Fast and precise DEM parameter calibration for Cucurbita ficifolia seeds

The lack of discrete element method (DEM) models and calibration parameters for Cucurbita ficifolia seeds, as well as low accuracy and efficiency of common parameters calibration methods, hinder the application of DEM for computer simulation in air-suction directional seeding equipment. In this study, the DEM parameters of the seeds were calibrated. The angle of repose (AOR), intrinsic parameters, and partial contact parameters of the seeds were experimentally measured. The seed 3D models were reconstructed based on the three-view profile information. The parameters and their value ranges were filtered through the Plackett–Burman design and steepest ascent test. The response surface method (RSM) and machine learning were utilised for optimisation inversion of the parameters. The experiments showed that the geometric relative error of the seed model was 0.69–6.54%, which meets the modelling requirements for DEM. The seed–seed static friction coefficient, the seed–seed and the seed–PVC rolling friction coefficient were 0.341, 0.026, and 0.059, respectively, which were obtained by inverting the GA-BP regression model via the Genetic Algorithm. The simulated AOR was 26.64°, with a relative error compared to the actual AOR of 1.64%, which was better than the simulated AOR obtained by RSM optimisation. The greater the smoothing value setting in EDEM software, the less the particle filling, resulting in improved simulation efficiency but reduced model accuracy. The CPU + GPU(CUDA) solver showed high DEM solution efficiency. The results reveal that the method can be used to quickly and accurately construct a 3D model of the seed, and the parameter optimisation accuracy of GA-BP-GA is better than that of RSM.