[DOC] Discussion for Background Research #60
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Documentation Type
Thesis Chapter/Section
Description
The history for every changes should be documented here with make sense reasons, either from author's thoughts or advisors.
Current State
No response
Proposed Changes
Possible structure section and subsection:
Background
Identification Probem
Most existing SHM frameworks require dense sensor arrays, which are costly and impractical for many field applications.
Many high-performing models rely on deep learning techniques, which demand high computational resources and may not be interpretable or easy to deploy.
There is a lack of simple, generalizable approaches that balance minimal sensor usage with reliable damage localization.
Key Issue
To guide the direction of the study, the following key issues are addressed:
Can vibration signals captured only at the top and bottom sensors of a column group retain sufficient features to accurately classify structural damage?
Does pooling data from multiple column paths enhance model generalization, even with limited sensors per path?
Can a simple classical machine learning algorithm compete with more complex models when applied to this reduced-input scenario?
Problem Limitation
This study is limited to the QUGS dataset, which is based on a specific laboratory steel-frame structure under controlled conditions.
The trained model is not intended for transfer to different simulators or physical structures without retraining.
The machine learning scope is limited to classical algorithms (e.g., SVM), excluding neural networks and deep learning models.
The research does not explore varying types of damage or multi-class classification beyond what the dataset provides.
Purpose
To develop a simplified SHM pipeline that uses only a pair of sensors (top and bottom) from grouped columns of sensors (e.g., columns: (1,6,11,16,21,26)).
To treat each sensor group as a simplified one-dimensional beam and assess whether damage characteristics are preserved in the vibration energy between endpoints.
To append each grouped column as a singular dataset and conduct five separate tests, with each group serving as a validation set in turn.
To include the column group’s own data in the training set, thereby generating one inclusive model for all column groups.
To explore the feasibility of generalizing a single model across multiple column paths using endpoint sensor data.
Novelty
Introduces a new pipeline that reduces sensor dependency by reorganizing the layout into vertical groups and using only the first and last sensors.
Demonstrates that vibration energy across a simplified vertical path retains enough information for effective classification.
Uses cross-group augmentation: each group column contributes to training while also serving as an independent validation path.
Proposes a cost-effective, scalable SHM method suitable for implementation in constrained environments.
Offers an interpretable, classical machine learning-based solution as an alternative to black-box deep learning models.
Documentation Location
latex/chapters/01_introduction.tex
Priority
Critical (required for thesis)
Target Audience
Thesis Committee/Reviewers
References
No response
Additional Notes
No response
Tujuan -> dengan sepasang sensor (yang dilimitasi)
Tujuan -> Pipeline -> Sebaiknya tidak disebut secara ekspilisit karena akan diperlukan pembahasan time complxity Abdeljaber2017 dan mencoba melakukan eksperimen metodologi yang disampaikan oleh Abdeljaber tetapi hanya dengan sensor yang dilimitasi (2 sensor)
Jumlah point pada rumusan masalah seharusnya sama dengan point tujuan dan kesimpulan (rata-rata dibuat dalam point: tugas akhir atau paragraf: paper.
Manfaat:
Dasar Teori: