Artificial Neural Network Surrogate Model Research Articles

Optimizing passive design strategies for energy efficient buildings using hybrid artificial neural network (ANN) and multi-objective evolutionary algorithm through a case study approach

Building energy reduction opportunities and thermal comfort of occupants are two major aspects that could be enhanced through passive design solutions. In tropical weather conditions such as in India, there have been very minimal efforts in developing simulation-based approaches to understand the impacts of common passive design solutions on energy needs and thermal comfort. Therefore, this study focuses on developing an Artificial Neural Network (ANN) and Building Energy Simulation and Optimization (BESO) based framework to improve building energy performance and thermal comfort using passive design solutions in Indian climate condition with mixed-mode operation strategy. The initial part of the study performs building energy simulation through a coupled approach of DesignBuilder and jEPlus to develop a valid ANN surrogate model. Eventually, ANN outputs are integrated with multi-objective evolutionary algorithm (Non-dominated Sorting Genetic Algorithm-II) to find the optimal solutions of predefined passive decision variables of the building for minimizing building energy consumption and maximizing thermal comfort. The optimized results in the study show decrease of 46% of building energy consumption and 7.58% of discomfort hours when the set of passive design solutions are opted. The findings of optimal passive design solutions could be helpful for designers and practitioners in developing energy-efficient buildings.

Neural Network Meta-Models for FPSO Motion Prediction From Environmental Data With Different Platform Loads

The current design process of mooring systems for Floating Production, Storage, and Offloading units (FPSOs) depends on the availability of the platform’s mathematical model and the accuracy of dynamic simulations. These simulations then provide the FPSO’s time series motion which is evaluated according to design constraints. This process can be time-consuming and present inaccurate results due to the mathematical model’s limitations and the overall complexity of the vessel’s dynamics. We propose a Neural Simulator, called NeuroSim, a set of data-based surrogate models with environmental data as input, each model specialized in predicting different motion statistics relevant to mooring system design: Maximum Roll, Platform Offset, and Fairlead Displacements. The surrogate models are trained by current, wind, and wave data given in 3 hours periods at a Brazilian Offshore Basin from 2003 to 2010, and the associated dynamic response of a spread-moored FPSO is obtained through time-domain simulations using the Dynasim software. Hyperparameter Optimization techniques are performed to obtain optimal Artificial Neural Network (ANN) models specialized in different platform drafts. Finally, the proposed models are shown to correctly capture platform dynamics, providing good results when compared to motion statistics obtained from Dynasim. We conclude that an ANN surrogate model can be trained directly on actual measured metocean conditions and corresponding FPSO motion statistics to provide increased accuracy and reduced computational time over traditional methods based on dynamic simulation. Moreover, the proposed architecture can be integrated into an automated learning framework: The data-based surrogate models can be continuously fine-tuned and updated with newly measured data, improving accuracy over time.

Stochastic natural frequency analysis of varying diameter functionally graded material pipe conveying fluid

This paper investigates the uncertainty of the structure and material parameters on the natural frequency of the varying diameter functionally graded material (FGM) pipe conveying fluid. Firstly, concentrating on the spatially dependent uncertainty of geometry varying in axial direction as well as the material varying in radial direction, the random field is employed to describe the uncertainty parameters. Then, a generalized eigenvalue equation of the varying diameter FGM pipe is derived by the dynamic stiffness method while introducing the random field and variables into the deterministic governing equation. Due to the input uncertainty, the natural frequencies of the varying diameter FGM pipe change to random variables. In this respect, the stochastic characteristics of natural frequencies can be approximated by the Monte Carlo simulation. However, the enormous computational cost is unbearable. The efficient artificial neural network surrogate model is introduced and incorporated for the natural frequency uncertainty quantification. Consequently, a stochastic natural frequency analysis approach based on the neural network is proposed for the varying diameter FGM pipe conveying fluid. Some examples are investigated to illustrate the effectiveness of the proposed method and the influence of input uncertainty on the probabilistic characterization of the natural frequencies.

