Time-series prediction of the deformation behavior of small circular ring based on residual structure and Bi-LSTM network

Purpose The purpose of this paper is to propose a novel deep learning approach to accurately predict the time-series deformation behavior and load-displacement response curves of small circular ring specimens under different displacement rates, using existing experimental data. This method aims to reduce experimental costs and improve efficiency compared to traditional testing methods. Design/methodology/approach A network model based on the residual structure and Bidirectional Long Short-Term Memory is established. The small ring tensile test data from the open data set is used as input. Data preprocessing is performed, including time point selection and standardization. Mean squared error is used as the loss function, and the model is compared with classical MLP, Long Short-Term Memory and Bidirectional Long Short-Term Memory models. Findings For the nine test samples, the proposed model achieves an R-squared of 0.9982 and a mean absolute error of 23.6810, outperforming the other five models, which had R-squared values ranging from 0.9805 to 0.9970 and mean absolute error values between 26.3013 and 96.7325. Compared to other models, the proposed model demonstrates excellent stability and accuracy in the initial and final stages of prediction. Originality/value This research pioneered the application of deep learning techniques to the field of small ring testing, proposing a novel data-driven approach distinct from traditional experimental methods. The model combines convolutional networks, fully connected layers and LSTM into a hybrid architecture capable of effectively analyzing time-series data. Compared to traditional methods requiring extensive testing, this deep learning approach uses limited existing data, reducing experimental costs and improving efficiency in studying small ring deformation behavior. It provides new ideas and methods for digital transformation and intelligent analysis in materials science.