Please use this identifier to cite or link to this item:
|Title:||Development of a graph convolutional network-based surface quality monitoring approach||Authors:||Peh, Gerald Zong Xian||Keywords:||Engineering::Mechanical engineering||Issue Date:||2022||Publisher:||Nanyang Technological University||Source:||Peh, G. Z. X. (2022). Development of a graph convolutional network-based surface quality monitoring approach. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/157257||Project:||B049||Abstract:||Many traditional quality monitoring approaches faced issues such as a huge number of uncontrollable parameters which leads to prediction inaccuracy. Other forms of modern monitoring system utilize Deep Learning (DL) models such Artificial Neural Networks (ANNs) and Convolutional Neural Networks (CNN). However, such models are unable to mine complex relations between each signal. To counter this issue, this study would introduce Graph Convolutional Networks (GCNs) to prediction of surface quality where it takes graph-structured data as an input instead of the usual Euclidean structured data. Such models can learn complex relationships which allows better feature representation of each sampled data. However, most GCNs have certain limitations such as a fixed kernel size which prevents learning of relationship from multi-hop neighborhood domains. Therefore, this project would implement a Multi-Hop Graph Convolutional Network (MHGCN) model to observe whether fused features from different receptive fields would provide better prediction results. The data samples are converted into weighted graphs to establish different importance between each sample relations. Also, the proposed model would utilize an attention mechanism to mine relationships among different hop domains and select the important ones. Simple averaging ensemble learning would be implemented to combine multiple learners and improve learning process. To verify the effectiveness of the proposed MHGCN model, it is compared with other DL and GCN models and the results shows that the MHGCN model with attention mechanism has the best performance.||URI:||https://hdl.handle.net/10356/157257||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||MAE Student Reports (FYP/IA/PA/PI)|
Updated on May 27, 2022
Updated on May 27, 2022
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.