Prediction of bearing capacity of cracked asymmetrical double-arch tunnels using the artificial neural networks

Abstract

Cracking in lining structures has the potential to seriously weaken the lining's carrying capacity. In this paper, the failure performance of cracked lining was firstly obtained by model tests and numerical simulations. An artificial neural network model was subsequently developed using extensive simulation data derived from the verified numerical model, to forecast the capacity loss in cracked double-arch tunnels. The results showed that: (1) The cracks caused the linings to undergo localized crushing failure, which greatly reduced the double-arch tunnels' load-bearing capacity. The cracked linings entered a stable working stage with cracks initially and then progressed directly to the destabilization stage, displaying hidden and sudden characteristics; (2) The neural network model, developed using a large number of numerical simulations, demonstrated outstanding predictive performance and accurately forecasted the load-bearing capacity of double-arch tunnels; (3) The depth of the cracks was the principal factor influencing the load carrying capacity decrease in double-arch tunnels, particularly in the vicinity of the void; (4) The structural capacity diminished swiftly when crack depth surpasses 30% of the lining's thickness, and larger voids and cracks closer to the voids amplified the lining-bearing capacity loss for equivalent crack depths.