Data-driven Surrogate Model Research Articles

Virtual sensing and virtual metrology for spatial error monitoring of roll-to-roll manufacturing systems

Roll-to-roll (R2R) manufacturing techniques are promising for the high-volume continuous production of substrate-based products for a variety of emerging technologies such as flexible printing electronics, biosensors, thin film batteries. However, meeting the tight tolerance requirements of the R2R printed products has become a major challenge in R2R. This paper focuses on integrating kinematics modelling and data-driven surrogate models to fundamentally understand the spatial error generation and variation propagation mechanism within a multi-step R2R system, and provides a novel method to devise accurate virtual sensing for effective process monitoring, fault detection and diagnosis with limited physical in-situ sensors, and how they can be used to achieve unprecedentedly accurate process control.

Adaptive surrogate model based multiobjective optimization for coastal aquifer management

In this study, a novel surrogate model assisted multiobjective memetic algorithm (SMOMA) is developed for optimal pumping strategies of large-scale coastal groundwater problems. The proposed SMOMA integrates an efficient data-driven surrogate model with an improved non-dominated sorted genetic algorithm-II (NSGAII) that employs a local search operator to accelerate its convergence in optimization. The surrogate model based on Kernel Extreme Learning Machine (KELM) is developed and evaluated as an approximate simulator to generate the patterns of regional groundwater flow and salinity levels in coastal aquifers for reducing huge computational burden. The KELM model is adaptively trained during evolutionary search to satisfy desired fidelity level of surrogate so that it inhibits error accumulation of forecasting and results in correctly converging to true Pareto-optimal front. The proposed methodology is then applied to a large-scale coastal aquifer management in Baldwin County, Alabama. Objectives of minimizing the saltwater mass increase and maximizing the total pumping rate in the coastal aquifers are considered. The optimal solutions achieved by the proposed adaptive surrogate model are compared against those solutions obtained from one-shot surrogate model and original simulation model. The adaptive surrogate model does not only improve the prediction accuracy of Pareto-optimal solutions compared with those by the one-shot surrogate model, but also maintains the equivalent quality of Pareto-optimal solutions compared with those by NSGAII coupled with original simulation model, while retaining the advantage of surrogate models in reducing computational burden up to 94% of time-saving. This study shows that the proposed methodology is a computationally efficient and promising tool for multiobjective optimizations of coastal aquifer managements.

Quantifying Roughness Coefficient Uncertainty in Urban Flooding Simulations through a Simplified Methodology

A methodology is presented which can be used in the evaluation of parametric uncertainty in urban flooding simulation. Due to the fact that such simulations are time consuming, the following methodology is proposed: (a) simplification of the description of the physical process; (b) derivation of a training data set; (c) development of a data-driven surrogate model; (d) use of a forward uncertainty propagation scheme. The simplification comprises the following steps: (a) unit hydrograph derivation using a 2D hydrodynamic model; (b) calculation of the losses in order to determine the effective rainfall depth; (c) flood event simulation using the principle of the proportionality and superposition. The above methodology was implemented in an urban catchment located in the city of Athens, Greece. The model used for the first step of the simplification was FLOW-R2D, whereas the well-known SWMM software (US Environmental Protection Agency, Washington, DC, USA) was used for the second step of the simplification. For the training data set derivation, an ensemble of 100 Unit Hydrographs was derived with the FLOW-R2D model. The parameters which were modified in order to produce this ensemble were the Manning coefficients in the two friction zones (residential and urban open space areas). The surrogate model used to replicate the unit hydrograph derivation, using the Manning coefficients as an input, was based on the Polynomial Chaos Expansion technique. It was found that, although the uncertainties in the derived results have to be taken into account, the proposed methodology can be a fast and efficient way to cope with dynamic flood simulation in an urban catchment.

