diff --git a/joss.06338/10.21105.joss.06338.crossref.xml b/joss.06338/10.21105.joss.06338.crossref.xml new file mode 100644 index 0000000000..2c2252e150 --- /dev/null +++ b/joss.06338/10.21105.joss.06338.crossref.xml @@ -0,0 +1,286 @@ + + + + 20240502T072653-3c66a98204fb10c4a7cd444ffd784f1af467b6ed + 20240502072653 + + JOSS Admin + admin@theoj.org + + The Open Journal + + + + + Journal of Open Source Software + JOSS + 2475-9066 + + 10.21105/joss + https://joss.theoj.org + + + + + 05 + 2024 + + + 9 + + 97 + + + + GrainLearning: A Bayesian uncertainty quantification +toolbox for discrete and continuum numerical models of granular +materials + + + + Hongyang + Cheng + https://orcid.org/0000-0001-7652-8600 + + + Luisa + Orozco + https://orcid.org/0000-0002-9153-650X + + + Retief + Lubbe + + + Aron + Jansen + https://orcid.org/0000-0002-4764-9347 + + + Philipp + Hartmann + https://orcid.org/0000-0002-2524-8024 + + + Klaus + Thoeni + https://orcid.org/0000-0001-7351-7447 + + + + 05 + 02 + 2024 + + + 6338 + + + 10.21105/joss.06338 + + + http://creativecommons.org/licenses/by/4.0/ + http://creativecommons.org/licenses/by/4.0/ + http://creativecommons.org/licenses/by/4.0/ + + + + Software archive + 10.5281/zenodo.11001174 + + + GitHub review issue + https://github.com/openjournals/joss-reviews/issues/6338 + + + + 10.21105/joss.06338 + https://joss.theoj.org/papers/10.21105/joss.06338 + + + https://joss.theoj.org/papers/10.21105/joss.06338.pdf + + + + + + An iterative bayesian filtering framework for +fast and automated calibration of DEM models + Cheng + Computer Methods in Applied Mechanics and +Engineering + 350 + 10.1016/j.cma.2019.01.027 + 0045-7825 + 2019 + Cheng, H., Shuku, T., Thoeni, K., +Tempone, P., Luding, S., & Magnanimo, V. (2019). An iterative +bayesian filtering framework for fast and automated calibration of DEM +models. Computer Methods in Applied Mechanics and Engineering, 350, +268–294. +https://doi.org/10.1016/j.cma.2019.01.027 + + + Probabilistic calibration of discrete element +simulations using the sequential quasi-monte carlo +filter + Cheng + Granular Matter + 20 + 10.1007/s10035-017-0781-y + 2018 + Cheng, H., Shuku, T., Thoeni, K., +& Yamamoto, H. (2018). Probabilistic calibration of discrete element +simulations using the sequential quasi-monte carlo filter. Granular +Matter, 20. +https://doi.org/10.1007/s10035-017-0781-y + + + Performance study of iterative bayesian +filtering to develop an efficient calibration framework for +DEM + Hartmann + Computers and Geotechnics + 141 + 10.1016/j.compgeo.2021.104491 + 2022 + Hartmann, P., Cheng, H., & +Thoeni, K. (2022). Performance study of iterative bayesian filtering to +develop an efficient calibration framework for DEM. Computers and +Geotechnics, 141. +https://doi.org/10.1016/j.compgeo.2021.104491 + + + Visco-elastic sintering kinetics in virgin +and aged polymer powders + Alvarez + Powder Technology + 397 + 10.1016/j.powtec.2021.11.044 + 0032-5910 + 2022 + Alvarez, J. E., Snijder, H., Vaneker, +T., Cheng, H., Thornton, A. R., Luding, S., & Weinhart, T. (2022). +Visco-elastic sintering kinetics in virgin and aged polymer powders. +Powder Technology, 397, 117000. +https://doi.org/10.1016/j.powtec.2021.11.044 + + + Machine learning in the calibration process +of discrete particle model + Nguyen + 2022 + Nguyen, Q. H. (2022). Machine +learning in the calibration process of discrete particle model. +http://essay.utwente.nl/91991/ + + + Discrete element modelling of uplift of rigid +pipes deeply buried in dense sand + Li + Computers and Geotechnics + 166 + 10.1016/j.compgeo.2023.105957 + 0266-352X + 2024 + Li, X., Kouretzis, G., & Thoeni, +K. (2024). Discrete element modelling of uplift of rigid pipes deeply +buried in dense sand. Computers and Geotechnics, 166, 105957. +https://doi.org/10.1016/j.compgeo.2023.105957 + + + Simulating industrial scenarios: With the +open-source software MercuryDPM + Thornton + 10.23967/c.particles.2023.015 + 2023 + Thornton, A., Nguyen, Q., Polman, H., +Bisschop, J., Weinhart-Mejia, R., Vesal, M., Weinhart, T., Post, M., +& Ostanin, I. (2023, January). Simulating industrial scenarios: With +the open-source software MercuryDPM. +https://doi.org/10.23967/c.particles.2023.015 + + + GrainLearning + Cheng + 10.5281/zenodo.8352544 + 2023 + Cheng, H., Orozco, L., Lubbe, R., +Jansen, A., Hartmann, P., & Thoeni, K. (2023). GrainLearning +(Version v2.0.2). Zenodo. +https://doi.org/10.5281/zenodo.8352544 + + + A calibration framework for discrete element +model parameters using genetic algorithms + Do + Advanced Powder Technology + 29 + 10.1016/J.APT.2018.03.001 + 0921-8831 + 2018 + Do, H. Q., Aragón, A. M., & +Schott, D. L. (2018). A calibration framework for discrete element model +parameters using genetic algorithms. Advanced Powder Technology, 29, +1393–1403. +https://doi.org/10.1016/J.APT.2018.03.001 + + + Application of taguchi methods to DEM +calibration of bonded agglomerates + Hanley + Powder Technology + 210 + 10.1016/j.powtec.2011.03.023 + 2011 + Hanley, K. J., O’Sullivan, C., +Oliveira, J. C., Cronin, K., & Byrne, E. P. (2011). Application of +taguchi methods to DEM calibration of bonded agglomerates. Powder +Technology, 210, 230–240. +https://doi.org/10.1016/j.powtec.2011.03.023 + + + Application of DEM-based metamodels in bulk +handling equipment design: Methodology and DEM case +study + Fransen + Powder Technology + 393 + 10.1016/J.POWTEC.2021.07.048 + 0032-5910 + 2021 + Fransen, M. P., Langelaar, M., & +Schott, D. L. (2021). Application of DEM-based metamodels in bulk +handling equipment design: Methodology and DEM case study. Powder +Technology, 393, 205–218. +https://doi.org/10.1016/J.POWTEC.2021.07.048 + + + Identification of DEM simulation parameters +by artificial neural networks and bulk experiments + Benvenuti + Powder Technology + 291 + 10.1016/j.powtec.2016.01.003 + 2016 + Benvenuti, L., Kloss, C., & +Pirker, S. (2016). Identification of DEM simulation parameters by +artificial neural networks and bulk experiments. Powder Technology, 291, +456–465. +https://doi.org/10.1016/j.powtec.2016.01.003 + + + + + + diff --git a/joss.06338/10.21105.joss.06338.jats b/joss.06338/10.21105.joss.06338.jats new file mode 100644 index 0000000000..85ff64a22e --- /dev/null +++ b/joss.06338/10.21105.joss.06338.jats @@ -0,0 +1,499 @@ + + +
+ + + + +Journal of Open Source Software +JOSS + +2475-9066 + +Open Journals + + + +6338 +10.21105/joss.06338 + +GrainLearning: A Bayesian uncertainty quantification +toolbox for discrete and continuum numerical models of granular +materials + + + +https://orcid.org/0000-0001-7652-8600 + +Cheng +Hongyang + + + + +https://orcid.org/0000-0002-9153-650X + +Orozco +Luisa + + +* + + + +Lubbe +Retief + + + + +https://orcid.org/0000-0002-4764-9347 + +Jansen +Aron + + + + +https://orcid.org/0000-0002-2524-8024 + +Hartmann +Philipp + + + + +https://orcid.org/0000-0001-7351-7447 + +Thoeni +Klaus + + + + + +Netherlands eScience center, Amsterdam, The +Netherlands + + + + +Soil Micro Mechanics (SMM), Faculty of Engineering +Technology, MESA+, University of Twente, Enschede, The +Netherlands + + + + +University of Newcastle, Callaghan, NSW, +Australia + + + + +* E-mail: + + +13 +1 +2024 + +9 +97 +6338 + +Authors of papers retain copyright and release the +work under a Creative Commons Attribution 4.0 International License (CC +BY 4.0) +2022 +The article authors + +Authors of papers retain copyright and release the work under +a Creative Commons Attribution 4.0 International License (CC BY +4.0) + + + +Bayesian inference +Calibration +Discrete element method +Granular materials +Uncertainty Quantification +Multi-particle simulation + + + + + + Summary +

