pith. sign in

arxiv: 1907.05928 · v1 · pith:HQ6L2747new · submitted 2019-07-12 · ⚛️ physics.data-an · cs.LG· physics.app-ph

A machine learning framework for computationally expensive transient models

Prashant Kumar , Kushal Sinha , Nandkishor Nere , Yujin Shin , Raimundo Ho , Ahmad Sheikh , Laurie Mlinar This is my paper

Pith reviewed 2026-05-24 22:15 UTC · model grok-4.3

classification ⚛️ physics.data-an cs.LGphysics.app-ph
keywords machine learningdiscrete element methodARIMAtransient simulationscientific computingensemble modelingtime series forecasting
0
0 comments X

The pith

Short DEM runs plus ARIMA and machine learning reproduce long-term transient dynamics at far lower cost.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes an ensemble method that runs the discrete element method only over short intervals, then uses ARIMA to forecast the next interval and a machine-learning model to correct the forecast. This hybrid keeps the accuracy of the full first-principles transient model while cutting the total compute time. A sympathetic reader would care because many industrial and scientific processes involve long-time dynamics that remain out of reach for direct simulation. If the approach holds, it makes previously intractable time scales or parameter sweeps feasible on ordinary hardware.

Core claim

The ensemble that interleaves limited discrete element method runs with ARIMA time-series forecasting and a trained machine-learning correction produces predictions in good agreement with literature results for the tested systems, while substantially lowering the computational burden of the original transient model.

What carries the argument

The ensemble that combines short discrete element method segments, ARIMA forecasting, and machine-learning correction to extend simulation length.

If this is right

  • The same hybrid structure can be attached to other expensive transient simulators.
  • Time horizons that are currently prohibitive become reachable with the same hardware budget.
  • Prediction accuracy remains comparable to the original first-principles model in the cases examined.
  • Process modeling and design studies gain a practical route to longer or higher-resolution transients.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the method proves robust across operating regimes, it could reduce reliance on large-scale parallel computing for routine transient studies.
  • Analogous hybrids could be tested on molecular dynamics or computational fluid dynamics codes that face similar cost barriers.
  • Periodic full-model checkpoints could serve as an online monitor for undetected error growth.

Load-bearing premise

Short segments of the full model plus statistical forecasting and learning suffice to capture the essential long-term behavior without accumulating errors that distort results at later times or different conditions.

What would settle it

A side-by-side run showing that hybrid predictions diverge systematically from full discrete element method results over longer times or under changed operating conditions would falsify the claim of retained accuracy.

read the original abstract

The promise of machine learning has been explored in a variety of scientific disciplines in the last few years, however, its application on first-principles based computationally expensive tools is still in nascent stage. Even with the advances in computational resources and power, transient simulations of large-scale dynamic systems using a variety of the first-principles based computational tools are still limited. In this work, we propose an ensemble approach where we combine one such computationally expensive tool, called discrete element method (DEM), with a time-series forecasting method called auto-regressive integrated moving average (ARIMA) and machine-learning methods to significantly reduce the computational burden while retaining model accuracy and performance. The developed machine-learning model shows good predictability and agreement with the literature, demonstrating its tremendous potential in scientific computing.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes an ensemble framework combining the discrete element method (DEM) for first-principles transient simulations with ARIMA time-series forecasting and machine-learning methods. The central claim is that this hybrid approach substantially reduces computational cost while retaining model accuracy, as evidenced by the developed ML model showing good predictability and agreement with the literature.

Significance. If the hybrid surrogate can be shown to preserve long-term accuracy without accumulating systematic errors, the framework would enable extended transient simulations that are currently intractable with pure DEM, offering a practical route to accelerate first-principles modeling in granular and dynamic systems.

major comments (2)
  1. [Abstract] Abstract: the assertion of 'good predictability and agreement with the literature' supplies no quantitative error metrics (e.g., RMSE, MAE), validation protocol, or baseline comparisons, making it impossible to judge whether post-hoc tuning or data selection affects the central claim of retained accuracy.
  2. [Results/Discussion] The central claim requires that short DEM runs plus ARIMA/ML faithfully capture long-term dynamics without systematic errors that grow over time or across operating conditions. No quantitative check is described on whether prediction residuals remain bounded as the forecast horizon increases, nor any cross-condition validation that would rule out drift or regime-specific bias.
minor comments (1)
  1. [Abstract] The abstract contains a minor grammatical issue ('however,' should be capitalized at the start of the sentence).

