Construction and validation of the predictive model for DEM parameters of northeastern loamy soil at varying moisture contents

A prediction model has been limited to discrete Element method (DEM) parameters of the northeastern loamy soil at varying moisture contents. However, conventional single-point calibration has caused tedious and repetitive operation, due to highly variable soil moisture levels. In this study, an efficient and direct prediction model was constructed to simulate the mechanical behaviors of cohesive wet soils. A comparison was also conducted on the applicability, accuracy, and reliability of the self-developed Computer-Aided Engineering (CAE) software, AgriDEM, and the widely used commercial software, EDEM. Physical experiments and simulations were employed for validations. Initially, physical stacking tests were conducted using the cylinder lifting at multiple moisture levels. A prediction was then constructed for the relationship between soil moisture content and the angle of repose. Subsequently, the Box-Behnken experiment was utilized to calibrate microscopic contact parameters, including the Hertz-Mindlin with Johnson-Kendall-Roberts (JKR) model in EDEM and the cohesive particle contact model in AgriDEM. The macroscopic angle of repose was selected to map the relationships between the angle of repose and microscopic parameters. Finally, a direct prediction model was established to link the moisture content with the parameters. This coupled model was then independently validated under various vertical loads using soil direct shear tests. The internal friction angle and cohesive force were evaluated between simulations and physical experiments. The results demonstrated that there was a highly significant quadratic polynomial relationship between the soil moisture content and the macroscopic angle of repose, providing for a reliable macroscopic target for parameter inversion. High-precision mapping relationships were established between the angle of repose and microscopic parameters—specifically, the JKR surface energy in EDEM and adhesion parameters in AgriDEM. A direct pathway was achieved from moisture content to parameters, indicating continuous parameter prediction under diverse moisture conditions. The direct shear tests validated that the high accuracy of the prediction model was achieved in AgriDEM software. The simulated moisture contents without the initial modeling were highly consistent with the physical measurement in the direct shear test. Both software platforms accurately reproduced the test. The internal friction angle decreased, while the cohesive force increased, as soil moisture content increased. Notably, the AgriDEM software exhibited a significantly higher degree of agreement with the experimental values in key mechanical indicators, compared with the commercial software EDEM. The relative errors for the internal friction angle and cohesive force also ranged of 0.90% to 4.52% and 0.64% to 6.71%, respectively, in AgriDEM, compared with the EDEM (relative errors of 0.80% to 7.19% and 4.78% to 6.49%, respectively). The parameters in the cohesive particle model were introduced for the maximum attraction adjustment and normal adhesion distance during separation. The adhesion force is maintained over an extended distance during particle separation, thereby indicating the plastic deformation, yield, and liquid bridge of wet cohesive soils. In conclusion, the direct prediction model can provide a highly efficient, reliable, and continuous parameter acquisition for the DEM modeling of soils under varying moisture conditions. The exceptional fidelity of AgriDEM software was validated to simulate the mechanical behavior of wet cohesive soils. These findings can offer robust data support for the agricultural engineering software and the adaptive optimization of machinery components in complete operational cycles.