SGER: Neural Networked Finite Element Methodology for Health Monitored Structural Systems

Kansas State University, Manhattan KS

Investigators

Abstract

PI: Yacoub M. Najjar & Hayder A. Rasheed; Kansas State University The objective of this proposed research is to transplant a Neural Network algorithm within a finite element analysis procedure to take over the function of its heart after a small jumpstart. The Neural Network will adapt to the internal functionality needed to re-produce the overall response of the body or the system, initially known from some external health monitoring measurements, through self-adjusting/learning analysis trials. This will build the memory and intelligence necessary to deal with similar analysis situations pretty much the same way the human mind and body interact and inter-respond without an interpreter. The specific objective of this research proposal is to develop, formulate, and implement a proof-of-concept for an innovative Neural Network-Finite Element (NN-FE) self-training approach. This approach can be used to accurately characterize the internal mechanical behavior of health monitored engineering systems such as buildings, bridges, offshore platforms, airplanes, space shuttles and other structures. Pre-generated finite element-based data-rich results simulating monitored 2D structures will be used to develop and validate the NN-FE technique. Accordingly, each database containing load and displacement history at some health monitoring/control points of the system could be used to generate the NN-FE internal behavior characterization model. The developed scheme will operate in two distinctive-alternating modes. First, it will operate in a predictor analysis mode to capture approximate stresses and strains at the integration points. Subsequently, and based on deviations between predictions and actual displacement values at the health monitoring points, the NN-FE self-adapting algorithm will operate in a training mode in order to reduce these deviations. As the algorithm is repeatedly applied in analysis and training modes to eliminate such deviations, it develops self-learning artificial intelligence and becomes smarter at making more accurate predictions. Intellectual Merit: The proposed research addresses the important issue of the self-sufficient and reliable internal characterization of the mechanical response of health monitored structural systems. Success of this research will lead to better assessment of current and future structures. Accordingly, this will lead to more economical designs, protect the national investments, and ensure higher safety measures to the public. More importantly, by integrating anticipated future development in non-contact measurements, this work is envisioned to pave the way to a new era of analysis in which computer systems can remotely scan objects for response data and use them to analyze the structural system by incorporating the self-training Neural Networked finite element procedure. This is expected to impact all aspects of engineering, science and life on the planet Earth, including aerospace, manufacturing, civil, mechanical, and chemical engineering fields. Furthermore, it may be extended for utilization in aeronautics and astronautics within man's quest to conquer the outer space. Broader Impact: The project.s broader impacts include the following aspects: 1) Full involvement of a PhD graduate student. Therefore, providing the student with intensive training/research experience to address similar complex engineering systems. 2) The proposed exploratory integrated research will most likely re-shape the interacting information technology-structural engineering mechanics educational curriculum. This will definitely intrigue the future engineering leaders to address other issues in similar integrated and systematic way. Moreover, it will help mature the young graduates into scholars with hands-on experience to tackle complex health monitored structural engineering systems.