nuluh/thesis
1
0
Fork 0
Code Issues 47 Pull Requests Actions Packages Projects Releases 1 Wiki Activity

[DOC] Original Paper of The Dataset #67

New Issue
Open
opened 2025-05-14 12:37:25 +00:00 by nuluh · 8 comments
nuluh commented 2025-05-14 12:37:25 +00:00 (Migrated from github.com)
Copy Link

Documentation Type

Thesis Chapter/Section

Description

This is the discussion original paper of the dataset that should be included and explained in-depth in the literature review #64 and create a brief of the paper for background #60

Current State

No response

Proposed Changes

🔍 1. Study Motivation & Problem Statement

Question:

  • What problem in SHM/damage detection is this paper addressing, and why is it significant?

Starter sentence:

The study by [Author(s), Year] addresses a key limitation in traditional structural health monitoring systems—namely, the reliance on hand-crafted features that often lack generalizability and demand high computational resources.

🧠 2. Proposed Method/Innovation

Question:

  • What is the core innovation or proposed solution in the paper?
  • How does it differ from or improve upon previous methods?

Starter sentence:

To overcome these limitations, the authors propose a novel 1D Convolutional Neural Network (CNN)-based approach that integrates feature extraction and classification into a unified learning framework, enabling real-time damage detection.

📊 3. Dataset and Experimental Setup

Question:

  • What structure is used to validate the approach (e.g., grandstand simulator)?
  • How was data collected (type of sensors, acceleration, noise type)?
  • Is the dataset publicly available?

Starter sentence:

The proposed method was validated using vibration data collected from a grandstand simulator, which served as a realistic testbed for assessing structural damage under various conditions.
The dataset, made publicly available by the authors, includes [insert sensor configuration, data type, etc.], making it a valuable resource for benchmarking alternative damage detection algorithms.

🚀 4. Performance and Results

Question:

  • How well did their method perform?
  • What were the key metrics or findings?
  • How did it compare to other methods, if any?

Starter sentence:

Experimental results demonstrated that the proposed CNN approach achieved outstanding accuracy in both damage detection and localization tasks, while maintaining a high level of computational efficiency suitable for real-time applications.

🔄 5. Relevance to Your Thesis

Question:

  • How does this paper’s dataset, approach, or finding relate to your thesis?
  • Are you using their data, method, or both?
  • What part of your thesis is inspired by or built upon their work?

Starter sentence:

In the context of this thesis, the dataset introduced by [Author(s)] was utilized to [state how you used it—e.g., apply a classical machine learning pipeline using SVM for zone-based damage localization], providing a comparative framework against modern deep learning-based SHM systems.

Documentation Location

latex/chapters/id/02_literature_review.tex

Priority

Critical (required for thesis)

Target Audience

Thesis Committee/Reviewers

References

  • Abdeljaber, O., Avci, O., Kiranyaz, S., Gabbouj, M., & Inman, D. J. (2017). Real-time vibration-based structural damage detection using one-dimensional convolutional neural networks. Journal of Sound and Vibration, 388, 154–170. https://doi.org/10.1016/j.jsv.2016.10.043

