Hybrid Mathematical-Informational Modeling of Structural Systems

Abstract eng:
Conventional modeling and computational simulation in mechanics and structures are based on mathematical or empirical equations that have been developed to represent the response of components or systems. Such idealization may lead to representations that exclude significant aspects of system behavior with specific loads and boundary conditions. An alternative approach is to represent system response based on information contained in the observed data. The data modeling approach is fundamentally different to the aforementioned mechanical modeling option. Computational intelligence methods have made the data modeling approach possible and effective. The information about the underlying mechanics is extracted from the observed system response data and stored in neural networks. Trained neural networks can then be used in the simulations of the system. This approach also has its limitations. This paper reports on a new hybrid modeling and simulation method where conventional mathematical models (of portions of the system for which an acceptably accurate solution exists) are complemented by the informational models (of portions of the system that are more challenging to model mathematically). The proposed hybrid modeling and simulation approach has potential applications in a wide range of fields in computational mechanics and structural analysis. The effectiveness of the hybrid method is illustrated for the hysteretic behavior of beam-column connections in steel frames where aspects of the response that are not amenable to mechanics- based modeling are represented by trained neural networks. The potential of the method in several fields of engineering and science is also explored.

• Export as: BibTeX | MARC | MARCXML | DC | EndNote | NLM | RefWorks
• Add to your basket