Optimized Neural Networks for Structural Damage Prediction Based on Modal Analysis

EasyChair Preprint no. 11186

Abstract

Damage detection and localization play a pivotal role in structural health monitoring endeavors. While the application of Artificial Neural Networks (ANNs) has yielded success in identifying damage within civil and mechanical infrastructures, it's worth noting that these ANNs come with certain inherent limitations. However, a promising avenue for enhancing the efficacy of ANNs lies in the modification of their architecture and refining their training strategies.

In this study, the authors introduce a novel approach that leverages a metaheuristic algorithm known as the Grasshopper Optimization Algorithm (GOA) to craft an optimized ANN tailored for the task of predicting multiple types of damage in aluminum bars. The input parameters for this ANN are drawn from the natural frequencies of the structure, while the output is focused on predicting the depths of cracks within the material. To generate the necessary dataset, an advanced Finite Element Model (FEM) is employed, which is adapted to accommodate various crack depths. This data collection is facilitated through the utilization of a simulation tool.

To ascertain the robustness of this proposed technique, real-world experimental data derived from the analysis of cracked beams is gathered, encompassing a range of distinct crack depths. The performance of this newly introduced approach is subsequently benchmarked against alternative methods that also harness metaheuristic algorithms, specifically the Ants Colony Optimization Algorithm (ACOI) and the Genetic Algorithm (GA).

Remarkably, the results showcase that the innovative methodology put forth by the authors exhibits a high level of predictive accuracy in the realm of damage prognosis. This underscores the potential of the Grasshoper Optimization Algorithm in optimizing the architecture of ANNs to effectively forecast various forms of damage in structural components.

Keyphrases: Damage prediction, neural networks, Optimization

@Booklet{EasyChair:11186,
  author = {Abdelwahhab Khatir and Roberto Capozucca and Erica Magagnini and Abdelmoumen Oulad Brahim and Laith Abualigah},
  title = {Optimized Neural Networks for Structural Damage Prediction Based on Modal Analysis},
  howpublished = {EasyChair Preprint no. 11186},

  year = {EasyChair, 2023}}