictive performance was evaluated using statistical criteria such as R2, MAE, and MSE. The results demonstrated that the proposed model can accurately reproduce both the overall trend and the detailed features of the experimental stress–strain curves, showing good agreement with the measured data. The findings of this study indicate that LSTM-based modeling is a promising and efficient tool for predicting the mechanical behavior of solder alloys, reducing dependence on costly experimental testing, and advancing intelligent data-driven approaches in materials engineering.
کلیدواژهها English