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments highlight important aspects of clarity and validation that we address below. We agree that strengthening the quantitative presentation will improve the work and plan revisions accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the assertion of 'good predictability and agreement with the literature' supplies no quantitative error metrics (e.g., RMSE, MAE), validation protocol, or baseline comparisons, making it impossible to judge whether post-hoc tuning or data selection affects the central claim of retained accuracy.

    Authors: We agree that the abstract lacks specific quantitative metrics. In the revised manuscript we will expand the abstract to include explicit error metrics (RMSE, MAE) from the ML model validation, a brief description of the validation protocol (train/test split on DEM-generated time series), and comparison against a pure ARIMA baseline. These additions will be drawn from the existing results without altering the underlying data or claims. revision: yes

  2. Referee: [Results/Discussion] The central claim requires that short DEM runs plus ARIMA/ML faithfully capture long-term dynamics without systematic errors that grow over time or across operating conditions. No quantitative check is described on whether prediction residuals remain bounded as the forecast horizon increases, nor any cross-condition validation that would rule out drift or regime-specific bias.

    Authors: The referee correctly notes that explicit checks for residual boundedness over long horizons and cross-condition validation are not presented. We will add a new subsection in Results that plots prediction residuals versus forecast horizon for multiple DEM runs, demonstrating that errors remain bounded within the reported MAE range. We will also include a cross-condition test using an additional operating point (different particle size or velocity) held out from training to address potential regime-specific bias. These analyses use the same trained models and will be reported with the existing dataset. revision: yes

Circularity Check

0 steps flagged

No circularity: standard application of ARIMA+ML surrogate to short DEM runs

full rationale

The paper describes an ensemble workflow that runs short DEM simulations, fits ARIMA and ML models on those outputs, and uses the surrogate for longer-term forecasting. No equations, uniqueness theorems, or self-citations are invoked to derive the long-term behavior from the short-run data by construction. The central claim is an empirical demonstration that the hybrid surrogate retains accuracy, which is an external validation task rather than a definitional reduction. No load-bearing step collapses to a fitted parameter renamed as a prediction or to a self-citation chain.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available, so the ledger is necessarily incomplete; typical free parameters would include ARIMA orders (p,d,q) and ML hyperparameters fitted to DEM output, but none are enumerated here.

pith-pipeline@v0.9.0 · 5677 in / 993 out tokens · 16591 ms · 2026-05-24T22:15:01.156765+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

38 extracted references · 38 canonical work pages

  1. [1]

    M., Rosario, B

    Oliver, N. M., Rosario, B. & Pentland, A. P. A Bayesian computer vision system for modeling human interactions. IEEE Trans. Pattern Anal. Mach. Intell. 22, 831–843 (2000)

    work page 2000

  2. [2]

    & Weston, J

    Collobert, R. & Weston, J. A unified architecture for natural language processing. in Proceedings of the 25th international conference on Machine learning - ICML ’08 160–167 (ACM Press, 2008). doi:10.1145/1390156.1390177

    work page doi:10.1145/1390156.1390177 2008

  3. [3]

    Bojarski, M. et al. End to End Learning for Self-Driving Cars. (2016)

    work page 2016

  4. [4]

    Machine-learning approaches in drug discovery: methods and applications

    Lavecchia, A. Machine-learning approaches in drug discovery: methods and applications. Drug Discov. Today 20, 318–331 (2015)

    work page 2015

  5. [5]

    Brockherde, F. et al. Bypassing the Kohn-Sham equations with machine learning. Nat. Commun. 8, (2017)

    work page 2017

  6. [6]

