On the Development of a Surrogate Modelling Toolbox for Virtual Assembly

Production from multi-bench unconventional development is time-consuming to model in physics-based simulators, requiring multiple runs. Each simulation takes minutes to hours, even with enhanced cloud computing power. This limits history matching to a single deterministic solution, as multi-scenario algorithms need significantly more simulations. An AI-based surrogate model expedites forecasting, enabling efficient multi-scenario algorithms. The AI surrogate relies on a large dataset of high-fidelity physics-based simulations created using modern high-performance computing (HPC) infrastructure. Simulations were based on Latin-hypercube sampling of over 100 parameters, covering geology, reservoir properties, well placements, and completion designs. Results were processed on HPC to extract features and train a surrogate model using advanced AI techniques. The model was iteratively trained to meet user-defined error metrics. The final surrogate was integrated with an automated history matching (AHM) package using a Gauss-Newton trust region optimizer to obtain multiple scenarios that align with field production. Developing a surrogate model with the desired fidelity was a challenging task. Reservoir models include several highly non-linear input-output relationships from various physical processes. Training a surrogate with an acceptable level of fidelity requires both a massive dataset and specialized AI techniques to enforce physical relationships in deep learning algorithms. The latter was critical to obtain correct physical trends from the surrogate model predictions. The final surrogate model's production forecasts accurately match the trends and magnitudes of the computationally expensive, high-fidelity numerical simulations across several real examples of multi-well and multi-bench developments. The surrogate model integration within the AHM package was straightforward and enabled the efficient determination of multiple scenarios in about one hour for a typical unconventional case. This expedites the determination of plausible scenarios, empowering end-users to better optimize the development plan. A fast reservoir surrogate model significantly reduces the cycle time for using physics-based models and workflows, capturing subsurface uncertainty more holistically. The surrogate model workflow benefits from standard features of machine learning systems, including interpretability and confidence scores, providing the user with richer information before making decisions. In particular, the work demonstrates that determining multiple reservoir scenarios matching production history is practically achievable by leveraging cutting-edge AI technologies.

Surrogate modelling is a powerful tool to replace computationally expensive nonlinear numerical simulations, with fast representations thereof, for inverse analysis, model-based control or optimization. For some problems, it is required that the surrogate model describes a complete output field. To construct such surrogate models, proper orthogonal decomposition (POD) can be used to reduce the dimensionality of the output data. The accuracy of the surrogate models strongly depends on the (pre)processing actions that are used to prepare the data for the dimensionality reduction. In this work, POD-based surrogate models with Radial Basis Function interpolation are used to model high-dimensional FE data fields. The effect of (pre)processing methods on the accuracy of the result field is systematically investigated. Different existing methods for surrogate model construction are compared with a novel method. Special attention is given to data fields consisting of several physical meanings, e.g. displacement, strain and stress. A distinction is made between the errors due to truncation and due to interpolation of the data. It is found that scaling the data per physical part substantially increases the accuracy of the surrogate model.

Virtual trial assembly of large steel members with bolted connections based on point cloud data

Virtual trial assembly of a complex steel structure by Generalized Procrustes Analysis techniques

Micro assembly involves the assembly of micron-sized devices. Given the complexity of this domain, the role of virtual environments becomes important as they provide a basis to propose and compare assembly alternatives virtually prior to physical assembly. This paper proposes an integrated approach which includes the use of virtual reality-based assembly environments that interface with physical micro assembly environments. Such an approach can be an intrinsic part of a collaborative manufacturing framework that seeks to support the rapid assembly of micro devices. In this paper, the design of VREM (Virtual Reality based Environment for Micro Assembly) is discussed which is based on this integrated approach involving use of virtual and physical resources.

Transport of contaminant species undergoing chemical reactions in groundwater aquifers is a complex physical and biochemical process. Simulating this transport process involves solving complex nonlinear equations and requires huge computational time for a given aquifer study area. Development of optimal remediation strategies in aquifers may require repeated solution of such complex numerical simulation models. To overcome this computational limitation and improve the computational feasibility of large number of repeated simulations, Genetic Programming based trained surrogate models are developed to approximately simulate such complex transport processes. Transport process of acid mine drainage, a hazardous pollutant is first simulated using a numerical simulated model: HYDROGEOCHEM 5.0 for a study area resembling a mine site. Simulation model solution results for an illustrative contaminated aquifer site is then approximated by training and testing a Genetic Programming (GP) based surrogate model. To decrease the total number of GP formulations, the coordinates of observation locations are implemented as input data in the surrogate models. Comparison of the surrogate models and numerical simulation results show that the surrogate models can provide acceptable approximations of this complex transport process in contaminated groundwater aquifers.

