An LSTM-PSO model for forecasting the flow behavior of a Ni-based superalloy during hot deformation

Abstract

Isothermal compressive experiments on a Ni-based superalloy were performed at strain rates from 0.001 to 1 s−1 and temperatures between 920 and 1040◦C to study its high-temperature deformation. Utilizing the experimental data, a Long Short-Term Memory (LSTM) model, optimized with the Particle Swarm Optimization (PSO) algorithm (LSTM-PSO), was developed to characterize this behavior. The LSTM component of the model effectively handles the complexity and nonlinear characteristics of time-series data, while the PSO component performs parameter optimization, enhancing the model’s accuracy and generalization capability. The model’s inputs include deformation temperature, strain rate, and true strain, with true stress as the output. A comparison of experimental and forecasted results revealed that the LSTM-PSO model accurately predicts high-temperature deformation, achieving a correction coefficient of 0.9988 and an average absolute relative error of 1.16, demonstrating superior performance compared to other advanced methods.