Hybrid finite element and machine learning approach for covered linkway sheltering factor assessment

• Main Author: Tok, Wei Sheng
• Other Authors: Mir Feroskhan
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering; Covered linkway; Sheltering factor; Machine learning; Finite element
• Online Access: https://hdl.handle.net/10356/177581
• Institution: Nanyang Technological University

The proliferation of Unmanned Aerial Vehicle (UAV) technology has resulted in an increase in UAV-related incidents, raising concerns about potential collisions with covered linkways. However, the protective capabilities of these linkways against UAV impacts remain uncertain. Therefore, the objective of this study is to conduct a comprehensive evaluation of the covered linkway sheltering factor. Given the computationally intensive nature of Finite Element Analysis (FEA), this research explores the feasibility of adopting a hybrid approach that integrates FEA with machine learning techniques to enhance efficiency and accuracy in assessing sheltering factors. The research methodology is structured into four distinct phases. In Phase 1, FEA is employed to simulate UAV impacts on covered linkways, generating an initial dataset using mass and velocity inputs. In Phase 2, a Gaussian Process (GP) model is applied as a machine learning technique to predict energy absorption values based on the FEA dataset. In Phase 3, deep learning methodologies are integrated into the assessment process. An Artificial Neural Network (ANN) model is implemented to predict penetration outcomes of UAV impacts, while an ANN-Convolutional Neural Network (CNN) model is used to predict FEA deformation images resulting from these impacts. Finally, in Phase 4, spatial data analysis technique is applied to compute the overall sheltering factor within specific subzones in Singapore. The main results of this study indicate promising performance of the hybrid FEA-machine learning approach. The GP model achieved a Root Mean Squared Error (RMSE) of 953, demonstrating its ability to predict energy absorption values with reasonable accuracy. The ANN model exhibited a notable accuracy rate of 90% in predicting penetration outcomes, while the ANN-CNN model achieved a Mean Squared Error (MSE) of 0.05 in predicting deformation images resulting from UAV impacts. In conclusion, this research underscores the potential of integrating computational simulations with advanced machine learning techniques to assess the sheltering factor of covered linkways against UAV collisions. The findings provide valuable insights for engineering analysis and design, contributing to improved understanding and design optimization of protective structures in UAV-prone environments.