关键词: 电弧熔丝增材制造, 混合模型, 驼峰缺陷, 深度学习, 人工特征

Abstract: Existing online monitoring algorithms for wire arc additive manufacturing lose a significant amount of useful information during feature extraction,which limits the model′s recognition performance. To address this issue,a hybrid model combining physics-driven hand crafted features and data-driven deep learning features is proposed to alleviate information loss during feature extraction,thereby achieving more accurate defect recognition. First,17-dimensional hand crafted features and 128-dimensional deep learning features are extracted from the input electrical signals through hand-crafted feature extraction and a Convolutional Neural Network (CNN),respectively. Then,the dimension of the hand crafted features is adjusted to 128 dimensions through a fully connected layer. Finally,the two types of features are concatenated and used as input to a classifier (a fully connected layer) to achieve accurate recognition of different deposition qualities (hump and good quality). Comparative experimental results show that the method based on deep learning features (CNN) (accuracy of 93.79 %) is superior to the method based on traditional hand-crafted features (accuracy of 91.26 %),and the method based on the hybrid model (accuracy of 96.41 %) is further superior to the method based on deep learning features. This indicates that hand-crafted features and deep learning features are complementary,and their effective combination can fully mine useful information in the data and obtain better recognition performance.

Key words: wire arc additive manufacturing, hybrid model, hump defect, deep learning, hand crafted features