Additional Notes

No response

### Documentation Type Thesis Chapter/Section ### Description This is the discussion original paper of the dataset that should be included and explained in-depth in the literature review #64 and create a brief of the paper for background #60 ### Current State _No response_ ### Proposed Changes ## 🔍 **1. Study Motivation & Problem Statement** **Question:** * What problem in SHM/damage detection is this paper addressing, and why is it significant? **Starter sentence:** > The study by \[Author(s), Year] addresses a key limitation in traditional structural health monitoring systems—namely, the reliance on hand-crafted features that often lack generalizability and demand high computational resources. --- ## 🧠 **2. Proposed Method/Innovation** **Question:** * What is the core innovation or proposed solution in the paper? * How does it differ from or improve upon previous methods? **Starter sentence:** > To overcome these limitations, the authors propose a novel 1D Convolutional Neural Network (CNN)-based approach that integrates feature extraction and classification into a unified learning framework, enabling real-time damage detection. --- ## 📊 **3. Dataset and Experimental Setup** **Question:** * What structure is used to validate the approach (e.g., grandstand simulator)? * How was data collected (type of sensors, acceleration, noise type)? * Is the dataset publicly available? **Starter sentence:** > The proposed method was validated using vibration data collected from a grandstand simulator, which served as a realistic testbed for assessing structural damage under various conditions. > The dataset, made publicly available by the authors, includes \[insert sensor configuration, data type, etc.], making it a valuable resource for benchmarking alternative damage detection algorithms. --- ## 🚀 **4. Performance and Results** **Question:** * How well did their method perform? * What were the key metrics or findings? * How did it compare to other methods, if any? **Starter sentence:** > Experimental results demonstrated that the proposed CNN approach achieved outstanding accuracy in both damage detection and localization tasks, while maintaining a high level of computational efficiency suitable for real-time applications. --- ## 🔄 **5. Relevance to Your Thesis** **Question:** * How does this paper’s dataset, approach, or finding relate to your thesis? * Are you using their data, method, or both? * What part of your thesis is inspired by or built upon their work? **Starter sentence:** > In the context of this thesis, the dataset introduced by \[Author(s)] was utilized to \[state how you used it—e.g., apply a classical machine learning pipeline using SVM for zone-based damage localization], providing a comparative framework against modern deep learning-based SHM systems. ### Documentation Location latex/chapters/id/02_literature_review.tex ### Priority Critical (required for thesis) ### Target Audience Thesis Committee/Reviewers ### References - Abdeljaber, O., Avci, O., Kiranyaz, S., Gabbouj, M., & Inman, D. J. (2017). Real-time vibration-based structural damage detection using one-dimensional convolutional neural networks. Journal of Sound and Vibration, 388, 154–170. https://doi.org/10.1016/j.jsv.2016.10.043 ### Additional Notes _No response_
nuluh commented 2025-05-14 13:37:15 +00:00 (Migrated from github.com)
Copy Link

Answers Question 1**:

"What problem in SHM/damage detection is this paper addressing, and why is it significant?"

Problem Identification

"It is imperative to extract damage-sensitive features..."
"Fixed features/classifiers... may not optimally characterize the acquired signal..."
"Feature extraction usually turns out to be a computationally costly operation..."

These clearly outline the problems:

  • Manual/trial-and-error feature selection
  • Poor generalizability
  • High computational cost
  • Incompatibility with real-time SHM applications

Significance

"Cannot accomplish a reliable performance level for damage detection"
"May hinder the usage... for a real-time SHM application"

These lines highlight the urgency and relevance of solving the above problems, especially for real-time damage detection.

Starter Sentence for Literature Review (Revised from Step 1):

Traditional structural health monitoring methods often rely on hand-crafted features and manually tuned classifiers, which pose challenges in terms of generalization, reliability, and computational efficiency. As highlighted by [Author(s), Year], these approaches frequently require a trial-and-error process for feature and classifier selection, which not only reduces their robustness across structures but also hinders their deployment in real-time applications due to the computational load of the feature extraction phase.

Answers Question 1**: **"What problem in SHM/damage detection is this paper addressing, and why is it significant?"** #### **Problem Identification** > "It is imperative to extract damage-sensitive features..." > "Fixed features/classifiers... may not optimally characterize the acquired signal..." > "Feature extraction usually turns out to be a computationally costly operation..." These clearly outline the **problems**: * Manual/trial-and-error feature selection * Poor generalizability * High computational cost * Incompatibility with real-time SHM applications #### **Significance** > "Cannot accomplish a reliable performance level for damage detection" > "May hinder the usage... for a real-time SHM application" These lines highlight the **urgency and relevance** of solving the above problems, especially for real-time damage detection. --- Starter Sentence for Literature Review (Revised from Step 1): > Traditional structural health monitoring methods often rely on hand-crafted features and manually tuned classifiers, which pose challenges in terms of generalization, reliability, and computational efficiency. As highlighted by \[Author(s), Year], these approaches frequently require a trial-and-error process for feature and classifier selection, which not only reduces their robustness across structures but also hinders their deployment in real-time applications due to the computational load of the feature extraction phase.
nuluh commented 2025-05-14 13:58:10 +00:00 (Migrated from github.com)
Copy Link

Answer question 2:

  • What is the core innovation or proposed solution in the paper?
  • How does it differ from or improve upon previous methods?

