Hybrid SSA-LSTM based mechanical property degradation prediction for corroded steel wires of long-span cable supported bridges

As an essential component of bridge cables, high-strength steel wires are highly susceptible to corrosion and other environmental factors during long-term service, leading to degradation of mechanical properties. Accurate prediction of such degradation is crucial for the maintenance and safety assessment of bridges. To address this issue, a hybrid SSA-LSTM prediction model was proposed, combining the sparrow search algorithm (SSA) with a long short-term memory (LSTM) neural network. The optimization capability of SSA was first validated through benchmark function tests, which demonstrated its superiority in convergence speed and optimization accuracy. Based on this, SSA was employed to optimize the hyperparameters of the LSTM model, thereby improving both training efficiency and prediction accuracy. The target mechanical properties include yield strength, yield strain, ultimate strength, ultimate strain. Furthermore, a series of mechanical property tests were conducted to obtain those indicators under different corrosion ages (0–96 days), enriching the dataset for training the SSA-LSTM. the proposed model was compared with a traditional LSTM, a gray wolf optimizer-based LSTM (GWO-LSTM), and a genetic algorithm-based LSTM (GA-LSTM). The comparison results indicate that SSA-LSTM consistently outperforms the other models in predicting the mechanical properties of corroded steel wires. This study provides a potential reference for the safety evaluation of corroded bridge cables for long-span cable supported bridges. • Developed a hybrid SSA-LSTM model for steel wire degradation prediction. • Integrated SSA to optimize LSTM hyperparameters with improved accuracy. • Built experimental dataset of corroded steel wires aged 0–96 days. • SSA-LSTM outperforms traditional LSTM, GWO-LSTM, and GA-LSTM models. • Provides a reliable tool for safety evaluation of corroded bridge cables.