研究目的
The assessment of structural damage for ensuring the service safety of carbon fiber reinforced plastics (CFRP) structures by predicting damage degree using fiber Bragg grating and epsilon-support vector regression.
研究成果
The ε-SVR model accurately predicts damage degree in CFRP structures with an absolute relative error less than 10% for most testing samples. The method leverages FBG sensors to minimize mass impact and uses regression for small sample prediction, showing promise for structural health monitoring. Future work could explore real damage types and larger datasets.
研究不足
The study uses simulated damage by mass loading, which may not fully represent real-world damage scenarios. The method requires active excitation and may be sensitive to sensor placement and environmental factors. The small sample size (60 training, 30 testing) could limit generalizability, and the feature selection threshold (0.05 to 0.15) is empirically chosen without extensive validation.
1：Experimental Design and Method Selection:
The study involves detecting structural dynamic response signals using fiber Bragg grating (FBG) sensors, applying Fourier transform to extract frequency response characteristics as damage features, reducing feature dimensionality with the RReliefF algorithm, and establishing a prediction model using epsilon-support vector regression (ε-SVR).
2：Sample Selection and Data Sources:
A CFRP plate of size 500 mm × 500 mm × 2 mm with fixed edges is used. Damage is simulated by loading concentrated mass blocks (0 g, 34.5 g, 69 g, 103.5 g, 138 g, 170 g, 202 g) in the damage area. Training samples include 60 groups (15 per damage degree for 0 g, 69 g, 138 g, 202 g), and testing samples include 30 groups (5 per damage degree for 34.5 g, 69 g, 103.5 g, 138 g, 170 g, 202 g).
3：5 g, 69 g, 5 g, 138 g, 170 g, 202 g) in the damage area. Training samples include 60 groups (15 per damage degree for 0 g, 69 g, 138 g, 202 g), and testing samples include 30 groups (5 per damage degree for 5 g, 69 g, 5 g, 138 g, 170 g, 202 g).
List of Experimental Equipment and Materials:
3. List of Experimental Equipment and Materials: Equipment includes FBG sensors (wavelengths: 1539.999 nm, 1536.360 nm, 1565.235 nm, 1531.866 nm), wavelength demodulation equipment (SM130 from MOI), data acquisition and processing device, CFRP plate, steel ball for active excitation (25 g, 0.06 J energy), and concentrated mass blocks.
4：999 nm, 360 nm, 235 nm, 866 nm), wavelength demodulation equipment (SM130 from MOI), data acquisition and processing device, CFRP plate, steel ball for active excitation (25 g, 06 J energy), and concentrated mass blocks.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: Active excitation is applied using a steel ball impact. Dynamic response signals are detected by FBG sensors, processed with Fourier transform to get frequency response, features are reduced with RReliefF, and ε-SVR model is trained and validated using cross-validation for parameter optimization.
5：Data Analysis Methods:
Feature weights are calculated using RReliefF, model parameters (regularization and kernel) are optimized via grid search and k-fold cross-validation, and prediction accuracy is evaluated using mean squared error and relative error.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
