Low Velocity Impact Monitoring of Composite Tubes Based on FBG Sensors

Shengsheng Huan, Linjiao Lu, Tao Shen, Jianke Du2024 · DOI

TL;DR

This paper presents a novel approach for pinpointing low-velocity impacts on carbon fiber composite tubes using Fiber Bragg Grating (FBG) sensors. By analyzing the subtle shifts in FBG sensor wavelengths and employing an error outlier-based algorithm, the researchers achieved significantly higher accuracy in impact localization compared to traditional neural network methods, requiring less training data.

What Problem Does This Paper Address?

Carbon fiber reinforced composites (CFRPS) are increasingly used in critical applications like aerospace and automotive due to their strength-to-weight ratio. However, they are susceptible to hidden damage from low-velocity impacts, which can compromise structural integrity without visible signs. Detecting and precisely locating these impact events is crucial for ensuring safety and preventing catastrophic failures. Existing methods for impact localization in composite structures often rely on strain gauges or accelerometers, but these can be bulky or susceptible to electromagnetic interference. While Fiber Bragg Grating (FBG) sensors offer advantages like small size and immunity to EMI, their application for precise impact localization on tubular composite structures, especially with low-velocity impacts, requires further investigation and refined methodologies.

Methods

The researchers employed a multi-step approach for impact monitoring:

  • FBG Sensor Integration: Four FBG sensors were strategically bonded to the surface of a carbon fiber composite tube. These sensors detect strain-induced changes in their reflected wavelength when an impact occurs.
  • Signal Acquisition and Denoising: The wavelength shifts from the FBG sensors were captured using an optical sensing interrogator. To clean up noisy signals, the Discrete Wavelet Transform (DWT) was utilized. Specifically, the db3 wavelet was chosen for its ability to preserve signal peaks and abrupt changes, and a hard thresholding function was applied to reconstruct the denoised signal.
  • Impact Localization Algorithms: Two primary methods were explored for pinpointing impact locations:
  • BP Neural Network: A standard Backpropagation (BP) neural network was trained using the maximum center wavelength shift as input and the impact coordinates (axial and radial) as output.
  • Error Outlier Algorithm: This novel approach involves calculating the error between a reference FBG signal and a test signal. Outliers in this error are identified by comparing them to a predefined threshold. The location with the fewest outliers is then identified as the impact site. This method focuses on the differences between signals from different impact locations versus those from the same location.

Datasets

The study utilized experimental data collected from low-velocity impact tests on a carbon fiber composite tube.

  • Training Data: For the BP neural network, three datasets, each containing 21 data points (representing 21 impact test points), were used for training, totaling 63 data points.
  • Testing Data: Four random cases were selected for testing both the BP neural network and the error outlier algorithm.
  • Reference Data: The error outlier algorithm specifically utilized a single data set as a reference signal for comparison.

The paper does not explicitly state the creation of a new benchmark dataset but rather uses data generated from their experimental setup.

Key Insights and Ideas

  • FBG Sensors' Sensitivity: FBG sensors are highly sensitive to the minute deformations caused by even low-velocity impacts, making them suitable for detecting these events.
  • DWT for Signal Integrity: The Discrete Wavelet Transform is effective in removing noise from FBG sensor signals without distorting crucial impact-related features.
  • Error Outlier Superiority: The error outlier-based localization method demonstrated a significant advantage over the BP neural network. This is attributed to its ability to effectively differentiate between signals from distinct impact locations by focusing on outlier detection, rather than relying on complex pattern mapping.
  • Data Efficiency: The error outlier method requires substantially less training data compared to the BP neural network, making it more practical for real-world applications where data acquisition can be time-consuming and resource-intensive.

Key Results

  • BP Neural Network: Achieved an average localization error of 20.68 mm with a maximum error of 22.82 mm.
  • Error Outlier Algorithm: Achieved a superior average localization error of 13.98 mm with a maximum error of 14.70 mm.
  • Comparison: The error outlier method showed higher predictive accuracy and a more stable performance, particularly in minimizing variance, when compared to the BP neural network.

Limitations and Future Work

  • Limited Impact Locations: The study considered a specific set of 21 impact test points. While the error outlier method showed promise, its performance across a wider and more continuous range of impact locations could be further investigated.
  • Single Tube Type: The experiments were conducted on a single type of carbon fiber composite tube. Future work could explore the applicability of these methods to tubes with different material compositions, layup configurations, or geometries.
  • Low-Velocity Focus: The research specifically addresses low-velocity impacts. Extending the methodology to monitor higher velocity impacts or different types of damage would be a valuable next step.
  • Further Optimization: While the error outlier method is effective, further refinement of the threshold selection and outlier identification criteria could potentially lead to even greater accuracy. The authors suggest that the latter approach (error outliers) has higher predictive accuracy and does not require a significant amount of data, implying its potential for broader adoption.

Summary updated

BibTeX
@misc{huan2024low,
  author = {Shengsheng Huan and Linjiao Lu and Tao Shen and Jianke Du},
  title = {{Low Velocity Impact Monitoring of Composite Tubes Based on FBG Sensors}},
  year = {2024},
  doi = {10.3390/s24041279},
}
Figure