Shahrood University of Technology Journal of Mining and Environment 2251-8592 Articles in Press 2026 05 04 An Integrated FDM–GA Framework for Automated Back Analysis in Tunneling–Part I: Methodology, Implementation, and Verification 3779 10.22044/jme.2026.17274.3421 EN Mohammad-Taghi Hamzaban Mining Engineering Faculty, Sahand University of Technology, Tabriz, Iran. Alireza Chehreghan Mining Engineering Faculty, Sahand University of Technology, Tabriz, Iran. Roozbeh Geraili Mikola Lead Associate, Delve Underground, Walnut Creek, CA, USA. Journal Article 2025 12 06 Back analysis of tunnel excavation plays a fundamental role in calibrating geomechanical parameters using field monitoring data. However, conventional direct back analysis procedures remain computationally demanding and highly dependent on operator supervision. This study presents an integrated Finite Difference Method–Genetic Algorithm (FDM–GA) framework for automated tunnel back analysis, implemented entirely within the FLAC environment using the embedded FISH programming language. The proposed approach eliminates the need for external optimization software and data transfer between numerical and artificial intelligence platforms. A simplified genetic algorithm is coupled directly with finite difference simulations to iteratively minimize the discrepancy between measured and computed tunnel convergences. The framework incorporates constrained parameter optimization, automated handling of non-convergent models, and a robust convergence-based stopping criterion that avoids predefined error thresholds. Verification is performed using two synthetic plane-strain tunnel models representing stiff cohesive soil and dense granular material. Six unknown parameters (<em>ρ</em>, <em>E</em>, <em>ν</em>, <em>c</em>, <em>φ</em>, and <em>K₀</em>) are back-calculated using only three convergence measurements. Results from multiple independent runs demonstrate stable convergence toward very small error values (on the order of 10^-6–10^-5) and consistent reproduction of synthetic monitoring data. The method successfully narrows broad initial parameter ranges and produces multiple acceptable parameter sets, explicitly acknowledging the non-uniqueness inherent in back analysis problems. The developed FDM–GA framework provides an efficient, self-contained, and adaptable tool for practical tunnel back analysis applications.