    & Hafner, J

    Kresse, G. & Hafner, J. Ab initio molecular dynamics for liquid metals. Phys. Rev. B 47, 558–561 (1993)

    work page 1993

  7. [7]

    (Springer Berlin Heidelberg, 2009)

    Computational Fluid Dynamics. (Springer Berlin Heidelberg, 2009). doi:10.1007/978-3-540- 85056-4

    work page doi:10.1007/978-3-540- 2009

  8. [8]

    W., Yang, W

    Ayers, P. W., Yang, W. & Yang, W. Density-Functional Theory. 103–132 (2003). doi:10.1201/9780203913390-9

    work page doi:10.1201/9780203913390-9 2003

  9. [9]

    The combined finite-discrete element method

    Munjiza, A. The combined finite-discrete element method. (Wiley, 2004)

    work page 2004

  10. [10]

    Hughes, T. J. R. The Finite Element Method: Linear Static and Dynamic Finite Element Analysis. Dover Publications Inc., Mineola, New York (2000)

    work page 2000

  11. [11]

    & Pantelides, C

    Bezzo, F., Macchietto, S. & Pantelides, C. C. General hybrid multizonal/CFD approach for bioreactor modeling. AIChE J. 49, 2133–2148 (2003)

    work page 2003

  12. [12]

    Vrábel, P. et al. CMA: integration of fluid dynamics and microbial kinetics in modelling of large- scale fermentations. Chem. Eng. J. 84, 463–474 (2001)

    work page 2001

  13. [13]

    H., Stephens, D

    Cooke, M. H., Stephens, D. J. & Bridgwater, J. Powder mixing — a literature survey. Powder Technol. 15, 1–20 (1976)

    work page 1976

  14. [14]

    Powder Technol

    Powder mixing: Some practical rules applied to agitated systems. Powder Technol. 68, 213–234 (1991)

    work page 1991

  15. [15]

    A., Jia, X

    Williams, R. A., Jia, X. & McKee, S. L. Development of slurry mixing models using resistance tomography. Powder Technol. 87, 21–27 (1996)

    work page 1996

  16. [16]

    & Ramachandran, R

    Sen, M. & Ramachandran, R. A multi-dimensional population balance model approach to continuous powder mixing processes. Adv. Powder Technol. 24, 51–59 (2013)

    work page 2013

  17. [17]

    & Ramachandran, R

    Sen, M., Dubey, A., Singh, R. & Ramachandran, R. Mathematical Development and Comparison of a Hybrid PBM-DEM Description of a Continuous Powder Mixing Process. J. Powder Technol. 2013, 1–11 (2013)

    work page 2013

  18. [18]

    Chaudhuri, B., Mehrotra, A., Muzzio, F. J. & Tomassone, M. S. Cohesive effects in powder mixing in a tumbling blender. Powder Technol. 165, 105–114 (2006)

    work page 2006

  19. [19]

    Conder, E. W. et al. The Pharmaceutical Drying Unit Operation: An Industry Perspective on Advancing the Science and Development Approach for Scale-Up and Technology Transfer. Org. Process Res. Dev. 21, 420–429 (2017)

    work page 2017

  20. [20]

    Fundamental powder mixing mechanisms

    Bridgwater, J. Fundamental powder mixing mechanisms. Powder Technol. 15, 215–236 (1976)

    work page 1976

  21. [21]

    & Marziano, I

    Birch, M. & Marziano, I. Understanding and Avoidance of Agglomeration During Drying Processes: A Case Study. Org. Process Res. Dev. 17, 1359–1366 (2013)

    work page 2013

  22. [22]

    (University of B

    Hoffmann, H. (University of B. Simple violin plot using matlab default kernel density estimation. (2015)

    work page 2015

  23. [23]

    L., Davies, M., Ingram, A

    Marigo, M., Cairns, D. L., Davies, M., Ingram, A. & Stitt, E. H. A numerical comparison of mixing efficiencies of solids in a cylindrical vessel subject to a range of motions. Powder Technol. 217, 540–547 (2012)

    work page 2012

  24. [24]

