A neural network-based integrated strength assessment method for mixed hard-soft rock masses at tunnel excavation faces

To address the challenge of accurately assessing the integrated surrounding rock mass strength at tunnel excavation faces due to mixed hard-soft rock masses distribution, this study proposes a backpropagation (BP) neural network-based strength assessment method. Finite element simulations were conducted to analyze tunnel deformation under varying uniaxial compressive strengths, area ratios, distribution positions, and overburden depths of heterogeneous rock masses. Numerical results demonstrate that lithological distribution exerts significant control over deformation behavior: increased hard rock proportion reduces overall tunnel deformation, while optimal load-bearing capacity occurs when hard rock is positioned at the crown or side walls. These findings indicate complex nonlinear relationships between lithological parameters and strength characteristics, necessitating a data-driven approach. Consequently, a BP neural network model was developed with lithological distribution position, area ratio, rock strength, and overburden depth as inputs, and lithological strength weighting factors as outputs. Applied to a railway tunnel project, the method computed the integrated surrounding rock mass strength and performed Basic Quality (BQ) classification. The classification results align closely with actual engineering conditions, validating the method’s practical applicability.