Limits of identifiability and uncertainty propagation in DEM parameter calibration using the angle of repose as the objective function

Discrete Element Method (DEM) simulations require the specification of numerous contact parameters, and their calibration remains a major practical challenge. This study examines the identifiability and uncertainty propagation structures inherent in DEM parameter calibration when the angle of repose is used as the sole objective function. A surrogate-based inverse analysis framework was constructed using 10,000 DEM simulations with seven contact parameters sampled from uniform distributions. Several regression-based surrogate models were trained, and the dominant parameters were extracted through permutation importance. Furthermore, Monte Carlo inverse analyses were conducted by varying the standard deviation of the target angle of repose σ from 0 . 5 ° to 3 . 0 ° to evaluate the resulting distributions and coefficients of variation (CV) of the estimated parameters. The results demonstrated that the rolling friction coefficient exhibits the highest contribution to the angle of repose but simultaneously shows strong amplification of uncertainty, resulting in highly unstable parameter estimates. In contrast, the CVs of the static friction coefficient and Young’s modulus remained nearly constant regardless of σ , indicating insufficient constraint rather than stable identifiability. These findings reveal that the inverse problem becomes ill-conditioned when the angle of repose is used as the sole scalar observable, leading to unreliable parameter estimates and inherent non-uniqueness. The study quantitatively clarifies the structural limitations of DEM parameter calibration based solely on the angle of repose and highlights the necessity of incorporating additional observables and multi-objective inverse analysis to improve identifiability and reliability. • 10,000 DEM runs and surrogate models were built for the angle of repose. • Rolling friction was the dominant parameter affecting the repose angle. • Static friction and stiffness showed weak influence on the repose angle. • Inverse analysis showed strong noise amplification in rolling friction. • A single repose angle is insufficient; multi-objective calibration is needed.