
A Data-Driven Tool for Uncertainty Quantification in the Calibration Process
Please login to view abstract download link
The Discrete Element Method (DEM) has become an indispensable tool for exploring granular flow and powder mechanics in both academic and industrial applications. These simulations rely on mathematical models to capture particle interactions, with contact parameters governing collisions and force transmission. The selection of these microscopic parameters is crucial, as they directly dictate the macroscopic bulk behavior of the simulated material. However, calibrating these parameters to align with experimental observations remains a significant challenge. Traditional approaches often involve manual tuning or optimization algorithms, both of which can be computationally expensive, time-consuming, and prone to human bias. This work introduces an automated calibration tool designed to simplify and accelerate this parameter selection process while minimizing subjectivity. Leveraging a dataset of several thousand artificial powders described in detail via the Luding model, we developed an interactive tool where users can input observed powder behavior and receive several optimized sets of contact parameters based on the closest matches within the data collection. Another key feature of our tool is its ability to handle incomplete samples. When one or more bulk characterization tests are missing or unreliable for the target powder, the tool can identify samples with similar behavior while highlighting the ones that exhibit the most variation in the missing tests. When working with expensive or hazardous materials, this workflow automatizes the detection of a given number of samples when additional tests must be added by focusing on those that most effectively depict the total variability and distances between less informative results. By quantifying the variance among similar powders, the tool offers a direct measure of uncertainty in the calibration process and, consequently, in the simulation results. This is achieved while providing deep insights into the effects of the powder description and bulk tests on processes, as well as offering a structured workflow with numerous potential applications.