    & Dennehy, R

    Hare, C., Ghadiri, M. & Dennehy, R. Prediction of attrition in agitated particle beds. Chem. Eng. Sci. 66, 4757–4770 (2011)

    work page 2011

  25. [25]

    Box, G. E. P., Jenkins, G. M., Reinsel, G. C. & Ljung, G. M. Time series analysis : forecasting and control. (Wiley, 2016)

    work page 2016

  26. [26]

    & Lin, C.-S

    Pai, P.-F. & Lin, C.-S. A hybrid ARIMA and support vector machines model in stock price forecasting. Omega 33, 497–505 (2005)

    work page 2005

  27. [27]

    Model selection and Akaike’s Information Criterion (AIC): The general theory and its analytical extensions

    Bozdogan, H. Model selection and Akaike’s Information Criterion (AIC): The general theory and its analytical extensions. Psychometrika 52, 345–370 (1987)

    work page 1987

  28. [28]

    A., Irrthum, A., Wehenkel, L

    Huynh-Thu, V. A., Irrthum, A., Wehenkel, L. & Geurts, P. Inferring regulatory networks from expression data using tree-based methods. PLoS One 5, e12776 (2010)

    work page 2010

  29. [29]

    T., Estrada, J

    Ahneman, D. T., Estrada, J. G., Lin, S., Dreher, S. D. & Doyle, A. G. Predicting reaction performance in C-N cross-coupling using machine learning. Science 360, 186–190 (2018)

    work page 2018

  30. [30]

    Dubey, A., Sarkar, A., Ierapetritou, M., Wassgren, C. R. & Muzzio, F. J. Computational Approaches for Studying the Granular Dynamics of Continuous Blending Processes, 1 - DEM Based Methods. Macromol. Mater. Eng. 296, 290–307 (2011)

    work page 2011

  31. [31]

    & Gonzalez, M

    Liu, Y., Gonzalez, M., Wassgren, C. & Gonzalez, M. Modeling granular material blending in a rotating drum using a finite element method and advection-diffusion equation multi-scale model

    work page

  32. [32]

    ARIMA Time Series Data Forecasting and Visualization in Python | DigitalOcean

    Thomas Vincent. ARIMA Time Series Data Forecasting and Visualization in Python | DigitalOcean. (2017). Available at: https://www.digitalocean.com/community/tutorials/a-guide-to- time-series-forecasting-with-arima-in-python-3. (Accessed: 15th November 2017)

    work page 2017

  33. [33]

    Anaconda Software Distribution. (2016)

    work page 2016

  34. [34]

    & Hastie, T

    Zou, H. & Hastie, T. Regularization and variable selection via the elastic net. J. R. Stat. Soc. Ser. B (Statistical Methodol. 67, 301–320 (2005)

    work page 2005

  35. [35]

    & Patranabis, D

    Basak, D., Pal, S. & Patranabis, D. C. Support Vector Regression. Neural Inf. Process. – Lett. Rev. 11, (2007)

    work page 2007

  36. [36]

    & Kowalski, B

    Geladi, P. & Kowalski, B. R. Partial least-squares regression: a tutorial. Anal. Chim. Acta 185, 1– 17 (1986)

    work page 1986

  37. [37]

    Random Forest: A Classification and Regression Tool for Compound Classification and QSAR Modeling

    Vladimir Svetnik, *,† et al. Random Forest: A Classification and Regression Tool for Compound Classification and QSAR Modeling. (2003). doi:10.1021/CI034160G

    work page doi:10.1021/ci034160g 2003

  38. [38]

    Pedregosa, F. et al. Scikit-learn: Machine Learning in Python. J. Mach. Learn. Res. 12, 2825–2830 (2011). Figures Figure 1: Flowchart of steps involved in applying machine -learning to computationally expensive high-fidelity scientific models. Availability to high -quality data is key to develop a good machine learning predictive model. Data transformatio...

    work page 2011