🔧 Key Points from Section 2 (Related work):

  1. Traditional ML Methods in SHM:

    • Parametric methods: mostly ANNs, OS-ELM, PNNs, FNNs.
    • Nonparametric methods: rely on signal-processing-based feature extraction (e.g., PCA, wavelets, AR modeling, SOMs, NARX) followed by a separate classifier (e.g., SVM, ANN, SVD).
    • Limitations: require careful manual feature extraction and may not generalize well.

  2. CNNs Offer Integrated Learning:

    • CNNs have shown state-of-the-art performance in deep learning fields (e.g., image and ECG signal classification).
    • Their convolutional layers automatically extract optimal features, unlike traditional hand-crafted techniques.
    • The feedforward layers perform classification in a single end-to-end training loop using backpropagation.

  3. 1D CNNs are now proven effective for time-series signals like ECG, and in this study, applied for structural vibration signals.

Starter Sentence Based on These Points:

Traditional structural damage detection systems have largely depended on either parametric classifiers such as artificial neural networks (ANNs), probabilistic neural networks (PNNs), or fuzzy systems, or nonparametric approaches that extract hand-crafted features through statistical or signal processing techniques like PCA, wavelet transforms, and autoregressive modeling. While these methods have shown some success, they are often limited by their reliance on manually engineered features and a two-stage pipeline that separates feature extraction from classification. In contrast, the study by [Author(s), Year] introduces a novel approach leveraging 1D Convolutional Neural Networks (CNNs), which unify both tasks into a single, end-to-end architecture. By automatically learning damage-sensitive features directly from raw vibration signals through its convolutional layers, and performing classification via fully connected layers, the proposed method reduces the need for domain-specific preprocessing and significantly enhances adaptability across different structural conditions.

**Answer question 2**: * What is the core innovation or proposed solution in the paper? * How does it differ from or improve upon previous methods? --- ### 🔧 Key Points from Section 2 (Related work): 1. **Traditional ML Methods in SHM:** * Parametric methods: mostly ANNs, OS-ELM, PNNs, FNNs. * Nonparametric methods: rely on signal-processing-based feature extraction (e.g., PCA, wavelets, AR modeling, SOMs, NARX) followed by a separate classifier (e.g., SVM, ANN, SVD). * Limitations: require careful manual feature extraction and may not generalize well. 2. **CNNs Offer Integrated Learning:** * CNNs have shown state-of-the-art performance in deep learning fields (e.g., image and ECG signal classification). * Their convolutional layers **automatically extract optimal features**, unlike traditional hand-crafted techniques. * The feedforward layers perform classification in a single end-to-end training loop using backpropagation. 3. **1D CNNs are now proven effective for time-series signals like ECG, and in this study, applied for structural vibration signals.** --- ### Starter Sentence Based on These Points: > Traditional structural damage detection systems have largely depended on either parametric classifiers such as artificial neural networks (ANNs), probabilistic neural networks (PNNs), or fuzzy systems, or nonparametric approaches that extract hand-crafted features through statistical or signal processing techniques like PCA, wavelet transforms, and autoregressive modeling. While these methods have shown some success, they are often limited by their reliance on manually engineered features and a two-stage pipeline that separates feature extraction from classification. In contrast, the study by \[Author(s), Year] introduces a novel approach leveraging 1D Convolutional Neural Networks (CNNs), which unify both tasks into a single, end-to-end architecture. By automatically learning damage-sensitive features directly from raw vibration signals through its convolutional layers, and performing classification via fully connected layers, the proposed method reduces the need for domain-specific preprocessing and significantly enhances adaptability across different structural conditions.
nuluh commented 2025-05-14 14:10:10 +00:00 (Migrated from github.com)
Copy Link

Answer question 3:

Q: What structure is used to validate the approach (e.g., grandstand simulator)?
Q: How was data collected (type of sensors, acceleration, noise type)?
Q: Is the dataset publicly available?

