Response prediction of nonlinear hysteretic systems by deep neural networks

Kim, Taeyong; Kwon, O’Dae; Song, Junho

doi:10.1016/j.neunet.2018.12.005

Cited by 77 publications

(55 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The authors have found that the capacity spectrum method shows a similar level of estimation error as compared with the coefficient method. 13,16 Therefore, just as the coefficient method, a similar amount of error is expected in the regional loss assessment based on HAZUS, 31 which adopted the capacity spectrum method for estimating structural responses given seismic intensities.…”

Section: Example 1: V-city Subjected To a Point Sourcementioning

confidence: 96%

“…To precisely predict the structural responses under ground motion, a deep neural network-based seismic response prediction framework, referred to as "DNN model" henceforth, was developed by merging the important features of structural information with those representing earthquake ground motion information. 13 The framework aimed to predict the maximum structural responses during the seismic excitation, which are considered the most important quantity in earthquake engineering practice. The important features of structural hysteretic behavior, which can be obtained by performing a quasi-static cyclic analysis, were extracted from the convolutional neural network (CNN).…”

Section: Probabilistic Evaluation Of Structural Responses Using Deementioning

confidence: 99%

See 1 more Smart Citation

Pre‐ and post‐earthquake regional loss assessment using deep learning

Kim

Song

Kwon

2020

Earthq Engng Struct Dyn

Self Cite

View full text Add to dashboard Cite

As urban systems become more highly sophisticated and interdependent, their vulnerability to earthquake events exhibits a significant level of uncertainties. Thus, community-level seismic risk assessments are indispensable to facilitate decision making for effective hazard mitigation and disaster responses. To this end, new frameworks for pre-and post-earthquake regional loss assessments are proposed using deep learning methods. First, to improve the accuracy of the response prediction of individual structures during the pre-earthquake loss assessment, a widely used nonlinear static procedure is replaced by the recently developed probabilistic deep neural network model. The variabilities of the nonlinear responses of a structural system given the seismic intensity can be quantified during the loss assessment process. Second, to facilitate near-real-time post-earthquake loss assessments, an adaptive algorithm, which identifies the optimal number and locations of sensors in a given urban area, is proposed. Using a deep neural network that estimates area-wide structural damage given the spatial distribution of the seismic intensity levels as a surrogate model, the algorithm adaptively places additional sensors at property lots at which errors from surrogate estimations of the structural damage are the greatest. Note that the surrogate model is constructed before earthquake events using simulated datasets. To test and demonstrate the proposed frameworks, the paper introduces thorough numerical investigations of two hypothetical urban communities. The proposed frameworks using the deep learning methods are expected to make critical advances in pre-and post-earthquake regional loss assessments. K E Y W O R D Sadaptive algorithm, deep learning, earthquake, optimal sensor placement, probabilistic seismic risk assessment, surrogate model

show abstract

Section: Example 1: V-city Subjected To a Point Sourcementioning

confidence: 96%

Section: Probabilistic Evaluation Of Structural Responses Using Deementioning

confidence: 99%

Pre‐ and post‐earthquake regional loss assessment using deep learning

Kim

Song

Kwon

2020

Earthq Engng Struct Dyn

Self Cite

View full text Add to dashboard Cite

show abstract

“…Figure 1 shows an example with three jobs (j 1 , j 2 , and j 3 ). Their deadlines and tardiness weights are (3,7,4) hours and (1.1, 0.3, 0.9), respectively. The N nodes (n 1 to n N ) can be configured with three VM types.…”

Section: Problem Statementmentioning

confidence: 99%

“…DL application training is computing intensive and for this reason it is usually backed by GPUs, which can perform matrix multiplications in parallel and are able to accelerate the models' training and evaluation in production environments [6]. As a matter of fact, CNN training is frequently offloaded from CPUs to GPUs [7] achieving a 5 to 40x performance gain [8]. It should not come as a surprise then that the GPU market that was already worth around 200 million USD in 2016 is experiencing an annual growth rate over 30% that analysts believe will remain unchanged until 2024 [9].…”

Section: Introductionmentioning

confidence: 99%

Optimizing on-demand GPUs in the Cloud for Deep Learning Applications Training

Jahani

Lattuada

Ciavotta

et al. 2019

2019 4th International Conference on Computing, Communications and Security (ICCCS)

View full text Add to dashboard Cite

Deep learning (DL) methods have recently gained popularity and been used in commonplace applications; voice and face recognition, among the others. Despite the growing popularity of DL and the associated hardware acceleration techniques, GPU-based systems still have very high costs. Moreover, while the cloud represents a cost-effective and flexible solution, in large settings operations costs can be further optimized by carefully managing and fostering resource sharing. This work addresses the online joint problem of capacity planning of virtual machines (VMs) and DL training jobs scheduling, and proposes a Mixed Integer Linear Programming (MILP) formulation. In particular, DL jobs are assumed to feature a deadline, while multiple VM types are available from a cloud provider catalog, and each VM has, possibly, multiple GPUs. Our solutions optimize the operations costs by (i) right-sizing the VM capacities; (ii) partitioning the set of GPUs among multiple concurrent jobs running on the same VM, and (iii) determining a deadline-aware job schedule. Our approach is evaluated using an ad-hoc simulator and a prototype environment, and compared against first-principle approaches, resulting in a cost reduction of 45-80%.

show abstract

“…Recently, machine learning has been intensively used in spam detection, speech and image detection, search engines, and disease and drug discoveries [20,21]. Applications of machine learning are not limited to the ones above; machine learning has been employed to predict the properties of different structural and material systems and to search for new materials with optimal designs [22,23,24,25] Additionally, Kim et al [28] argued that deep learning networks could be employed to capture the nonlinear hysteretic systems without compromising accuracy. Although the adopted approach is generic and is suitable for any system with hysteresis, they applied it to the prediction of structural responses under earthquake excitations.…”

Section: Introductionmentioning

confidence: 99%

Prediction and optimization of mechanical properties of composites using convolutional neural networks

Abueidda

Al-Masri

Ammourah

et al. 2019

Composite Structures

156

View full text Add to dashboard Cite

In this paper, we develop a convolutional neural network model to predict the mechanical properties of a two-dimensional checkerboard composite quantitatively. The checkerboard composite possesses two phases, one phase is soft and ductile while the other is stiff and brittle. The ground-truth data used in the training process are obtained from finite element analyses under the assumption of plane stress. Monte Carlo simulations and central limit theorem are used to find the size of the dataset needed. Once the training process is completed, the developed model is validated using data unseen during training. The developed neural network model captures the stiffness, strength, and toughness of checkerboard composites with high accuracy. Also, we integrate the developed model with a genetic algorithm optimizer to identify the optimal microstructural designs. The genetic algorithm optimizer adopted here has several operators, selection, crossover, mutation, and elitism. The optimizer converges to configurations with highly enhanced properties. For the case of the modulus and starting from randomly-initialized generation, the GA optimizer converges to the global maximum which involves no soft elements. Also, the GA optimiz- * ers, when used to maximize strength and toughness, tend towards having soft elements in the region next to the crack tip.

show abstract

Response prediction of nonlinear hysteretic systems by deep neural networks

Cited by 77 publications

References 23 publications

Pre‐ and post‐earthquake regional loss assessment using deep learning

Pre‐ and post‐earthquake regional loss assessment using deep learning

Optimizing on-demand GPUs in the Cloud for Deep Learning Applications Training

Prediction and optimization of mechanical properties of composites using convolutional neural networks

Contact Info

Product

Resources

About