Application of an Experimentally Validated High-dimensional Surrogate Model to Performance Prediction of a CHP Unit

This research study mainly presents the development of a data-driven surrogate model as a baseline for efficient and intelligent control and monitoring of a micro gas turbine driven combined heat and power system. Micro gas turbines are considered to be an efficient alternative to costly generation and transmission of electricity, especially in remote areas and cogeneration applications where effective tools for remote monitoring and diagnostics would be required for the realization of distributed generation. Given the extensive data obtained during the experimental tests from our gas turbine test rig, an accurate surrogate model is developed in this paper. The predictability of the metamodel is validated against data that is not used during modeling process. A sensitivity analysis is also performed to investigate the dependency between initially selected input and output parameters. The performance of concluding mathematical model is then compared with previously developed and validated optimum artificial neural network. The developed model does not require detailed and exact component characteristics, making it applicable for monitoring and control in a sophisticated smart grid.

An efficient MILP framework for integrating nonlinear process dynamics and control in optimal production scheduling calculations

The emphasis currently placed on enterprise-wide decision making and optimization has led to an increased need for methods of integrating nonlinear process dynamics and control information in scheduling calculations. The inevitable high dimensionality and nonlinearity of first-principles dynamic process models makes incorporating them in scheduling calculations challenging. In this work, we describe a general framework for deriving data-driven surrogate models of the closed-loop process dynamics. Focusing on Hammerstein–Wiener and finite step response (FSR) model forms, we show that these models can be (exactly) linearized and embedded in production scheduling calculations. The resulting scheduling problems are mixed-integer linear programs with a special structure, which we exploit in a novel and efficient solution strategy. A polymerization reactor case study is utilized to demonstrate the merits of this method. Our framework compares favorably to existing approaches that embed dynamics in scheduling calculations, showing considerable reductions in computational effort.

Low-Cost Data-Driven Surrogate Modeling of Antenna Structures by Constrained Sampling

Full-wave electromagnetic analysis has become one of the major design tools for contemporary antenna structures. Although reliable, it is computationally expensive, which makes automated simulation-driven antenna design (e.g., parametric optimization) difficult. This difficulty can be alleviated by utilization of fast and accurate replacement models (surrogates). Unfortunately, conventional data-driven modeling of antennas is usually prohibitively expensive in terms of data acquisition or even infeasible (due to numerical problems with handling large training datasets). In this letter, a technique for reduced-cost antenna modeling is discussed. It exploits constrained sampling within a restricted region of the design space, defined by the reference designs optimized for selected sets of performance requirements. The proposed approach permits dramatic reduction of the sampled portion of the design space without formally reducing geometry parameter ranges. It is demonstrated using two antenna structures and favorably compared to conventional modeling using kriging interpolation.

Data-driven surrogates for rapid simulation and optimization of WAG injection in fractured carbonate reservoirs

Conventional simulation of fractured carbonate reservoirs is computationally expensive because of the multiscale heterogeneities and fracture–matrix transfer mechanisms that must be taken into account using numerical transfer functions and/or detailed models with a large number of simulation grid cells. The computational requirement increases significantly when multiple simulation runs are required for sensitivity analysis, uncertainty quantification and optimization. This can be prohibitive, especially for giant carbonate reservoirs. Yet, sensitivity analysis, uncertainty quantification and optimization are particularly important to analyse, determine and rank the impact of geological and engineering parameters on the economics and sustainability of different Enhanced Oil Recovery (EOR) techniques. We use experimental design to set up multiple simulations of a high-resolution model of a Jurassic carbonate ramp, which is an analogue for the highly prolific reservoirs of the Arab D Formation in Qatar. We consider CO 2 water-alternating-gas (WAG) injection, which is a successful EOR method for carbonate reservoirs. The simulations are employed as a basis for generating data-driven surrogate models using polynomial regression and polynomial chaos expansion. Furthermore, the surrogates are validated by comparing surrogate predictions with results from numerical simulation and estimating goodness-of-fit measures. In the current work, parameter uncertainties affecting WAG modelling in fractured carbonates are evaluated, including fracture network properties, wettability and fault transmissibility. The results enable us to adequately explore the parameter space, and to quantify and rank the interrelated effect of uncertain model parameters on CO 2 WAG efficiency. The results highlight the first-order impact of the fracture network properties and wettability on hydrocarbon recovery and CO 2 utilization during WAG injection. In addition, the surrogate models enable us to calculate quick estimates of probabilistic uncertainty and to rapidly optimize WAG injection, while achieving significant computational speed-up compared with the conventional simulation framework.