Artificial Neural Network based Particle Swarm Optimization solution approach for the inverse depletion of used nuclear fuel

This work builds upon previous efforts to apply inverse analysis in the determination of the initial conditions and burnup history of used nuclear fuel for forensic purposes. In this work, the inverse depletion problem is defined and solved using a surrogate-based approach equipped with the Particle Swarm Optimization algorithm and an Artificial Neural Network surrogate model. The proposed approach is outlined and verified via a series of mockup forensic case studies based upon a VVER-1000 assembly model depleted via SCALE6.1 KENO VI module.The case studies considered retrieving the fuel initial enrichment and burnup using the final used nuclear fuel isotopic content. Results indicate that the proposed approach can estimate the initial fuel enrichment and maximum burnup with reasonable computational cost and acceptable accuracy. The relative error in estimating the fuel initial enrichment was <6% while that of the actual fuel burnup was <3% for all mockup case studies.

Optimization of an explosive waste incinerator via an artificial neural network surrogate model

In this study, the operating conditions for a novel method of explosive waste disposal have been optimized to minimize the formation of NOx emissions. Previous methods of disposal include rotary kilns but have disadvantages, especially associated with the formation of NOx. To address this, development of a fluidized bed reactor as an incinerator has been proposed. However, as the number of input variables to consider increases, optimization of the design through computational fluid dynamics (CFD) simulations alone becomes challenging, because of the significant computational time required. Thus, we proposed to build a surrogate model of CFD output using an artificial neural network (ANN) that can be used to locate the optimal operating conditions. To perform the optimization on five variables, finite ranges of these variables were selected and sampled using a Latin Hypercube Method, giving a total of 300 random input samples. CFD simulations were performed using each of these conditions, and the ANN surrogate model was designed to match the results. Due to its higher accuracy, the surrogate model from the Bayesian Regularization method was selected, which had an error rate of around 1% when compared to the CFD results. Using this surrogate model, a 34% reduction in NOx emissions from the reactor was achievable. Overall, the fluidized bed reactors show promise with regard to the reduction of NOx emissions from explosive waste disposal, and surrogate modeling with ANNs represents a useful method for reducing the CFD computation time used for multivariable optimization of such a system.

Adaptive Artificial Neural Network Surrogate Model of Nonlinear Hydraulic Adjustable Damper for Automotive Semi-Active Suspension System

Artificial neural network surrogate models are often used in the design optimization of the automotive semi-active suspension system. To realize the desired damping force, a surrogate model needs to be constructed to approximate the regulating mechanism of the hydraulic adjustable damper. However, very few of the existing studies discuss how to guarantee the modeling accuracy. To this end, this work constructs a novel surrogate model by using radial basis function neural network. Meanwhile, an adaptive modeling method based on modified hyperband and trust-region-based mode pursuing sampling is presented. Concretely, modified hyperband is used to fast select a seed model by early-stopping and dynamic resource allocation. Mode pursuing sampling is then performed in the neighborhood of the seed model, to systematically generate more sample points while statistically covering the entire neighborhood (i.e., trust region). In particular, in the mode pursuing sampling procedure, quadratic regression is performed around the current optimum as the second detection. Moreover, as the position or size of the trust region changes, the sampling and detection process iterate until the accuracy of the model is no longer improved. To avoid falling into the local optimum, the seed model selection and mode pursuing sampling process iterate until the termination criterion is met. The experimental results show that compared with the benchmarks, the modeling accuracy of hydraulic adjustable damper can be improved by up to 48%, and the iteration resources can be reduced by up to 84%.

Facilitating Bayesian analysis of combustion kinetic models with artificial neural network

Bayesian analysis provides a framework for the inverse uncertainty quantification (UQ) of combustion kinetic models. As the workhorse of the Bayesian approach, the Markov chain Monte Carlo (MCMC) methods, however, incur a substantial computational cost. In this work, a surrogate model is employed to improve the traditional MCMC algorithm. Specifically, the test errors of three typical surrogate models are compared, namely Polynomial Chaos Expansion (PCE), High Dimensional Model Representation (HDMR) and Artificial Neural Network (ANN); and ANN is shown to be a relatively more efficient surrogate model for the approximation of combustion reaction systems under extensive conditions. An inverse UQ method, which is the combination of the ANN and traditional MCMC method, and as such termed ANN–MCMC, is adopted. The calculation is performed on the methanol oxidation system and a series of ignition delay data are employed to optimize the rate coefficients of the kinetic model. The estimated posterior distributions of the rate coefficients and the model predictions using the ANN–MCMC are compared with the traditional MCMC methods, with the results showing that the ANN–MCMC can significantly reduce the computational cost needed for the MCMC algorithm, especially on the estimation of the posterior distributions of the input parameters. The rejection rate of the samples in a Markov chain is very high, especially for the calculation of the posterior distribution of less sensitive parameters, thus a large number of samples are needed to reach a desired accuracy for traditional MCMC process. While no samples are rejected when training the ANN surrogate model. Therefore, fewer original samples are needed to get a converged ANN surrogate, which can then generate a large number of low-cost ANN samples for a better accuracy of the MCMC process. The errors for the estimated posterior distributions using ANN–MCMC depend on the accuracy of converged ANN surrogates and more accurate results are obtained with improved settings of ANN. The ANN–MCMC is especially suitable to the computational systems when the computational ability is limited.

