Translator Disclaimer
17 May 2005 A new approach to designing multilayer feedforward neural networks for modeling nonlinear restoring forces
Author Affiliations +
Abstract
This study investigates the possibility of injecting parametric features into nonparametric identification techniques like neural networks in modeling nonlinear dynamic restoring forces. This affords the potential of creating relationships between model parameters in data-driven techniques and phenomenological behaviors in physics-based modeling, which is prompted by the needs in structural health monitoring and damage detections. Here a linear sum of sigmoidal basis functions is used in modeling nonlinear hysteretic restoring forces of single-degree-of-freedom oscillators under the force-state mapping formulation to showcase this idea. A constructive approach is proposed to guide the neural network initial design, where the number of hidden layers and hidden nodes as well as the initial values of the weights and biases are decided upon the characteristics of the nonlinear restoring force to be modeled rather than through indiscriminate numerical initialization schemes. Numerical simulations are presented to demonstrate the efficiency and engineered feature of this approach. A training example is provided to show that this approach enables neural networks to carry either physical or phenomenological "meaning" while remaining adaptive and thus powerful in system identification.
© (2005) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Jin-Song Pei and Andrew W. Smyth "A new approach to designing multilayer feedforward neural networks for modeling nonlinear restoring forces", Proc. SPIE 5765, Smart Structures and Materials 2005: Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems, (17 May 2005); https://doi.org/10.1117/12.600301
PROCEEDINGS
9 PAGES


SHARE
Advertisement
Advertisement
Back to Top