SADEA-II: A generalized method for efficient global optimization of antenna design

Abstract Efficiency improvement is of great significance for simulation-driven antenna design optimization methods based on evolutionary algorithms (EAs). The two main efficiency enhancement methods exploit data-driven surrogate models and/or multi-fidelity simulation models to assist EAs. However, optimization methods based on the latter either need ad hoc low-fidelity model setup or have difficulties in handling problems with more than a few design variables, which is a main barrier for industrial applications. To address this issue, a generalized three stage multi-fidelity-simulation-model assisted antenna design optimization framework is proposed in this paper. The main ideas include introduction of a novel data mining stage handling the discrepancy between simulation models of different fidelities, and a surrogate-model-assisted combined global and local search stage for efficient high-fidelity simulation model-based optimization. This framework is then applied to SADEA, which is a state-of-the-art surrogate-model-assisted antenna design optimization method, constructing SADEA-II. Experimental results indicate that SADEA-II successfully handles various discrepancy between simulation models and considerably outperforms SADEA in terms of computational efficiency while ensuring improved design quality. Highlights An EFFICIENT antenna design global optimization method for problems requiring very expensive EM simulations. A new multi-fidelity surrogate-model-based optimization framework to perform RELIABLE efficient global optimization A data mining method to address distortions of EM models of different fidelities (bottleneck of multi-fidelity design).

Data-driven Surrogate Model Approach for Improving the Performance of Reactive Transport Simulations

Geochemical simulation models are the computational bottleneck for coupled reactive transport simulations. We investigate the use of a data-driven surrogate model in place of a geochemical simulation model to speed up the run-times of reactive transport simulations. This is a challenge because the surrogate model needs to use results of its predictions as inputs at each subsequent simulation step. We test the suitability of surrogate model approach on a popular reactive transport benchmark problem, often used for evaluating simulation models. We show that the concept is feasible and can make the simulations many times faster, however several open areas for future work remain.

History matching and production optimization of water flooding based on a data-driven interwell numerical simulation model

Production optimization of a water flooding project aims to determine the optimum production strategy for maintaining the reservoir under real-time optimal controls, and maximizing the oil production or net present value (NPV) of the project. Current production optimization method with a conventional grid-based simulator bears a high computational cost. Also, cumbersome history matching procedures are often required to build a reliable reservoir model for reducing the uncertainty of the optimized production strategies. To address the drawbacks of such conventional method, data-driven surrogate models, e.g., the correlation-based model, are developed to acquire interwell connectivity; such history-matched model can be then used for production optimization. It is noted that, however, the correlation-based models cannot handle the case where the conversion of duty occurs for some wells draining the reservoir, e.g., a production well is converted to a producer. Recently, we have developed a physics-based data-driven interwell numerical simulation model (INSIM) which can predict the production rates of individual phases and handle conversion-of-duty scenarios between any well pair. INSIM is derived by solving the material balance equation and augmenting the Buckley–Leverett theory. In this study, the INSIM is applied for the first time to optimize the production of water flooding projects. We derive and improve the INSIM by considering the active bottom or edge aquifers that are often encountered in real reservoirs. The improved INSIM can handle two types of aquifers: constant pressure aquifer and constant volume aquifer. By applying the inversion theory based on a finite-difference algorithm, we estimate the model parameters by history matching the historic production data. Subsequently, the history-matched INSIM is used for optimizing NPV of water flooding projects in both a synthetic reservoir and a real reservoir with an active aquifer. The injection rate of each injection well is treated as a tuning variable. As for the synthetic model, the simulation results show that the inversely derived model parameters are able to characterize the interwell formation properties, and the optimization result based on INSIM is in a good agreement with that obtained with the commercial simulator Eclipse 100. Similar history-matching results are also obtained when INSIM is applied to a real fracture-vuggy reservoir with an active aquifer. After production optimization of this real reservoir with INSIM, we achieve a 4% reduction in the water cut but a 20% increase in the oil production rate, which verifies the soundness of the proposed method.

