scholarly journals Hybrid Modelling by Machine Learning Corrections of Analytical Model Predictions towards High-Fidelity Simulation Solutions

Materials ◽  
2021 ◽  
Vol 14 (8) ◽  
pp. 1883
Author(s):  
Frederic E. Bock ◽  
Sören Keller ◽  
Norbert Huber ◽  
Benjamin Klusemann
Keyword(s):  
Neural Network ◽  
Machine Learning ◽  
Analytical Model ◽  
Data Driven ◽  
High Fidelity ◽  
Prediction Errors ◽  
Hybrid Modelling ◽  
Model Predictions ◽  
Physical Laws ◽  
Modelling Approach

Within the fields of materials mechanics, the consideration of physical laws in machine learning predictions besides the use of data can enable low prediction errors and robustness as opposed to predictions only based on data. On the one hand, exclusive utilization of fundamental physical relationships might show significant deviations in their predictions compared to reality, due to simplifications and assumptions. On the other hand, using only data and neglecting well-established physical laws can create the need for unreasonably large data sets that are required to exhibit low bias and are usually expensive to collect. However, fundamental but simplified physics in combination with a corrective model that compensates for possible deviations, e.g., to experimental data, can lead to physics-based predictions with low prediction errors, also despite scarce data. In this article, it is demonstrated that a hybrid model approach consisting of a physics-based model that is corrected via an artificial neural network represents an efficient prediction tool as opposed to a purely data-driven model. In particular, a semi-analytical model serves as an efficient low-fidelity model with noticeable prediction errors outside its calibration domain. An artificial neural network is used to correct the semi-analytical solution towards a desired reference solution provided by high-fidelity finite element simulations, while the efficiency of the semi-analytical model is maintained and the applicability range enhanced. We utilize residual stresses that are induced by laser shock peening as a use-case example. In addition, it is shown that non-unique relationships between model inputs and outputs lead to high prediction errors and the identification of salient input features via dimensionality analysis is highly beneficial to achieve low prediction errors. In a generalization task, predictions are also outside the process parameter space of the training region while remaining in the trained range of corrections. The corrective model predictions show substantially smaller errors than purely data-driven model predictions, which illustrates one of the benefits of the hybrid modelling approach. Ultimately, when the amount of samples in the data set is reduced, the generalization of the physics-related corrective model outperforms the purely data-driven model, which also demonstrates efficient applicability of the proposed hybrid modelling approach to problems where data is scarce.

scholarly journals Data-driven predictions of a multiscale Lorenz 96 chaotic system using machine-learning methods: reservoir computing, artificial neural network, and long short-term memory network

2020 ◽  
Vol 27 (3) ◽  
pp. 373-389 ◽  
Author(s):  
Ashesh Chattopadhyay ◽  
Pedram Hassanzadeh ◽  
Devika Subramanian
Keyword(s):  
Neural Network ◽  
Machine Learning ◽  
Short Term Memory ◽  
Data Driven ◽  
Reservoir Computing ◽  
Short Term ◽  
Term Memory ◽  
Machine Learning Methods ◽  
Long Short Term Memory ◽  
Lorenz 96

Abstract. In this paper, the performance of three machine-learning methods for predicting short-term evolution and for reproducing the long-term statistics of a multiscale spatiotemporal Lorenz 96 system is examined. The methods are an echo state network (ESN, which is a type of reservoir computing; hereafter RC–ESN), a deep feed-forward artificial neural network (ANN), and a recurrent neural network (RNN) with long short-term memory (LSTM; hereafter RNN–LSTM). This Lorenz 96 system has three tiers of nonlinearly interacting variables representing slow/large-scale (X), intermediate (Y), and fast/small-scale (Z) processes. For training or testing, only X is available; Y and Z are never known or used. We show that RC–ESN substantially outperforms ANN and RNN–LSTM for short-term predictions, e.g., accurately forecasting the chaotic trajectories for hundreds of numerical solver's time steps equivalent to several Lyapunov timescales. The RNN–LSTM outperforms ANN, and both methods show some prediction skills too. Furthermore, even after losing the trajectory, data predicted by RC–ESN and RNN–LSTM have probability density functions (pdf's) that closely match the true pdf – even at the tails. The pdf of the data predicted using ANN, however, deviates from the true pdf. Implications, caveats, and applications to data-driven and data-assisted surrogate modeling of complex nonlinear dynamical systems, such as weather and climate, are discussed.