Completed Starter Sentences:

The proposed method was validated using a large-scale grandstand simulator constructed at Qatar University, designed to reflect the structural configuration and vibration behavior of real stadia. The simulator, measuring 4.2 m × 4.2 m, consists of 8 main girders and 25 removable filler beams supported by 4 columns. This modular design allows simulation of various structural damage scenarios, such as loosening or replacing beam-to-girder connections. To simulate slight damage cases, the authors introduced connection loosening at one or more of the 30 beam-to-girder joints. For data acquisition, 30 accelerometers—27 PCB model 393B04 and 3 B&K model 8344—were mounted using magnetic bases directly onto the steel joints. Excitation was applied using a modal shaker (Model 2100E11), powered by a SmartAmp 2100E21-400 amplifier, with input and output signals recorded using dual 16-channel data acquisition systems. The authors made the resulting vibration dataset publicly available, offering a realistic and high-resolution benchmark for further research in vibration-based structural damage detection.

The experimental validation was carried out in two phases. In the first phase, only a single girder was monitored (5 joints), and in the second phase, the entire structure was tested (30 joints). To generate the training dataset, the signals were first recorded under random excitation, with a sampling frequency of 1024 Hz and a recording duration of 256 seconds, resulting in 262,144 samples per signal. The data for each joint were then segmented into frames of 128 samples, producing 2048 frames per joint. For each joint, these frames were divided into undamaged and damaged categories based on the state of the joints. The undamaged data were obtained from experiments where all joints were undamaged, while damaged data were acquired by loosening the bolts of specific joints.

To ensure the effectiveness of the CNNs, each joint’s data was processed and divided into training vectors. A critical step in the data preparation was to balance the undamaged and damaged frames for each joint. The undamaged frames, U, and damaged frames, D, were grouped separately and normalized for input into the CNNs. The final training set for each joint consisted of equal numbers of undamaged and damaged frames. The dataset was shuffled to ensure a balanced distribution, and only the first portion of the frames were used for training, maintaining a consistent structure across all joint datasets.

Answer question 3: **Q: What structure is used to validate the approach (e.g., grandstand simulator)?** **Q: How was data collected (type of sensors, acceleration, noise type)?** **Q: Is the dataset publicly available?** **Completed Starter Sentences:** > The proposed method was validated using a large-scale grandstand simulator constructed at Qatar University, designed to reflect the structural configuration and vibration behavior of real stadia. The simulator, measuring 4.2 m × 4.2 m, consists of 8 main girders and 25 removable filler beams supported by 4 columns. This modular design allows simulation of various structural damage scenarios, such as loosening or replacing beam-to-girder connections. To simulate slight damage cases, the authors introduced connection loosening at one or more of the 30 beam-to-girder joints. For data acquisition, 30 accelerometers—27 PCB model 393B04 and 3 B\&K model 8344—were mounted using magnetic bases directly onto the steel joints. Excitation was applied using a modal shaker (Model 2100E11), powered by a SmartAmp 2100E21-400 amplifier, with input and output signals recorded using dual 16-channel data acquisition systems. The authors made the resulting vibration dataset publicly available, offering a realistic and high-resolution benchmark for further research in vibration-based structural damage detection. > The experimental validation was carried out in two phases. In the first phase, only a single girder was monitored (5 joints), and in the second phase, the entire structure was tested (30 joints). To generate the training dataset, the signals were first recorded under random excitation, with a sampling frequency of 1024 Hz and a recording duration of 256 seconds, resulting in 262,144 samples per signal. The data for each joint were then segmented into frames of 128 samples, producing 2048 frames per joint. For each joint, these frames were divided into undamaged and damaged categories based on the state of the joints. The undamaged data were obtained from experiments where all joints were undamaged, while damaged data were acquired by loosening the bolts of specific joints. > To ensure the effectiveness of the CNNs, each joint’s data was processed and divided into training vectors. A critical step in the data preparation was to balance the undamaged and damaged frames for each joint. The undamaged frames, U, and damaged frames, D, were grouped separately and normalized for input into the CNNs. The final training set for each joint consisted of equal numbers of undamaged and damaged frames. The dataset was shuffled to ensure a balanced distribution, and only the first portion of the frames were used for training, maintaining a consistent structure across all joint datasets.
nuluh commented 2025-05-14 14:12:16 +00:00 (Migrated from github.com)
Copy Link

Answer question 3:

Q: What structure is used to validate the approach (e.g., grandstand simulator)? Q: How was data collected (type of sensors, acceleration, noise type)? Q: Is the dataset publicly available?