ABSTRACTLinked simulation-optimisation (S–O) models need to simulate the physical processes either by using a rigorous numerical model, or a trained surrogate model approximating the physical processes. A methodology is proposed to evolve Pareto optimal management strategies for a multi-layered coastal aquifer system using a trained and tested Multivariate Adaptive Regression Spline (MARS) surrogate model linked to a multi-objective saltwater intrusion management model. Performance of the developed methodology is evaluated using an illustrative multi-layered coastal aquifer system. Solution results indicate that MARS is capable of approximately replacing the more rigorous numerical simulation model within the linked S–O model to ensure computational efficiency and feasibility in applying such linked S–O models for coastal aquifer management problems. Furthermore, the ability of MARS to recognise the most relevant input variables in predicting the responses as outputs enables the construction of an efficient and robust surrogate model. Integration of parallel processing capabilities within the optimisation model is shown to improve computational efficiency and feasibly of solving such large scale multi-objective problems. Therefore, the developed methodology utilising the MARS based surrogate model is potentially applicable for developing optimal groundwater extraction strategies for sustainable regional scale management of coastal aquifers.

Understanding and predicting complex environmental and hydrosystem processes is a central challenge in Earth system science. These systems are governed by interacting physical mechanisms across scales, are only partially observed, and are often characterized by limited data and substantial uncertainty. As a result, machine learning (ML) has emerged along two complementary development paths for environmental modeling.In a first branch, physical process models remain the backbone of simulation, while ML is employed as a surrogate to approximate expensive numerical solvers. Surrogate modeling approaches based on Gaussian process emulators, polynomial chaos expansions, support vector regression, and related probabilistic representations are particularly well suited for data-poor settings. Neural networks are used more selectively in this context, as uncertainty-aware and sample-efficient methods are often preferred. In surrogate modeling, considerable effort is devoted to optimal sampling strategies, including active learning, which adaptively select informative simulations and help preserve scarce computational resources. These surrogate models enable efficient uncertainty quantification, sensitivity analysis, and Bayesian inference, while preserving physical interpretability.A second, increasingly important branch emerges when physical models are incomplete, unavailable, or deliberately omitted, and ML models replace the governing equations altogether. This branch is most commonly based on neural network representations, but has recently also been explored using Gaussian processes and polynomial chaos–based learning concepts. In this setting, purely data-driven learning is insufficient, as unconstrained models tend to violate physical principles and extrapolate poorly. In this second branch, physical principles such as conservation laws or balance relations are embedded directly into learning architectures. Complementarily, constraint-driven learning strategies enforce physical laws, admissibility conditions, and structural consistency during training. By restricting the hypothesis space, these methods stabilize learning and support robust inference under incomplete physical knowledge.Taken together, surrogate modeling for physics-based simulations and physics-aware ML for equation-free learning represent two coherent and complementary branches of modern environmental machine learning. We observe a growing convergence between these two branches, as physics-based surrogate modeling and equation-free machine learning increasingly borrow concepts from each other. This convergence is not accidental, but a direct response to fundamental model limitations and the challenge of making reliable predictions under scarce data and knowledge constraints. By integrating physics, probabilistic reasoning and constraints, emerging approaches increasingly focus on robustness and interpretability rather than unconstrained expressive power.

The availability of consumer-facing virtual reality (VR) headsets makes virtual training an attractive alternative to expensive traditional training. Recent works showed that virtually trained workers perform bimanual assembly tasks equally well as ones trained with traditional methods. This paper presents a study that investigated how levels of immersion affect learning transfer between virtual and physical bimanual gearbox assembly tasks. The study used a with-in subject design and examined three different virtual training systems i.e., VR training with direct 3D inputs (HTC VIVE Pro), VR training without 3D inputs (Google Cardboard), and passive video-based training. 23 participants were recruited. The training effectiveness was measured by participant’s performance of assembling 3D-printed copies of the gearboxes in two different timings: immediately after and 2 weeks after the training. The result showed that participants preferred immersive VR training. Surprisingly, despite being less favourable, the subjects’ performance of video-based training were similar to training on HTC VIVE Pro. However, video training led to a significant performance decrease in the retention test session 2 weeks after the training.

