Predicting spatiotemporal coke growth in an industrial ethane cracker with a hybrid mechanistic–neural surrogate of its plant-tuned SPYRO model
5 affiliations
- 1Tianjin Haitai Testing Technology Co., Ltd.
- 2Refining, Petrochemical and New Materials Business Unit
- 3SINOPEC
- 4School of Astronautics
- 5Beihang University
Abstract
Coke accumulation on radiant coils sets the run length of steam-cracking furnaces, yet the thickness field cannot be measured in service, and rigorous simulators are proprietary and non-differentiable. We introduce a grey-box surrogate of a plant-calibrated SPYRO model for an industrial ethane cracker, forecasting the complete coke field across 52 axial segments over a 90-day run. It learns the coking rate as an explicit Arrhenius core modulated by a positive neural component and coke self-feedback, integrated over time to inherently ensure monotonic growth. On six held-out cases outside the training window in at least two operating variables, it reaches 0.034 ± 0.001 mm mean absolute error (R² = 0.9994, five seeds) with zero monotonicity violations; ablations show the guarantee costs no accuracy. Plant infrared data reproduce the encoded temperature trend. We close with limitations (unidentifiable embedded kinetics, narrow validity window, exogenous wall temperature) and demonstrate an online-estimator deployment using the analytic Jacobian.
Related
- A hybrid Fourier neural operator–convolutional surrogate for predicting geometrically necessary dislocation density in polycrystalline aluminum
- Mechanistic Framework for Multicomponent Nanoparticle Assembly: Predicting RNA-lipid and PEI-DNA nanoparticle assembly
- Mechanism-informed dynamic surrogate modeling for rapid thermal-aberration prediction in bonded optical systems
- A hybrid GAN-CNN surrogate framework for wave-structure interaction: Coupled prediction of free-surface fields and hydrodynamic forces
- Koopman Neural Operator-based surrogate modelling for high-frequency dynamics of high-speed rail pantographs
BibTeX
@misc{li2026predicting,
title = {{Predicting spatiotemporal coke growth in an industrial ethane cracker with a hybrid mechanistic–neural surrogate of its plant-tuned SPYRO model}},
author = {Hongtao Li and Hao Zhou and Jinshen Wang and Xiaochen Sun},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15009054/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009054/v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Predicting spatiotemporal coke growth in an industrial ethane cracker with a hybrid mechanistic–neural surrogate of its plant-tuned SPYRO model
AU - Hongtao Li
AU - Hao Zhou
AU - Jinshen Wang
AU - Xiaochen Sun
PY - 2026
DA - 2026/09/18
PB - ChemRxiv
DO - 10.26434/chemrxiv.15009054/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009054/v1
AB - Coke accumulation on radiant coils sets the run length of steam-cracking furnaces, yet the thickness field cannot be measured in service, and rigorous simulators are proprietary and non-differentiable. We introduce a grey-box surrogate of a plant-calibrated SPYRO model for an industrial ethane cracker, forecasting the complete coke field across 52 axial segments over a 90-day run. It learns the coking rate as an explicit Arrhenius core modulated by a positive neural component and coke self-feedback, integrated over time to inherently ensure monotonic growth. On six held-out cases outside the training window in at least two operating variables, it reaches 0.034 ± 0.001 mm mean absolute error (R² = 0.9994, five seeds) with zero monotonicity violations; ablations show the guarantee costs no accuracy. Plant infrared data reproduce the encoded temperature trend. We close with limitations (unidentifiable embedded kinetics, narrow validity window, exogenous wall temperature) and demonstrate an online-estimator deployment using the analytic Jacobian.
ER -