Braces are widely used in structural systems to provide lateral load resistance. Thus, the structural response of the concentrically braced frames under extreme events (such as earthquake, wind, blast) strongly depends on its brace behavior. In scientific literature, a significant amount of studies have been investigated the brace behavior both numerically and experimentally. However, due to the highly nonlinear and non-uniform characteristics of the braces, which also depends on the loading histories, it is still an open question on how to model their dynamic behaviors accurately \[1,2\]. A decent model has to capture bucking, deterioration, and failure due to low-cycle fatigue. These effects strongly depend on the nonlinear behavior of the braces, which is still not well understood, and thus difficult to numerically model \[1\]. There are recent efforts to improve the numerical model capabilities by benefiting from the high fidelity experimental data. 

Machine learning (ML) provides new opportunities for scientists to answer long-standing questions and serves promising solutions in many dynamic engineering problems. It is a developing field in structural engineering where a wide range of applications considered, such as predicting the global structural response under dynamic excitations \[3\]. In this study, we aim to further improve the analytical models by using the advantages of physics informed machine learning. For this purpose, small-scale brace responses, which are, tested both under cyclic loading and earthquake loading with real-time hybrid simulation will be used as the training dataset. Several deep neural network algorithms, such as one-dimensional convolutional neural networks (1D CONV), long-short term memory (LSTM) networks that could predict the dynamic responses of the braces will be compared. 

To summarize, this research seeks answers to the following scientific questions. (i) Which ML algorithm is capable and is more efficient to learn such a complex and nonlinear problem? (ii) Is it possible to artificially reproduce the brace behavior that is able to represent real physics to further decrease the number of necessary experiments? (iii) How can our findings be extended to the different engineering problems that are driven by similar nonlinear dynamics? To answer these questions, the presented methods will be validated by using experimental brace data. The capability of the ML algorithms to generate the brace behavior is evaluated by using root mean square error. Besides, the effects of tuning the hyper parameters on the models, such as layer numbers, neuron numbers, and the activation functions, will be presented. Finally, the ability to learn nonlinear dynamics by using deep neural network algorithms and their advantages are discussed.

References:

\[1\] E. E. Bas, M. A. Moustafa, G. Pekcan “*Compact Hybrid Simulation System for Research and Education: System Validation and New Applications for Braced Frames Seismic Testing”, *DOI: 10.1080/13632469.2020.1733138

\[2\] A. Hammad, M.A. Moustafa, “*Shake Table tests of special concentric braced frames under short and long duration earthquakes”,* Engineering Structures, * *DOI: https://doi.org/10.1016/j.engstruct.2019.109695

\[3\] E. E. Bas, M. A. Moustafa, D. Feil-Seifer, J. Blankenburg, *“Using a Machine Learning Approach for Computational Substructure in Real-Time Hybrid Simulation”*, presented in IMAC XXXVIII, Houston TX, 2020.


Predicting Nonlinear Seismic Response of Structural Braces Using Machine Learning

Division / Sponsors: 
Aerospace, Bioengineering, Applied Mechanics, Design Engineering Division, Advanced Energy Systems, Manufacturing Engineering Division, Noise Control and Acoustics, Nondestructive Evaluation, Diagnosis, and Prognosis (NDPD), Nuclear Engineering Division, Process Industries Division, Fluids Engineering, Heat Transfer, Materials, Microelectromechanical Systems, Safety Engineering and Risk Analysis, Computers & Information in Engineering


Paper Proceedings: https://asmedigitalcollection.asme.org/IMECE/IMECE2020/volume/84478

IMECE2020 - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.