Completed Starter Sentences:

The proposed method was validated using a large-scale grandstand simulator constructed at Qatar University, designed to reflect the structural configuration and vibration behavior of real stadia. The simulator, measuring 4.2 m × 4.2 m, consists of 8 main girders and 25 removable filler beams supported by 4 columns. This modular design allows simulation of various structural damage scenarios, such as loosening or replacing beam-to-girder connections. To simulate slight damage cases, the authors introduced connection loosening at one or more of the 30 beam-to-girder joints. For data acquisition, 30 accelerometers—27 PCB model 393B04 and 3 B&K model 8344—were mounted using magnetic bases directly onto the steel joints. Excitation was applied using a modal shaker (Model 2100E11), powered by a SmartAmp 2100E21-400 amplifier, with input and output signals recorded using dual 16-channel data acquisition systems. The authors made the resulting vibration dataset publicly available, offering a realistic and high-resolution benchmark for further research in vibration-based structural damage detection.

The detail model of equipment serial type and the CNN architecture details setup shouldn't be included to keep the readibility and simplicity:

The proposed method was validated using a large-scale grandstand simulator constructed at Qatar University, designed to reflect the structural configuration and vibration behavior of real stadia. The simulator, measuring 4.2 m × 4.2 m, consists of 8 main girders and 25 removable filler beams supported by 4 columns. This modular design allows simulation of various structural damage scenarios, such as loosening or replacing beam-to-girder connections. To simulate slight damage cases, the authors introduced connection loosening at one or more of the 30 beam-to-girder joints. For data acquisition, 30 accelerometers were mounted using magnetic bases directly onto the steel joints. Excitation was applied using a modal shaker, with input and output signals recorded using dual 16-channel data acquisition systems. The authors made the resulting vibration dataset publicly available, offering a realistic and high-resolution benchmark for further research in vibration-based structural damage detection.

The experimental validation was carried out in two phases. In the first phase, only a single girder was monitored (5 joints), and in the second phase, the entire structure was tested (30 joints). To generate the training dataset, the signals were first recorded under random excitation, with a sampling frequency of 1024 Hz and a recording duration of 256 seconds, resulting in 262,144 samples per signal. The data for each joint were then segmented into frames of 128 samples, producing 2048 frames per joint. For each joint, these frames were divided into undamaged and damaged categories based on the state of the joints. The undamaged data were obtained from experiments where all joints were undamaged, while damaged data were acquired by loosening the bolts of specific joints.

> Answer question 3: > > **Q: What structure is used to validate the approach (e.g., grandstand simulator)?** **Q: How was data collected (type of sensors, acceleration, noise type)?** **Q: Is the dataset publicly available?** > > **Completed Starter Sentences:** > > > The proposed method was validated using a large-scale grandstand simulator constructed at Qatar University, designed to reflect the structural configuration and vibration behavior of real stadia. The simulator, measuring 4.2 m × 4.2 m, consists of 8 main girders and 25 removable filler beams supported by 4 columns. This modular design allows simulation of various structural damage scenarios, such as loosening or replacing beam-to-girder connections. To simulate slight damage cases, the authors introduced connection loosening at one or more of the 30 beam-to-girder joints. For data acquisition, 30 accelerometers—27 PCB model 393B04 and 3 B&K model 8344—were mounted using magnetic bases directly onto the steel joints. Excitation was applied using a modal shaker (Model 2100E11), powered by a SmartAmp 2100E21-400 amplifier, with input and output signals recorded using dual 16-channel data acquisition systems. The authors made the resulting vibration dataset publicly available, offering a realistic and high-resolution benchmark for further research in vibration-based structural damage detection. The detail model of equipment serial type and the CNN architecture details setup shouldn't be included to keep the readibility and simplicity: > The proposed method was validated using a large-scale grandstand simulator constructed at Qatar University, designed to reflect the structural configuration and vibration behavior of real stadia. The simulator, measuring 4.2 m × 4.2 m, consists of 8 main girders and 25 removable filler beams supported by 4 columns. This modular design allows simulation of various structural damage scenarios, such as loosening or replacing beam-to-girder connections. To simulate slight damage cases, the authors introduced connection loosening at one or more of the 30 beam-to-girder joints. For data acquisition, 30 accelerometers were mounted using magnetic bases directly onto the steel joints. Excitation was applied using a modal shaker, with input and output signals recorded using dual 16-channel data acquisition systems. The authors made the resulting vibration dataset publicly available, offering a realistic and high-resolution benchmark for further research in vibration-based structural damage detection. > The experimental validation was carried out in two phases. In the first phase, only a single girder was monitored (5 joints), and in the second phase, the entire structure was tested (30 joints). To generate the training dataset, the signals were first recorded under random excitation, with a sampling frequency of 1024 Hz and a recording duration of 256 seconds, resulting in 262,144 samples per signal. The data for each joint were then segmented into frames of 128 samples, producing 2048 frames per joint. For each joint, these frames were divided into undamaged and damaged categories based on the state of the joints. The undamaged data were obtained from experiments where all joints were undamaged, while damaged data were acquired by loosening the bolts of specific joints.
nuluh commented 2025-05-14 17:39:14 +00:00 (Migrated from github.com)
Copy Link

