Ground Settlement Estimation in HDD: Lessons from Empirical, Analytical, Numerical, and Machine Learning Models

https://doi.org/10.65161/rec25shnVknb1467m

Horizontal directional drilling (HDD) can induce ground settlement that poses a threat to roads, railways, and buried utilities. This study reviews and compares four approaches for estimating ground settlement: empirical, analytical, numerical, and machine learning methods. Eight representative studies are evaluated, encompassing clay, sand, layered ground, HDD, pipe jacking, and shield tunnelling. Each study is assessed using six criteria: (1) closeness to measured settlement, (2) number of input parameters, (3) overall complexity, (4) time required for calculation, (5) principal strengths, and (6) key methodological limitations. Evidence from published field measurements indicates that empirical formulas can reproduce settlement behaviour within the soils and project conditions for which they were calibrated, but their validity outside those conditions may be limited or uncertain. Analytical methods provide explicit relationships among ground loss, soil properties, geometry, and settlement-trough behaviour, but rely on idealized assumptions regarding soil behaviour, geometry, and boundary conditions. Finite-element and machine learning models demonstrated strong agreement with measurements in studies that included field validation. However, direct accuracy rankings among the reviewed approaches cannot be made reliably because the methods were applied to different projects and evaluated using different datasets and performance metrics. The study concludes that no single method is universally optimal. Instead, a combined strategy is recommended: simple empirical or geometry-based formulas may be used for preliminary HDD screening, while calibrated analytical or numerical models are more appropriate for critical crossings with strict settlement limits. Machine learning methods may provide additional predictive capability when sufficient high-quality, project-specific monitoring data are available.