声发射
分层(地质)
失效模式及影响分析
材料科学
小波
稳健性(进化)
小波变换
复合数
结构工程
计算机科学
卷积神经网络
复合材料
压缩(物理)
抗压强度
开裂
灾难性故障
模式识别(心理学)
热塑性塑料
人工智能
作者
Mohammad Mehdi Shahzamanian,Li Ai,Paul H. Ziehl
标识
DOI:10.1177/14759217261443692
摘要
Composite materials in aircraft structures can suffer impact damage that leaves barely visible yet structurally significant defects, which degrade mechanical performance, especially under compressive loads. Existing methods for identifying failure modes in compression after impact (CAI) tests using acoustic emission (AE) data are limited in accuracy and scope. This study introduces an approach combining AE sensing with a heterogeneous ensemble convolutional neural network (CNN) to detect and classify failure mechanisms in impacted composite specimens. The novelty of this work lies in employing Red, Green, and Blue (RGB) wavelet images, produced through continuous wavelet transform (CWT) of AE signals, as inputs to a heterogeneous ensemble of CNN architectures for the impacted composite specimens being used in urban and advanced air mobility vehicles. This approach enables more accurate classification of failure modes than conventional feature-based machine learning (ML) methods such as XGBoost and random forest. By leveraging CWT, failure mode prediction accounts for mixed-mode signals rather than relying solely on peak-frequency ranges, particularly where matrix cracking occurs alongside delamination and matrix–fiber debonding. The CNN model further evaluates the relatedness of different failure mode clusters by analyzing the wavelet shapes corresponding to each mechanism. To address the scarcity of experimental AE data, data augmentation techniques were applied to enlarge the dataset artificially, enhancing model robustness and generalization. CAI tests were performed on thermoplastic composite panels impacted at varying energy levels, emphasizing the critical 30 Joule case where damage is barely visible yet significantly compromises compressive strength. AE signals recorded during testing validated the proposed method. Results demonstrate that the ensemble CNN, aided by data augmentation, outperforms traditional ML models in identifying failure mechanisms. This approach offers a promising path toward real-time structural health monitoring of composites, improving safety and informing more effective aerospace design strategies.
科研通智能强力驱动
Strongly Powered by AbleSci AI