How to keep dikes safe with rising sea levels? Why are ripples + formed in sand? What can we prepare for landing on Mars? At the center + of these questions is the understanding of how the grains, as a + self-organizing material, collide, flow, or get jammed and compressed. + State-of-the-art algorithms allow for simulating millions of grains + individually in a computer. However, such computations can take very + long and produce complex data difficult to interpret and be upscaled + to large-scale applications such as sediment transport and debris + flows. GrainLearning is an open-source toolbox with machine learning + and statistical inference modules allowing for emulating granular + material behavior and learning material uncertainties from real-life + observations.

+

To understand what GrainLearning does, let us consider a mechanical + test performed on a granular material. The macroscopic response of + such material, in terms of stress-strain evolution curves, is obtained + from the test. It would be interesting to have a digital equivalent + material to further investigate, using numerical simulations such as + the discrete element method (DEM), how such material would behave + under other mechanical constraints. To do so, the first step is + defining a contact model governing interactions between grains in DEM. + This involves multiple a-priori unknown constants, such as friction + coefficients or Young’s modulus, whose chosen values will determine + the macroscopic behavior of the simulation. By repeatedly comparing + the simulation results with provided experimental data, GrainLearning + allows one to calibrate or infer these values such that the mechanical + response in the DEM simulation is the closest to that observed in the + real-world experiment.

+

While it was initially developed for DEM simulations of granular + materials, GrainLearning can be extended to other simulation + frameworks such as FEM, CFD, LBM, and even other techniques such as + agent-based modeling. In the same vein, the framework is not exclusive + for granular materials.

+ + + Statement of need +

Understanding the link from particle motions to the macroscopic + material response is essential to develop accurate models for + processes such as 3D printing with metal powders, pharmaceutical + powder compaction, flow and handling of cereals in the alimentary + industry, grinding and transport of construction materials. Discrete + Element Method (DEM) has been used widely as the fundamental tool to + produce the data to understand such link. However, DEM simulations are + highly computationally intensive and some of the parameters used in + the contact laws cannot be directly measured experimentally.

+

GrainLearning + (Cheng + et al., 2023) arises as a tool for Bayesian calibration of such + computational models, which means the model parameters are estimated + with a certain level of uncertainty, constrained on (noisy) real-world + observations. Effectively, this makes the simulations digital twins of + real-world processes with uncertainties propagated on model outputs, + which then can be used for optimization or decision-making.