Structural damage identification via physics-guided machine learning: a methodology integrating pattern recognition with finite element model updating

2020 ◽  
pp. 147592172092748 ◽  
Author(s):  
Zhiming Zhang ◽  
Chao Sun
Keyword(s):  
Neural Network ◽  
Machine Learning ◽  
Finite Element ◽  
Finite Element Model ◽  
Health Monitoring ◽  
Damage Identification ◽  
Model Updating ◽  
Element Model ◽  
Data Driven ◽  
Finite Element Model Updating

Structural health monitoring methods are broadly classified into two categories: data-driven methods via statistical pattern recognition and physics-based methods through finite elementmodel updating. Data-driven structural health monitoring faces the challenge of data insufficiency that renders the learned model limited in identifying damage scenarios that are not contained in the training data. Model-based methods are susceptible to modeling error due to model idealizations and simplifications that make the finite element model updating results deviate from the truth. This study attempts to combine the merits of data-driven and physics-based structural health monitoring methods via physics-guided machine learning, expecting that the damage identification performance can be improved. Physics-guided machine learning uses observed feature data with correct labels as well as the physical model output of unlabeled instances. In this study, physics-guided machine learning is realized with a physics-guided neural network. The original modal-property based features are extended with the damage identification result of finite element model updating. A physics-based loss function is designed to evaluate the discrepancy between the neural network model output and that of finite element model updating. With the guidance from the scientific knowledge contained in finite element model updating, the learned neural network model has the potential to improve the generality and scientific consistency of the damage detection results. The proposed methodology is validated by a numerical case study on a steel pedestrian bridge model and an experimental study on a three-story building model.

scholarly journals A wavelet-neural network hybrid modelling approach for estimating and predicting river monthly flows

2013 ◽  
Vol 58 (2) ◽  
pp. 374-389 ◽  
Author(s):  
Shouke Wei ◽  
Hong Yang ◽  
Jinxi Song ◽  
Karim Abbaspour ◽  
Zongxue Xu
Keyword(s):  
Neural Network ◽  
Wavelet Neural Network ◽  
Hybrid Modelling ◽  
Modelling Approach

scholarly journals Reaction Wheels Fault Isolation Onboard 3-Axis Controlled Satel-lite using Enhanced Random Forest with Multidomain Features

2021 ◽  
Vol 12 (2) ◽  
Author(s):  
Afshin Rahimi ◽  
Mofiyinoluwa O. Folami
Keyword(s):  
Machine Learning ◽  
Fault Isolation ◽  
Data Driven ◽  
Machine Learning Techniques ◽  
Superior Performance ◽  
Alternative Methods ◽  
Model Parameters ◽  
High Fidelity ◽  
Novel Approach ◽  
Learning Techniques

As the number of satellite launches increases each year, it is only natural that an interest in the safety and monitoring of these systems would increase as well. However, as a system becomes more complex, generating a high-fidelity model that accurately describes the system becomes complicated. Therefore, imploring a data-driven method can provide to be more beneficial for such applications. This research proposes a novel approach for data-driven machine learning techniques on the detection and isolation of nonlinear systems, with a case-study for an in-orbit closed loop-controlled satellite with reaction wheels as actuators. High-fidelity models of the 3-axis controlled satellite are employed to generate data for both nominal and faulty conditions of the reaction wheels. The generated simulation data is used as input for the isolation method, after which the data is pre-processed through feature extraction from a temporal, statistical, and spectral domain. The pre-processed features are then fed into various machine learning classifiers. Isolation results are validated with cross-validation, and model parameters are tuned using hyperparameter optimization. To validate the robustness of the proposed method, it is tested on three characterized datasets and three reaction wheel configurations, including standard four-wheel, three-orthogonal, and pyramid. The results prove superior performance isolation accuracy for the system under study compared to previous studies using alternative methods (Rahimi & Saadat, 2019, 2020).

scholarly journals Acoustic Classification of Surface and Underwater Vessels in the Ocean Using Supervised Machine Learning

Sensors ◽  
2019 ◽  
Vol 19 (16) ◽  
pp. 3492 ◽  
Author(s):  
Jongkwon Choi ◽  
Youngmin Choo ◽  
Keunhwa Lee
Keyword(s):  
Neural Network ◽  
Machine Learning ◽  
Density Matrix ◽  
Spectral Density ◽  
Input Data ◽  
Supervised Machine Learning ◽  
Data Driven ◽  
Matrix Elements ◽  
Spectral Density Matrix ◽  
Cross Spectral Density

Four data-driven methods—random forest (RF), support vector machine (SVM), feed-forward neural network (FNN), and convolutional neural network (CNN)—are applied to discriminate surface and underwater vessels in the ocean using low-frequency acoustic pressure data. Acoustic data are modeled considering a vertical line array by a Monte Carlo simulation using the underwater acoustic propagation model, KRAKEN, in the ocean environment of East Sea in Korea. The raw data are preprocessed and reorganized into the phone-space cross-spectral density matrix (pCSDM) and mode-space cross-spectral density matrix (mCSDM). Two additional matrices are generated using the absolute values of matrix elements in each CSDM. Each of these four matrices is used as input data for supervised machine learning. Binary classification is performed by using RF, SVM, FNN, and CNN, and the obtained results are compared. All machine-learning algorithms show an accuracy of >95% for three types of input data—the pCSDM, mCSDM, and mCSDM with the absolute matrix elements. The CNN is the best in terms of low percent error. In particular, the result using the complex pCSDM is encouraging because these data-driven methods inherently do not require environmental information. This work demonstrates the potential of machine learning to discriminate between surface and underwater vessels in the ocean.

Physics-Aware Deep-Learning-Based Proxy Reservoir Simulation Model Equipped with State and Well Output Prediction

2021 ◽  
Author(s):  
Emilio J. R. Coutinho ◽  
Marcelo J. Aqua and Eduardo Gildin
Keyword(s):  
Neural Network ◽  
Machine Learning ◽  
Deep Learning ◽  
Reservoir Simulation ◽  
Data Driven ◽  
Training Procedure ◽  
State Variables ◽  
Input Output ◽  
Well Control ◽  
Reduced Order

Abstract Physics-aware machine learning (ML) techniques have been used to endow data-driven proxy models with features closely related to the ones encountered in nature. Examples span from material balance and conservation laws. Physics-based and data-driven reduced-order models or a combination thereof (hybrid-based models) can lead to fast, reliable, and interpretable simulations used in many reservoir management workflows. We built on a recently developed deep-learning-based reduced-order modeling framework by adding a new step related to information of the input-output behavior (e.g., well rates) of the reservoir and not only the states (e.g., pressure and saturation) matching. A Combination of data-driven model reduction strategies and machine learning (deep- neural networks – NN) will be used here to achieve state and input-output matching simultaneously. In Jin, Liu and Durlofsky (2020), the authors use a NN architecture where it is possible to predict the state variables evolution after training an autoencoder coupled with a control system approach (Embed to Control - E2C) and adding some physical components (Loss functions) to the neural network training procedure. In this paper, we extend this idea by adding the simulation model output, e.g., well bottom-hole pressure and well flowrates, as data to be used in the training procedure. Additionally, we added a new neural network to the E2C transition model to handle the connections between state variables and model outputs. By doing this, it is possible to estimate the evolution in time of both the state variables as well as the output variables simultaneously. The method proposed provides a fast and reliable proxy for the simulation output, which can be applied to a well-control optimization problem. Such a non-intrusive method, like data-driven models, does not need to have access to reservoir simulation internal structure. So it can be easily applied to commercial reservoir simulations. We view this as an analogous step to system identification whereby mappings related to state dynamics, inputs (controls), and measurements (output) are obtained. The proposed method is applied to an oil-water model with heterogeneous permeability, 4 injectors, and 5 producer wells. We used 300 sampled well control sets to train the autoencoder and another set to validate the obtained autoencoder parameters.

Data-Driven Computing With Convolutional Neural Networks for Two-Phase Flows: Application to Wave-Structure Interaction

2018 ◽  
Author(s):  
X. Mao ◽  
V. Joshi ◽  
T. P. Miyanawala ◽  
Rajeev K. Jaiman
Keyword(s):  
Neural Network ◽  
Free Surface ◽  
Bluff Body ◽  
Data Driven ◽  
Model Parameters ◽  
Wave Forces ◽  
Approximation Procedure ◽  
High Fidelity ◽  
Bluff Bodies ◽  
Fluid Forces

Fluctuating wave force on a bluff body is of great significance in many offshore and marine engineering applications. We present a Convolutional Neural Network (CNN) based data-driven computing to predict the unsteady wave forces on bluff bodies due to the free-surface wave motion. For the full-order modeling and high-fidelity data generation, the air-water interface for such wave-body problems must be captured accurately for a broad range of physical and geometric parameters. Originated from the thermodynamically consistent theories, the physically motivated Allen-Cahn phase-field method has many advantages over other interface capturing techniques such as level-set and volume-of-fluid methods. The Allen-Cahn equation is solved in the mass-conservative form by imposing a Lagrange multiplier technique. While a tremendous amount of wave-body interaction data is generated in offshore engineering via both CFD simulations and experiments, the results are generally underutilized. Design space exploration and flow control of such practical scenarios are still time-consuming and expensive. An alternative to semi-analytical modeling, CNN is a class of deep neural network for solving inverse problems which is efficient in parametric data-driven computation and can use the domain knowledge. It establishes a model with arbitrarily generated model parameters, makes predictions using the model and existing input parametric settings, and adjusts the model parameters according to the error between the predictions and existing results. The computational cost of this prediction process, compared with high-fidelity CFD simulation, is significantly reduced, which makes CNN an accessible tool in design and optimization problems. In this study, CNN-based data-driven computing is utilized to predict the wave forces on bluff bodies with different geometries and distances to the free surface. The discrete convolution process with a non-linear rectification is employed to approximate the mapping between the bluff-body shape, the distance to the free-surface and the fluid forces. The wave-induced fluid forces on bluff bodies of different shapes and submergences are predicted by the trained CNN. Finally, a convergence study is performed to identify the effective hyper-parameters of the CNN such as the convolution kernel size, the number of kernels and the learning rate. Overall, the proposed CNN-based approximation procedure has a profound impact on the parametric design of bluff bodies experiencing wave loads.

Sign in / Sign up

Export Citation Format

Share Document