Transfer Learning as a Tool for Reducing Simulation Bias: Application to Inertial Confinement Fusion

We adopt a technique, known in the machine learning community as transfer learning, to reduce the bias of computer simulation using very sparse experimental data. Unlike the Bayesian calibration, which is commonly used to estimate the simulation bias, the transfer learning approach discussed in this article involves calculating an artificial neural network surrogate model of the simulations. Assuming that the simulation code correctly predicts the trends in the experimental data but it is subject to unknown biases, we then partially retrain, or transfer learn, the initial surrogate model to match the experimental data. This process eliminates the bias while still taking advantage of the physics relations learned from the simulation. Transfer learning can be easily adapted to a wide range of problems in science and engineering. In this article, we carry out numerical tests to investigate the applicability of this technique to predict the observable outcomes of inertial confinement fusion (ICF) experiments under new conditions. Using our synthetic validation data set, we demonstrate that an accurate predictive model can be built by retraining an initial surrogate model with experimental data volumes so small that they are relevant to the ICF problem. This opens up new opportunities for knowledge transfer and building predictive models in physics. After implementing transfer learning in a standard neural network, we successfully extended the method to a more complex, generative adversarial network architecture, which will be needed for predicting not only scalars but also diagnostic images in our future work.

Optimization Approaches for Durable Reinforced Concrete Structures considering Interval and Stochastic Parameter Uncertainty

AbstractThe durability of reinforced concrete structures is dominated by steel reinforcement corrosion and uncertain service loads, which have to be considered in the lifetime oriented structural design. The transport of corrosive substances into the structure is considerably influenced by load induced cracking. The crack width therefore is a major controlling factor for the lifetime of reinforced concrete structures. To improve the design for durability, finite element models in combination with optimization approaches for polymorphic uncertain data are presented. Here, the crack width at the reinforcement layer is used as the optimization objective to be minimized. The structural reliability is treated as a constraint of the optimization task in terms of the accepted failure probability. The concrete covers of the reinforcement layers are chosen as interval design parameters, with midpoints to be optimized and a given radius to take construction imprecision into account. The structural loading, the Young's modulus of concrete and the corresponding tensile strength are considered as stochastic a priori parameters within the optimization, which is solved by a particle swarm optimization approach in combination with an artificial neural network surrogate model. The polymorphic uncertain structural response is computed by a combination of Monte Carlo simulations and optimization‐based interval analysis to consider the stochastic and interval parameters within the optimization task, respectively. An application example demonstrates the performance of the proposed approach.

Hydropower Optimization Using Artificial Neural Network Surrogate Models of a High‐Fidelity Hydrodynamics and Water Quality Model

AbstractHydropower operations optimization subject to environmental constraints is limited by challenges associated with dimensionality and spatial and temporal resolution. The need for high‐fidelity hydrodynamic and water quality models within optimization schemes is driven by improved computational capabilities, increased requirements to meet specific points of compliance with greater resolution, and the need to optimize operations of not just single reservoirs but systems of reservoirs. This study describes an important advancement for computing hourly power generation schemes for a hydropower reservoir using high‐fidelity models, surrogate modeling techniques, and optimization methods. The predictive power of the high‐fidelity hydrodynamic and water quality model CE‐QUAL‐W2 is successfully emulated by an artificial neural network, then integrated into a genetic algorithm optimization approach to maximize hydropower generation subject to constraints on dam operations and water quality. This methodology is applied to a multipurpose reservoir near Nashville, Tennessee, USA. The model successfully reproduced high‐fidelity reservoir information while enabling 6.8% and 6.6% increases in hydropower production value relative to actual operations for dissolved oxygen (DO) limits of 5 and 6 mg/L, respectively, while witnessing an expected decrease in power generation at more restrictive DO constraints. Exploration of simultaneous temperature and DO constraints revealed capability to address multiple water quality constraints at specified locations. The reduced computational requirements of the new modeling approach demonstrated an ability to provide decision support for reservoir operations scheduling while maintaining high‐fidelity hydrodynamic and water quality information as part of the optimization decision support routines.