The detail model of equipment serial type and the CNN architecture details setup shouldn't be included to keep the readibility and simplicity:

The proposed method was validated using a large-scale grandstand simulator constructed at Qatar University, designed to reflect the structural configuration and vibration behavior of real stadia. The simulator, measuring 4.2 m × 4.2 m, consists of 8 main girders and 25 removable filler beams supported by 4 columns. This modular design allows simulation of various structural damage scenarios, such as loosening or replacing beam-to-girder connections. To simulate slight damage cases, the authors introduced connection loosening at one or more of the 30 beam-to-girder joints. For data acquisition, 30 accelerometers were mounted using magnetic bases directly onto the steel joints. Excitation was applied using a modal shaker, with input and output signals recorded using dual 16-channel data acquisition systems. The authors made the resulting vibration dataset publicly available, offering a realistic and high-resolution benchmark for further research in vibration-based structural damage detection.

The experimental validation was carried out in two phases. In the first phase, only a single girder was monitored (5 joints), and in the second phase, the entire structure was tested (30 joints). To generate the training dataset, the signals were first recorded under random excitation, with a sampling frequency of 1024 Hz and a recording duration of 256 seconds, resulting in 262,144 samples per signal. The data for each joint were then segmented into frames of 128 samples, producing 2048 frames per joint. For each joint, these frames were divided into undamaged and damaged categories based on the state of the joints. The undamaged data were obtained from experiments where all joints were undamaged, while damaged data were acquired by loosening the bolts of specific joints.

Blend into one paragraph:

The proposed method was validated using vibration data collected from a purpose-built grandstand simulator at Qatar University, designed to mimic realistic structural configurations found in modern stadia. The steel-framed structure was equipped with 30 accelerometers placed at each beam-to-girder joint to capture localized vibration responses. Structural damage was simulated experimentally by loosening bolts at beam connections, enabling precise control over damage severity and location. Data were collected using a modal shaker that applied band-limited white noise (0–512 Hz), while acceleration signals were sampled at 1024 Hz over 256 seconds, yielding 262,144 time-domain samples per experiment. For each joint, signals were segmented into 128-sample frames, resulting in 2048 frames, evenly split between undamaged and damaged classes for training. Each joint had its own independently trained 1D CNN, creating a decentralized detection architecture that only required local acceleration input. This dataset, rich in structural variability and high-resolution signal data, provides an essential benchmark for evaluating modern SHM algorithms.