+

Conventionally, the calibration of contact parameters at the grain + scale is accomplished by trial and error, by comparing the macroscopic + responses between simulation and experiments. This is due to the + difficulty of obtaining precise measurements at the contact level and + the randomness of grain properties (e.g., shape, stiffness, and + asphericity). In the last decade, optimization + (Do et + al., 2018) and design-of-experiment + (Hanley + et al., 2011) approaches such as Latin Hypercube sampling and + genetic algorithms have been used. However, the amount of model runs + is still too large. For this reason, Gaussian process regression + (Fransen + et al., 2021) or artificial neural networks + (Benvenuti + et al., 2016) were tested as surrogate- or meta-models for the + DEM. GrainLearning combines probabilistic learning of parameter space + and sampling to achieve Bayesian optimization efficiently.

+ + + Functionality +

GrainLearning’s core functionality is illustrated in + [fig:calibration_diagram]. + GrainLearning started in the geotechnical engineering community and + was primarily used for granular materials in quasi-static, laboratory + conditions + (Cheng + et al., 2018, + 2019). + These include triaxial + (Hartmann + et al., 2022; + Li + et al., 2024) and oedometric + (Cheng + et al., 2019) compressions of soil samples. In the particle + technology community, attempts with GrainLearning have been made to + identify contact parameters for polymer and pharmaceutical powders + against angle-of-repose + (Nguyen, + 2022), shear cell + (Thornton + et al., 2023), and sintering experiments + (Alvarez + et al., 2022). Satisfactory results have been obtained in + simulation cases where the grains were in dynamic regimes or treated + under multi-physical processes.

+ + +

Calibration or parameter inference: By means of + Sequential Monte Carlo filtering GrainLearning can infer and + update model parameters. By learning the underlying distribution + using a variational Gaussian model, highly probable zones are + identified and sampled iteratively until a tolerance for the + overall uncertainty is reached. This process requires the input + of: a time series reference data, the ranges of the parameters to + infer and a tolerance. The software iteratively minimizes the + discrepancy between the model solution and the reference data.

+ + + +

Elements of the trade in the calibration process. 1. + Draw initial values of the parameters to calibrate. 2. Run the + dynamic system with the parameters. 3. With the reference data or + observation, estimate the posterior distribution via the Bayesian + filtering. 4. Check convergence of the parameter inference, if the + process has not converged: 5. Define a Gaussian mixture from the + examples of this iteration and sample the parameters for the next + iteration. 6. Next iteration step. For more details check + the + iterative Bayesian filter section of GrainLearning’s + documentation. +

+ + + + +

Surrogate modeling: Besides using direct + simulation results (e.g. DEM) GrainLearning offers the capability + of building surrogates (e.g. recurrent neural networks) as an + alternative to computationally expensive DEM simulations, + effectively reducing the cost by several orders of magnitude.

+ + + + + Acknowledgements +

The last author would like to thank the Netherlands eScience Center + for the funding provided under grant number NLESC.OEC.2021.032.

+ + + Author contributions +

All authors have contributed substantially to the development of + GrainLearning. H. Cheng was responsible for the main idea and is the + main contributor. R. Lubbe and H. Cheng designed the code structure, + including the inference and sampling modules and tutorials and + examples to facilitate understanding. A. Jansen and L. Orozco + contributed to the conceptualization, implementation, testing and + documentation of the machine learning module, as well as the + improvement of best software practices.

