DynNet: Physics-based neural architecture design for nonlinear structural response modeling and prediction

Eshkevari, Soheil Sadeghi; Takáč, Martin; Pakzad, Shamim N.; Jahani, Majid

doi:10.1016/j.engstruct.2020.111582

Cited by 26 publications

(9 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…DynNet [25] is a network motivated by implicit equation of motion solvers in a recurrent cell for full response prediction of nonlinear multi degrees of freedom systems. The architecture of the model is based on ResNet [23], and the residual block is the only component of the network capable of learning the nonlinear behavior of the dynamic system.…”

Section: Using Deep Neural Network For Differential Equationsmentioning

confidence: 99%

A deep neural network for oxidative coupling of methane trained on high-throughput experimental data

et al. 2022

Self Cite

View full text Add to dashboard Cite

In this work, we develop a deep neural network model of the reaction rate of oxidative coupling of methane from published high-throughput experimental catalysis data. The neural network is formulated so that the rate model satisfies the plug flow reactor design equation. The model is then employed to understand the variation of reactant and product composition within the reactor for the reference catalyst Mn−Na2WO4/SiO2 at different temperatures and to identify new catalysts and combination of known catalysts that would increase yield and selectivity relative to the reference catalyst. The model revealed that methane is converted in the first half of the catalyst bed while the second part largely consolidates the products (i.e. increases ethylene to ethane ratio). A screening study of ≥ 3400 combinations of pairs of previously studied catalysts of the form M1(M2)1−2M3Ox/support (where M1, M2, and M3 are metals) revealed that a reactor configuration comprising two sequential catalyst beds leads to synergistic effects resulting in increased yield of C2 compared to the reference catalyst at identical conditions and contact time. Finally, an expanded screening study of 7400 combinations (comprising previously studied metals but with several new permutations) revealed multiple catalyst choices with enhanced yields of C2 products. This study shows the value of learning a deep neural network model of the instantaneous reaction rate directly from high throughput data and represents a first step in constraining a data-driven reaction model to satisfy domain information.

show abstract

Section: Using Deep Neural Network For Differential Equationsmentioning

confidence: 99%

A deep neural network for oxidative coupling of methane trained on high-throughput experimental data

et al. 2022

Self Cite

View full text Add to dashboard Cite

show abstract

“…Han et al [27] set up an error correction step, additionally training the deep neural network to approximate the model error, so as to increase the network performance. Eshkevari et al [28] designed a physics-based recurrent neural network model to estimate system dynamics. They proposed several techniques to significantly accelerate the learning process only using smaller datasets.…”

Section: Introductionmentioning

confidence: 99%

Neural network modeling and dynamic behavior prediction of nonlinear dynamic systems

et al. 2023

View full text Add to dashboard Cite

In this paper, data-driven models of nonlinear dynamic systems are built through neural network. The numerical solution of state equations is used to simulate the experimental data as well as training data of neural network. The loss function with two coefficients is constructed according to the relationship between the data. Forward Euler method and the trained neural network are combined to predict the response data of different excitation frequencies. Then, whole data-driven modeling and substructure data-driven modeling are presented to model and analyse a fivedegree-of-freedom duffing oscillator. By comparison, the prediction accuracy of substructure data-driven modeling is higher. Furthermore, the influence of loss function, hidden layer, training data and noise on the accuracy of the model is studied.The results show that the selection of training data and the number of hidden layers have a great impact on the prediction ability, and verify accuracy, generalization and robustness of the model. In order to improve the prediction accuracy of substructure data-driven modeling, adjustable learning rate is adopted in the optimizer. The effects of adding control parameters to the network input (APNI) and not adding control parameters to the network input (NAPNI) on the dynamic characteristics are considered respectively. The prediction accuracy of the model can be effectively improved by adding control parameters to the network input.

show abstract

“…Instead of establishing a purely black-box model, more and more data-driven algorithms intend to embed more physical/engineering interpretations towards an explainable mathematical model with implicit use of physical information, which can be found in literature. 16,17 Despite great progress in seeking accurate numerical approximator to nonlinear structural seismic response prediction using DL approaches, tedious training process and large volume of structural response data under earthquakes for training and validation are often prohibitively accessible. In this work, the main innovation can be seen in the incorporation of deep neural networks (DNN) into a classical numerical integration method by using a hybridized integration time-stepper.…”

Section: Introductionmentioning

confidence: 99%

“…Specifically, considerable progress has been made in physics‐informed learning strategy for modeling and prediction of dynamics. Instead of establishing a purely black‐box model, more and more data‐driven algorithms intend to embed more physical/engineering interpretations towards an explainable mathematical model with implicit use of physical information, which can be found in literature 16,17 …”

Section: Introductionmentioning

confidence: 99%

Combination of physics‐based and data‐driven modeling for nonlinear structural seismic response prediction through deep residual learning

Guo

Enokida

et al. 2023

Earthq Engng Struct Dyn

View full text Add to dashboard Cite

Despite great progress in seeking accurate numerical approximator to nonlinear structural seismic response prediction using deep learning approaches, tedious training process and large volume of structural response data under earthquakes for training and validation are often prohibitively accessible. In our methodology, the main innovation can be seen in the incorporation of deep neural networks (DNNs) into a classical numerical integration method by using a hybridized integration time-stepper. In this way, the linear physics information of the structure and the obscure nonlinear dynamics are smoothly combined. We propose to use residual network (ResNet) to learn time-stepping schemes specifically for the nonlinear state variables of the system. Our Physics-DNN hybridized integration (PDHI) time-stepping scheme provides important advantages over current pure data-driven approaches, including (i) a flexible framework incorporating known time-invariant physics information, (ii) requirement of structural seismic response data being circumvented by simple short bursts of trajectories collected from underlying nonlinear components, and (iii) efficiency in training and validation process. Besides, our results indicate that a simple feedforward or convolutional architecture outperforms recurrent networks to fulfill the requirement of prediction accuracy as well as long-range memory in structural dynamic analysis. Several numerical and experimental examples are presented to demonstrate the performance of the method.

show abstract

DynNet: Physics-based neural architecture design for nonlinear structural response modeling and prediction

Cited by 26 publications

References 34 publications

A deep neural network for oxidative coupling of methane trained on high-throughput experimental data

A deep neural network for oxidative coupling of methane trained on high-throughput experimental data

Neural network modeling and dynamic behavior prediction of nonlinear dynamic systems

Combination of physics‐based and data‐driven modeling for nonlinear structural seismic response prediction through deep residual learning

Contact Info

Product

Resources

About