> The detail model of equipment serial type and the CNN architecture details setup shouldn't be included to keep the readibility and simplicity: > > > The proposed method was validated using a large-scale grandstand simulator constructed at Qatar University, designed to reflect the structural configuration and vibration behavior of real stadia. The simulator, measuring 4.2 m × 4.2 m, consists of 8 main girders and 25 removable filler beams supported by 4 columns. This modular design allows simulation of various structural damage scenarios, such as loosening or replacing beam-to-girder connections. To simulate slight damage cases, the authors introduced connection loosening at one or more of the 30 beam-to-girder joints. For data acquisition, 30 accelerometers were mounted using magnetic bases directly onto the steel joints. Excitation was applied using a modal shaker, with input and output signals recorded using dual 16-channel data acquisition systems. The authors made the resulting vibration dataset publicly available, offering a realistic and high-resolution benchmark for further research in vibration-based structural damage detection. > > > The experimental validation was carried out in two phases. In the first phase, only a single girder was monitored (5 joints), and in the second phase, the entire structure was tested (30 joints). To generate the training dataset, the signals were first recorded under random excitation, with a sampling frequency of 1024 Hz and a recording duration of 256 seconds, resulting in 262,144 samples per signal. The data for each joint were then segmented into frames of 128 samples, producing 2048 frames per joint. For each joint, these frames were divided into undamaged and damaged categories based on the state of the joints. The undamaged data were obtained from experiments where all joints were undamaged, while damaged data were acquired by loosening the bolts of specific joints. Blend into one paragraph: > The proposed method was validated using vibration data collected from a purpose-built grandstand simulator at Qatar University, designed to mimic realistic structural configurations found in modern stadia. The steel-framed structure was equipped with 30 accelerometers placed at each beam-to-girder joint to capture localized vibration responses. Structural damage was simulated experimentally by loosening bolts at beam connections, enabling precise control over damage severity and location. Data were collected using a modal shaker that applied band-limited white noise (0–512 Hz), while acceleration signals were sampled at 1024 Hz over 256 seconds, yielding 262,144 time-domain samples per experiment. For each joint, signals were segmented into 128-sample frames, resulting in 2048 frames, evenly split between undamaged and damaged classes for training. Each joint had its own independently trained 1D CNN, creating a decentralized detection architecture that only required local acceleration input. This dataset, rich in structural variability and high-resolution signal data, provides an essential benchmark for evaluating modern SHM algorithms.
nuluh commented 2025-05-14 17:43:18 +00:00 (Migrated from github.com)
Copy Link

Answer to question 4:

  • How well did their method perform?
  • What were the key metrics or findings?
  • How did it compare to other methods, if any?

The experimental evaluation was conducted in two phases. In the first phase, focusing on a single girder with 5 joints, the trained CNNs accurately distinguished between undamaged, single-damage, and double-damage cases, consistently assigning high probability-of-damage (PoD) scores to damaged joints and low values to intact ones. In the second phase, the method scaled up to monitor all 30 joints across the full structure, involving 31 training experiments and 24 test cases (1 undamaged, 18 single-damage, 5 double-damage). The average training classification error was only 0.54%, and heatmap visualizations of PoD distributions confirmed high localization accuracy for single damage cases. For double damage cases, particularly when damages were adjacent or symmetric, the CNN performance showed slight degradation—sometimes falsely indicating damage at an undamaged joint—suggesting room for improvement in more complex damage scenarios. Nevertheless, the method delivered high accuracy and computational efficiency, making it a promising approach for real-time, vibration-based SHM systems.

Answer to question 4: - How well did their method perform? - What were the key metrics or findings? - How did it compare to other methods, if any? > The experimental evaluation was conducted in **two phases**. In the first phase, focusing on a **single girder with 5 joints**, the trained CNNs accurately distinguished between undamaged, single-damage, and double-damage cases, consistently assigning high probability-of-damage (PoD) scores to damaged joints and low values to intact ones. In the second phase, the method scaled up to monitor all **30 joints across the full structure**, involving **31 training experiments** and **24 test cases** (1 undamaged, 18 single-damage, 5 double-damage). The average **training classification error was only 0.54%**, and heatmap visualizations of PoD distributions confirmed high localization accuracy for single damage cases. For double damage cases, particularly when damages were adjacent or symmetric, the CNN performance showed slight degradation—sometimes falsely indicating damage at an undamaged joint—suggesting room for improvement in more complex damage scenarios. Nevertheless, the method delivered **high accuracy and computational efficiency**, making it a promising approach for real-time, vibration-based SHM systems.
nuluh commented 2025-05-14 17:59:43 +00:00 (Migrated from github.com)
Copy Link

