Quality Prediction for Wire Arc Additive Manufacturing Based on Multi-source Signals, Whale Optimization Algorithm–Variational Modal Decomposition, and One-Dimensional Convolutional Neural Network

• Published in: Journal of materials engineering and performance Vol. 33; no. 20; pp. 11351 - 11364
• Main Authors: Huang, Yong, Yue, Chenkai, Tan, Xiaxin, Zhou, Ziyuan, Li, Xiaopeng, Zhang, Xiaoyong, Zhou, Chundong, Peng, Yong, Wang, Kehong
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.10.2024; Springer Nature B.V
• Subjects: Additive manufacturing; Algorithms; Artificial neural networks; Characterization and Evaluation of Materials; Chemistry and Materials Science; Corrosion and Coatings; Decomposition; Defects; Electric potential; Engineering Design; Entropy; Lagrange multiplier; Manufacturing; Materials Science; Multisensor applications; Neural networks; Optimization; Optimization algorithms; Original Research Article; Prediction models; Quality Control; Reliability; Safety and Risk; Signal processing; Slagging; Time series; Tribology; Unevenness; Voltage; Wavelet transforms; arc additive manufacturing; energy entropy; variational modal decomposition; multi-sensor; convolutional neural network
• ISSN: 1059-9495; 1544-1024
• DOI: 10.1007/s11665-023-08768-7

A deep-neural-network-based multi-sensor data and defect prediction algorithm for gas metal arc additive manufacturing (GMA-AM) is proposed. The core idea is to collect current, voltage and sound signals during GMA-AM using multiple sensors and to combine with a one-dimensional convolutional neural network (1D-CNN) model to identify different defect states. First, the current, voltage and sound signals are adaptively decomposed by optimizing the variational modal decomposition (VMD) parameters through the whale optimization algorithm (WOA), and the energy entropy of the decomposition components is used as the feature vector to detect different defect states. Then, a defect prediction model is established using 1D-CNN to classify and discriminate five types of typical defects: trajectory deviation, pore, slagging, thermal deformation, and surface unevenness. Experimental verification shows WOA-VMD can adaptively decompose the signals of arc additive manufacturing and detect different defect states with 92.22% accuracy.