Damage clasification in composite materials using neural networks

Abstract

Composite materials are widely employed in critical industrial applications, where their use has surged due to their numerous advantages over traditional materials. However, these benefits can be compromised if adequate quality control techniques are not implemented, particularly for detecting structural damage. Acoustic emission is a nondestructive technique commonly used for damage detection. By leveraging artificial intelligence tools to efficiently process emitted signals, the detection and classification process can be automated. This study utilizes sound pressure levels to diagnose failures in fiberglass-reinforced (GFRP) epoxy composite beams. A pattern recognition system based on Artificial Neural Network (ANN) algorithms is employed for diagnosis. To ensure data variability, the classifier was trained and validated using preprocessed acoustic signals from multiple healthy and damaged beams in various locations. Testing was conducted using test results from specimens not used for training and validation, ensuring the ANN's robustness. The results demonstrate a high fault detection percentage, confirming the reliability of the ANN.

Composite materials are widely employed in critical industrial applications, where their use has surged due to their numerous advantages over traditional materials. However, these benefits can be compromised if adequate quality control techniques are not implemented, particularly for detecting structural damage. Acoustic emission is a nondestructive technique commonly used for damage detection. By leveraging artificial intelligence tools to efficiently process emitted signals, the detection and classification process can be automated. This study utilizes sound pressure levels to diagnose failures in fiberglass-reinforced (GFRP) epoxy composite beams. A pattern recognition system based on Artificial Neural Network (ANN) algorithms is employed for diagnosis. To ensure data variability, the classifier was trained and validated using preprocessed acoustic signals from multiple healthy and damaged beams in various locations. Testing was conducted using test results from specimens not used for training and validation, ensuring the ANN's robustness. The results demonstrate a high fault detection percentage, confirming the reliability of the ANN.