Failure node prediction study of in-service tunnel concrete for sulfate attack by PSO-LSTM based on Markov correction

Sulfate attack seriously affects the durability of concrete structures, and many tunnels constructed in China in the early years did not adequately take it into account, resulting in structural degradation of many currently operating tunnels within their service life. In this study, we integrated experimental and machine learning (ML) approaches to assess tunnel concrete. Specifically, we compared concrete deteriorated by sulfate attack with freshly poured lining concrete that was built using the same mix ratio. By analyzing ultrasonic velocity data for different periods in wet and dry cyclic accelerated erosion experiments, we evaluated the effectiveness of multiple time series statistical methods, ML techniques, and their optimization algorithms. Through the analysis, we derived the best prediction model Particle Swarm Optimization - Long Short Term Memory (PSO-LSTM) and its Markov-corrected prediction results. Based on the damage time nodes of concrete compressive strength corrosion resistance coefficient (CCSCRC) and its actual damage in the test, the damage nodes of the target lining concrete of the existing tunnel are predicted. The experimental results show that: (1) The PSO-LSTM model is well adapted to the prediction of the deterioration data for concrete subjected to sulfate attack, with a model correlation coefficient R2 of 0.9739. (2) The PSO-LSTM model corrected by Markov chain can effectively capture the trend of the predicted data of concrete subjected to sulfate attack and improve the prediction accuracy. (3) When the CCSCRC decreased to 82 %, the loosening and granulation of the tunnel concrete occurred. In addition, actual failure of the concrete structure was observed at 677.6 days after sampling, less than two years. The outcomes of this research enhance the precision in predicting the service life of tunnel linings subjected to sulfate attack, providing valuable insights for on-site maintenance and construction, and contributing as a reference for similar studies.