Machine Learning-Based Substructuring Technique for Multi-Element Hybrid Simulation of Steel Braced Frames

Dr. Imanpour

Fardad Mokhtari


Hybrid simulation is an advanced structural testing method, which enables researchers to assess the response of structural systems by combining both experimental testing and numerical analysis. In hybrid simulation, the critical part of the structure expected to experience instability or high nonlinearity is tested physically in the laboratory whilst the rest of the structure is numerically analyzed using the finite element method. However, the results obtained from hybrid simulation can easily become biased in structures that consist of more than one critical element such as seismic fuses in multi-story structures or bridge piers where only one or a few - in the case of geographically-distributed hybrid simulations - of such critical elements are experimentally tested because of laboratory limitations. Furthermore, errors arising from test controller, hydraulic actuators, signal delays, etc. may affect reliability of hybrid simulation. To overcome these constraints of hybrid simulation, this paper proposes a numerical substructuring technique where the experimental test or continuum finite element analysis results are used to train critical elements such as seismic fuses with the aid of machine learning algorithms. A multi-storey buckling restrained braced frame is first selected and designed in accordance with the Canadian steel design standard. The inelastic cyclic response of the braces is then predicted using the proposed technique and fed to the hybrid numerical model of the frame that consists of elements other than the braces. Nonlinear dynamic analyses are finally performed to verify the reliability of the proposed technique when compared to the results obtained from a baseline pure numerical model of the frame. The results confirm that the proposed substructuring technique that benefits from a wealth of test data can appropriately simulate the nonlinear seismic response of buckling restrained braced frames.
Hybrid Simulation, Machine Learning, Artificial Intelligence, Seismic Response, Steel Braced Frames

Flowchart of the conventional hybrid simulation and the proposed machine learning-based hybrid simulation

Blue line indicates the results of conventional hybrid simulation while the red line indicates the results of machine learning-based hybrid simulation under a seismic excitation

Donadeo Innovation Centre for Engineering

University of Alberta

Edmonton, AB, Canada