+ + + + + + + + ChengH. + ShukuTakayuki + ThoeniKlaus + TemponePamela + LudingStefan + MagnanimoVanessa + + An iterative bayesian filtering framework for fast and automated calibration of DEM models + Computer Methods in Applied Mechanics and Engineering + 2019 + 350 + 0045-7825 + https://www.sciencedirect.com/science/article/pii/S0045782519300520 + 10.1016/j.cma.2019.01.027 + 268 + 294 + + + + + + ChengH. + ShukuT. + ThoeniK. + YamamotoH. + + Probabilistic calibration of discrete element simulations using the sequential quasi-monte carlo filter + Granular Matter + 2018 + 20 + 10.1007/s10035-017-0781-y + + + + + + HartmannP. + ChengH. + ThoeniK. + + Performance study of iterative bayesian filtering to develop an efficient calibration framework for DEM + Computers and Geotechnics + Elsevier Ltd + 2022 + 141 + 10.1016/j.compgeo.2021.104491 + + + + + + AlvarezJ. E. + SnijderH. + VanekerT. + ChengH. + ThorntonA. R. + LudingS. + WeinhartT. + + Visco-elastic sintering kinetics in virgin and aged polymer powders + Powder Technology + 2022 + 397 + 0032-5910 + https://www.sciencedirect.com/science/article/pii/S0032591021009980 + 10.1016/j.powtec.2021.11.044 + 117000 + + + + + + + NguyenQ. H. + + Machine learning in the calibration process of discrete particle model + 2022 + http://essay.utwente.nl/91991/ + + + + + + LiXin + KouretzisGeorge + ThoeniKlaus + + Discrete element modelling of uplift of rigid pipes deeply buried in dense sand + Computers and Geotechnics + 2024 + 166 + 0266-352X + https://www.sciencedirect.com/science/article/pii/S0266352X23007140 + 10.1016/j.compgeo.2023.105957 + 105957 + + + + + + + ThorntonAnthony + NguyenQ. + PolmanH. + BisschopJ. + Weinhart-MejiaR. + VesalM. + WeinhartThomas + PostM. + OstaninIgor + + Simulating industrial scenarios: With the open-source software MercuryDPM + 202301 + 10.23967/c.particles.2023.015 + + + + + + + + ChengH. + OrozcoL. + LubbeR. + JansenA. + HartmannP. + ThoeniK. + + GrainLearning + Zenodo + 202309 + https://doi.org/10.5281/zenodo.8352544 + 10.5281/zenodo.8352544 + + + + + + DoHuy Q. + AragónAlejandro M. + SchottDingena L. + + A calibration framework for discrete element model parameters using genetic algorithms + Advanced Powder Technology + Elsevier + 201806 + 29 + 0921-8831 + https://www.sciencedirect.com/science/article/pii/S0921883118300773?via%3Dihub + 10.1016/J.APT.2018.03.001 + 1393 + 1403 + + + + + + HanleyKevin J. + O’SullivanCatherine + OliveiraJorge C. + CroninKevin + ByrneEdmond P. + + Application of taguchi methods to DEM calibration of bonded agglomerates + Powder Technology + 201107 + 210 + http://www.sciencedirect.com/science/article/pii/S0032591011001380 + 10.1016/j.powtec.2011.03.023 + 230 + 240 + + + + + + FransenMarc P. + LangelaarMatthijs + SchottDingena L. + + Application of DEM-based metamodels in bulk handling equipment design: Methodology and DEM case study + Powder Technology + Elsevier + 202111 + 393 + 0032-5910 + 10.1016/J.POWTEC.2021.07.048 + 205 + 218 + + + + + + BenvenutiL. + KlossC. + PirkerS. + + Identification of DEM simulation parameters by artificial neural networks and bulk experiments + Powder Technology + 201604 + 291 + http://www.sciencedirect.com/science/article/pii/S003259101630002X + 10.1016/j.powtec.2016.01.003 + 456 + 465 + + + + +
diff --git a/joss.06338/10.21105.joss.06338.pdf b/joss.06338/10.21105.joss.06338.pdf new file mode 100644 index 0000000000..bfd23c1dd7 Binary files /dev/null and b/joss.06338/10.21105.joss.06338.pdf differ diff --git a/joss.06338/media/calibration_diagram.png b/joss.06338/media/calibration_diagram.png new file mode 100644 index 0000000000..9933f677a2 Binary files /dev/null and b/joss.06338/media/calibration_diagram.png differ