Surrogate Modeling and Explainable Artificial Intelligence for Complex Systems: A Workflow for Automated Simulation Exploration
Abstract
Complex systems are increasingly explored through simulation-driven engineering workflows that combine physics-based and empirical models with optimization and analytics. Despite their power, these workflows face two central obstacles: a high computational cost, since accurate exploration requires many expensive simulator runs; and limited transparency and reliability when decisions rely on opaque blackbox components. We propose a workflow that addresses both challenges by training lightweight emulators on compact designs of experiments that (i) provide fast, low-latency approximations of expensive simulators, (ii) enable rigorous uncertainty quantification, and (iii) are adapted for global and local eXplainable Artificial Intelligence (XAI) analyses. This paper formalizes this integration by providing a methodological bridge that handles continuous, categorical, and integer variables across diverse simulation domains, from engineering design to socio-environmental agent-based models. To validate surrogate explanations and analyze how they reflect simulator behavior, we introduce a comparative methodology incorporating active validation filters, such as explanation consistency and aleatoric uncertainty quantification. Furthermore, this study provides a practical approach for coupling designs of experiments with emulator training, offering practical guidelines for selecting surrogate architectures that balance computational efficiency with global and local interpretability. We illustrate our framework's applicability across two contrasting paradigms: a deterministic multidisciplinary design simulation of a hybrid-electric aircraft and a stochastic agent-based simulation of urban segregation. Results show that the coupling of surrogate with XAI enables large-scale exploration in seconds, helps identify nonlinear interactions, and highlights potential design or policy levers. By evaluating explanation consistency across multiple surrogate architectures, this workflow contributes to scalable and explainability-driven machine learning for complex-system exploration.
Related
- Artificial intelligence-based surrogate modeling of hemodynamic indicators in abdominal aortic aneurysms using computational fluid dynamics
- Causal Inference with Generative Artificial Intelligence: Application to Texts as Treatments
- The Akaike Information Criterion and Artificial Intelligence
- Questionable mediated effect of artificial intelligence anxiety on academic burnout: A comment on Long et al. (2026)
- Buyer Artificial Intelligence-Enabled Environmental Governance and Supplier Environmental Controversies: An Organizational Information Processing and Signaling
BibTeX
@misc{saves2026surrogate,
title = {{Surrogate Modeling and Explainable Artificial Intelligence for Complex Systems: A Workflow for Automated Simulation Exploration}},
author = {Paul Saves and Pramudita Satria Palar and Muhammad Robani and Nicolas Verstaevel and Moncef Garouani and Jordan Levy and Julien Aligon and Benoit Gaudou and Koji Shimoyama and Joseph Morlier},
howpublished = {HAL},
year = {2026},
month = oct,
eprint = {2510.16742},
archivePrefix = {arXiv},
url = {https://hal.science/hal-05321914v2},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Surrogate Modeling and Explainable Artificial Intelligence for Complex Systems: A Workflow for Automated Simulation Exploration
AU - Paul Saves
AU - Pramudita Satria Palar
AU - Muhammad Robani
AU - Nicolas Verstaevel
AU - Moncef Garouani
AU - Jordan Levy
AU - Julien Aligon
AU - Benoit Gaudou
AU - Koji Shimoyama
AU - Joseph Morlier
PY - 2026
DA - 2026/10/31
PB - HAL
UR - https://hal.science/hal-05321914v2
AB - Complex systems are increasingly explored through simulation-driven engineering workflows that combine physics-based and empirical models with optimization and analytics. Despite their power, these workflows face two central obstacles: a high computational cost, since accurate exploration requires many expensive simulator runs; and limited transparency and reliability when decisions rely on opaque blackbox components. We propose a workflow that addresses both challenges by training lightweight emulators on compact designs of experiments that (i) provide fast, low-latency approximations of expensive simulators, (ii) enable rigorous uncertainty quantification, and (iii) are adapted for global and local eXplainable Artificial Intelligence (XAI) analyses. This paper formalizes this integration by providing a methodological bridge that handles continuous, categorical, and integer variables across diverse simulation domains, from engineering design to socio-environmental agent-based models. To validate surrogate explanations and analyze how they reflect simulator behavior, we introduce a comparative methodology incorporating active validation filters, such as explanation consistency and aleatoric uncertainty quantification. Furthermore, this study provides a practical approach for coupling designs of experiments with emulator training, offering practical guidelines for selecting surrogate architectures that balance computational efficiency with global and local interpretability. We illustrate our framework's applicability across two contrasting paradigms: a deterministic multidisciplinary design simulation of a hybrid-electric aircraft and a stochastic agent-based simulation of urban segregation. Results show that the coupling of surrogate with XAI enables large-scale exploration in seconds, helps identify nonlinear interactions, and highlights potential design or policy levers. By evaluating explanation consistency across multiple surrogate architectures, this workflow contributes to scalable and explainability-driven machine learning for complex-system exploration.
ER -