Universal design space exploration for building energy design

Intelligent vectorial surrogate modeling framework for multi-objective reliability estimation of aerospace engineering structural systems

In this paper, the study on aerodynamic shape optimization of contra rotating open rotor is presented. Computational fluid dynamics simulations combined with Neural networks (radial basis function) surrogate model is applied to solve the problem. The difficulty is associated with both time-consuming computational fluid dynamics simulations and the large number of design variables involved. The contra rotating open rotor optimization problem is to find the variables that produces the maximum propulsive efficiency for a limited range of thrust requirement. The DOE (deign of experiment) process is completed by uniform design and computational fluid dynamics simulations. The non-linear programming by quadratic lagrangian (NLPQL) is used to tackle the optimization problem. Infill sampling criteria is used in this process to reduce demand on the number of initial samples while keep the accuracy of surrogate model. After building the initial surrogate from DOE sampling, the update point is selected according to the infill sampling criteria. It shows that the infill sampling criteria is helpful in surrogate model building process and surrogate-based optimization method is valid for aerodynamic shape optimization of contra rotating open rotor.

Surrogate models of numerical relativity simulations of merging black holes provide the most accurate tools for gravitational-wave data analysis. Neural network-based surrogates promise evaluation speedups, but their accuracy relies on (often obscure) tuning of settings such as the network architecture, hyperparameters, and the size of the training dataset. We propose a systematic optimization strategy that formalizes setting choices and motivates the amount of training data required. We apply this strategy on NRSur7dq4Remnant, an existing surrogate model for the properties of the remnant of generically precessing binary black hole mergers and construct a neural network version, which we label NRSur7dq4Remnant_NN. The systematic optimization strategy results in a new surrogate model with comparable accuracy and provides insights into the meaning and role of the various network settings and hyperparameters as well as the structure of the physical process. Moreover, NRSur7dq4Remnant_NN results in evaluation speedups of up to 8 times on a single CPU and a further improvement of 2000 times when evaluated in batches on a GPU. To determine the training-set size, we propose an iterative enrichment strategy that efficiently samples the parameter space using much smaller training sets than naive sampling. NRSur7dq4Remnant_NN requires $\mathcal{O}({10}^{4})$ training data, so neural-network-based surrogates are ideal for speeding up models that support such large training datasets, but at the moment cannot directly be applied to numerical relativity catalogs that are $\mathcal{O}({10}^{3})$ in size. The optimization strategy is available through the gwbonsai package.

Two types of angular rotation deviations are observed in our optical CT scanner for gel dosimetry: a large deviation at the starting position, and small deviations during the scanning. In order to investigate the influence of these angular rotation deviations on the reconstructed OD maps, we applied the angular deviations on a synthetic phantom for optical CT simulation. Meanwhile, three reconstruction schemes are proposed for compensation. The simulation results show that the influence of both types of rotation deviation is as small as less than 1% over the central reconstruction phantom region. Based on the reconstructed results via different methods, we can conclude: 1) reconstruction with the regridded projections can better decrease the errors caused by the large angle deviation; 2) for small angle deviations, reconstruction with actual-angle compensation delivers less pronounced results; 3) for real cases, the hybrid approach is shown to reduce reconstruction errors.

The scene graph structure of Fischer-Tropsch synthesis (FTS) was constructed. In the meanwhile, the assembly unit model of equipment was designed. The main FTS reactor and its mathematical physical model were described. Platform for chemical process simulation was established, on base of the virtual assembly for FTS. The method of data transfer in common chemical process simulation was proposed. The composition and flow rate of components, as well as the temperature and pressure of the FTS reactor was investigated by the one dimensional quasi homogeneous model. The relative error is less than 3% between the calculation and experiment. The accuracy of the system process simulation is reliable. It is verified that the simulation method of chemical process is feasible on base of the virtual assembly. This work studies a new way for simulation of the FTS, and also provides technologic supports for the simulation of chemical process based on virtual assembly.