Alternative short paragraph:

In their study, [Author(s), Year] introduced a CNN-based structural damage detection approach validated through a large-scale grandstand simulator at Qatar University. The structure, designed to replicate modern stadiums, was equipped with 30 accelerometers and subjected to controlled damage by loosening beam-to-girder bolts. Acceleration data, collected under band-limited white noise excitation and sampled at 1024 Hz, were segmented into 128-sample frames for training localized 1D CNNs—one per joint—creating a decentralized detection system. Across two experimental phases, involving both partial and full-structure monitoring, the method demonstrated high accuracy in damage localization, achieving a training classification error of just 0.54%. While performance remained strong even under double-damage scenarios, some misclassifications occurred in symmetric or adjacent damage cases. Overall, the proposed method presents a highly efficient and accurate solution for real-time SHM applications.

Alternative short paragraph: > In their study, [Author(s), Year] introduced a CNN-based structural damage detection approach validated through a large-scale grandstand simulator at Qatar University. The structure, designed to replicate modern stadiums, was equipped with 30 accelerometers and subjected to controlled damage by loosening beam-to-girder bolts. Acceleration data, collected under band-limited white noise excitation and sampled at 1024 Hz, were segmented into 128-sample frames for training localized 1D CNNs—one per joint—creating a decentralized detection system. Across two experimental phases, involving both partial and full-structure monitoring, the method demonstrated high accuracy in damage localization, achieving a training classification error of just 0.54%. While performance remained strong even under double-damage scenarios, some misclassifications occurred in symmetric or adjacent damage cases. Overall, the proposed method presents a highly efficient and accurate solution for real-time SHM applications.
meowndor commented 2025-05-14 23:56:04 +00:00 (Migrated from github.com)
Copy Link

Added paragraph to add the paper's conclusion and its future works:

In the context of this thesis, the dataset and experimental setup introduced by [Author(s), Year] form the foundation for comparative analysis and algorithm testing. The authors have not only demonstrated the efficacy of a compact 1D CNN-based system for vibration-based structural damage detection, but also highlighted the value of using output-only acceleration data—a constraint shared in this thesis’s methodology. The decentralized design of their system, which allows each CNN to process only locally available data, is particularly aligned with this thesis's focus on efficient, sensor-level data analysis without requiring full-system synchronization. Furthermore, since the authors indicate plans to publicly release their dataset and source code, this thesis leverages that open data for applying alternative analysis methods such as support vector machines (SVM) or frequency domain feature extraction techniques, allowing a direct performance comparison between classical and deep learning-based SHM approaches. Thus, this work serves as both a benchmark reference and a data source in the development and evaluation of more accessible, lower-complexity alternatives for structural health monitoring systems.

Added paragraph to add the paper's conclusion and its future works: > In the context of this thesis, the dataset and experimental setup introduced by [Author(s), Year] form the foundation for comparative analysis and algorithm testing. The authors have not only demonstrated the efficacy of a compact 1D CNN-based system for vibration-based structural damage detection, but also highlighted the value of using output-only acceleration data—a constraint shared in this thesis’s methodology. The decentralized design of their system, which allows each CNN to process only locally available data, is particularly aligned with this thesis's focus on efficient, sensor-level data analysis without requiring full-system synchronization. Furthermore, since the authors indicate plans to publicly release their dataset and source code, this thesis leverages that open data for applying alternative analysis methods such as support vector machines (SVM) or frequency domain feature extraction techniques, allowing a direct performance comparison between classical and deep learning-based SHM approaches. Thus, this work serves as both a benchmark reference and a data source in the development and evaluation of more accessible, lower-complexity alternatives for structural health monitoring systems.
👍 1
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
bug
Something isn't working
 critical
Block Progress
 documentation
Improvements or additions to documentation
 documentclass
Improvements or additions to LaTeX documentclass
 duplicate
This issue or pull request already exists
 enhancement
New feature or request
 good first issue
Good for newcomers
 help wanted
Extra attention is needed
 invalid
This doesn't seem right
 latex
Related to the writings
 question
Further information is requested
 wontfix
This will not be worked on
No Label documentation latex
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
nuluh
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: nuluh/thesis#67
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?