Surrogate-enhanced stochastic search algorithms to identify implicitly defined functions for reliability analysis

The paper proposes two adaptive stochastic search algorithms to locate and trace an implicitly defined function with samples that are used to construct a surrogate model for reliability analysis. Both methods begin by propagating a series of surrogate-informed stochastic processes toward the implicit performance function. Having located the function, the first method conducts a “global” tracing of the function while the second traces the function by propagating samples locally. A key feature of both proposed algorithms is that the surrogate model evolves continuously with the sample selection and is used to inform the selection of new samples such that it converges rapidly to an accurate representation of the limit surface. In the present implementation, an artificial neural network surrogate model is employed but the method can, in principle, be applied with any surrogate model form. Performance of the algorithms is illustrated through problems with highly nonlinear limit state functions and high dimensional non-Gaussian random variables.

Delamination detection with error and noise polluted natural frequencies using computational intelligence concepts

Delaminations are one of the most prevalent defects noticed in laminated composite structures. The existence of delaminations changes the vibration characteristics of laminates and hence such indicators can be used to quantify the health characteristics of laminates and detect potential risk of catastrophic failures. An approach for online health monitoring of in-service composite laminates is presented in this paper that relies on methods based on computational intelligence. Typical changes in the observed vibration characteristics (i.e. change in natural frequencies) are considered as inputs to identify the existence, location and magnitude of delaminations. The performance of the proposed approach is demonstrated using both numerical models and experimental studies of composite laminates. Since this identification problem essentially involves the solution of an optimization problem, the use of finite element (FE) methods as the underlying tool for analysis turns out to be computationally expensive. A surrogate assisted optimization approach is hence introduced to contain the computational time within affordable limits. An artificial neural network (ANN) model with Bayesian regularization is used as the underlying approximation scheme while an improved rate of convergence is achieved using a memetic algorithm. However, building of ANN surrogate models usually requires large training datasets. K-means clustering is effectively employed to reduce the size of datasets. ANN is also used for the solution of the inverse problem to determine the interface, size and location of delaminations using changes in measured natural frequencies before and after damage. The algorithms successfully performed delamination detection given limited amount of training datasets. Since in all practical problems, noise and error are inherently present, the performance of the proposed approach is also evaluated under varying levels of noise and error. The results clearly highlight the efficiency and the robustness of the approach. Hence, a delamination prediction strategy via K-means clustering, ANN and optimization algorithms integrated with surrogate models based on ANN for computational enhancement have been successfully developed and found efficient for detection of the interface of delamination, its size and location in FRP composite laminates using variations in natural frequencies. The developed algorithms are applied to composite beams. Results clearly indicate the delamination detection capability of the algorithms. The results demonstrate that the proposed techniques provide remarkable accurate detection of delamination damage with error and noise corrupted natural frequency validation data. The algorithms are quite promising, successful, flexible and practical with significant increase in terms of efficiency and effectiveness for SHM.

High-Degree-of-Freedom Engine Modelling for Control Design Using a Crank-Angle-Resolved Flame Propagation Simulation and Artificial Neural Network Surrogate Models

Non-linearity of the engine system creates a challenge in building a reliable control-oriented model (COM). The main source of non-linearity is the complex nature of the combustion process. Modern engine system configurations are increasingly complex and predicting their transient response poses additional difficulty. In the present paper, a COM is developed to address the challenges and capture the behaviour of a high-degree-of-freedom engine system. Engine combustion models are created by utilizing the high-fidelity engine cycle simulation to characterize the effects of main parameters, such as turbulence, air—fuel ratio, and residual fraction, and subsequently capturing the interrelationships with artificial neural networks. Then, system dynamics are accounted for by adding manifold and actuator dynamics models. The capabilities of the proposed COM are demonstrated using a spark-ignition engine with a dual-independent cam phasing as a test case. The results indicate the model's ability to accurately predict engine responses to an arbitrary schedule of engine control inputs over the feasible operating range.