This paper proposed a method for predicting composite laminates’ compressive residual strength after impact based on convolutional neural networks. Laminates made by M21E/IMA prepreg were used to introduce low-velocity impact damage and construct a non-destructive testing image dataset. The dataset images characterized the impact damage details, including dents, delamination, and matrix cracking. The convolution kernel automatically extracted and identified these complex features that could be used for classification. The model took the images as input and compressive residual strength labels as output for iterative training, and the final prediction accuracy reached more than 90%, the highest 96%. This method introduced overall damage into the model in the form of images utilizing convolution, which can quickly and accurately predicted laminates’ compression performance after impact.

提出了一种基于卷积神经网络的复合材料层板冲击后抗压残余强度预测方法。M21E / IMA预浸料制成的层压板用于引入低速冲击损伤并构建非破坏性测试图像数据集。数据集图像描述了冲击损伤的详细信息，包括凹痕，分层和基体开裂。卷积核自动提取并识别了可用于分类的这些复杂特征。该模型以图像作为输入，压缩残余强度标签作为输出进行迭代训练，最终预测精度达到90％以上，最高为96％。这种方法利用卷积以图像的形式将整体损坏引入模型中，