Contact parameter calibration for flax threshing materials using machine learning and the Brazil nut effect

Contact parameters between agricultural granular mixtures are crucial for the development and simulation optimization of agricultural machinery but are challenging to measure experimentally. This study utilizes the Brazil nut effect (BNE) to use particle volume concentration as a macroscopic response indicator and to calibrate six contact parameters of flax threshing material. By combining optimized Latin hypercube sampling with vibration segregation simulation tests, the study generates input parameters for training the PSO-BP model, which is optimized using an improved Non-dominated Sorting Genetic Algorithm (NSGA-II). The objective is to minimize the volume concentration error of flax seeds, short stems, and capsule shells between simulation and physical tests, resulting in an optimal Pareto set. Validation through static and dynamic repose angle tests shows errors within 5% (static) and 4% (dynamic), respectively, confirming the reliability of the calibration results and the research method. [Display omitted] •A NSGA II parameter calibration method based on the BNE.•This method is more standardized, unified, and accurate.•PSO-BP model and NSGA II have good applicability in this study.•DEM virtual simulation tests and numerical analysis were conducted.•Technical support is provided for calibration of relevant granular materials.