Data mining and statistical inference in selective laser melting

Selective laser melting (SLM) is an additive manufacturing process that builds a complex three-dimensional part, layer-by-layer, using a laser beam to fuse fine metal powder together. The design freedom afforded by SLM comes associated with complexity. As the physical phenomena occur over a broad range of length and time scales, the computational cost of modeling the process is high. At the same time, the large number of parameters that control the quality of a part make experiments expensive. In this paper, we describe ways in which we can use data mining and statistical inference techniques to intelligently combine simulations and experiments to build parts with desired properties. We start with a brief summary of prior work in finding process parameters for high-density parts. We then expand on this work to show how we can improve the approach by using feature selection techniques to identify important variables, data-driven surrogate models to reduce computational costs, improved sampling techniques to cover the design space adequately, and uncertainty analysis for statistical inference. Our results indicate that techniques from data mining and statistics can complement those from physical modeling to provide greater insight into complex processes such as selective laser melting.

Surrogate Modeling of Ultrasonic Nondestructive Evaluation Simulations

Ultrasonic testing (UT) is used to detect internal flaws in materials or to characterize material properties. Computational simulations are an important part of the UT process. Fast models are essential for UT applications such as inverse design or model-assisted probability of detection. This paper presents investigations of using surrogate modeling techniques to create fast approximate models of UT simulator responses. In particular, we propose to use data-driven surrogate modeling techniques (kriging interpolation), and physics-based surrogate modeling techniques (space mapping), as well a mixture of the two approaches. These techniques are investigated for two cases involving UT simulations of metal components immersed in a water bath during the inspection process.

A multi-fidelity surrogate-model-assisted evolutionary algorithm for computationally expensive optimization problems

Integrating data-driven surrogate models and simulation models of different accuracies (or fidelities) in a single algorithm to address computationally expensive global optimization problems has recently attracted considerable attention. However, handling discrepancies between simulation models with multiple fidelities in global optimization is a major challenge. To address it, the two major contributions of this paper include: (1) development of a new multi-fidelity surrogate-model-based optimization framework, which substantially improves reliability and efficiency of optimization compared to many existing methods, and (2) development of a data mining method to address the discrepancy between the low- and high-fidelity simulation models. A new efficient global optimization method is then proposed, referred to as multi-fidelity Gaussian process and radial basis function-model-assisted memetic differential evolution. Its advantages are verified by mathematical benchmark problems and a real-world antenna design automation problem.

Nonlinear surrogate modeling of tibio-femoral joint interactions

Musculoskeletal modeling can predict muscle forces and the resulting motion and loading during human ambulatory activities. A better understanding of the loading environment on hard and soft tissues can enhance our understanding of ligament injury and prevention, tissue engineering, prosthetic design, osteoporosis, and osteoarthritis. The current state-of-the-art in movement simulation is to use simplified representation of the joints, such as representing the knee as a simple hinge joint. The aim of this study is to produce data-driven surrogate models which effectively capture the complex three-dimensional behavior of tibio-femoral joint interactions and that have the ease of use and computational efficiency required for incorporation in existing neuromusculoskeletal simulations. In order to meet our objective, we explored and compared the performance and sensitivity of nonlinear Hammerstein–Wiener, nonlinear autoregressive, and time delay neural network models under different configurations, individually and in ensembles. These models learned from solutions calculated by a validated multibody model of the knee. Inputs to the surrogate models were positions and orientations of the tibia relative to the femur, and the outputs were resulting forces and torques at the tibia with respect to the femur. Models were mixed using mean (sum) rule, weighted mean, and stacked generalization ensemble methods. It was observed that individually, time delay neural network models performed better than other models with normalized mean square errors between 0.0509 and 0.0889 on test data. Among the ensembles, stacked generalization provided the best results reducing test errors by 13–40%.