CNN-BiLSTM enabled prediction on molten pool width for thin-walled part fabrication using Laser Directed Energy Deposition. (June 2022)

• Record Type: Journal Article
• Authors: Hu, Kaixiong; Wang, Yanghui; Li, Weidong; Wang, Lihui
• Is Part Of: Journal of manufacturing processes. Volume 78(2022)
• Page Start: 32
• Page End: 45
• Publication Date: 2022-06
• Subjects: Laser Directed Energy Deposition (LDED) -- Molten pool width -- Data driven approach -- Additive manufacturing (AM); Production management -- Data processing -- Periodicals; Manufacturing processes -- Periodicals; Procestechnologie; Productietechniek; Production -- Gestion -- Informatique -- Périodiques; Fabrication -- Périodiques; Manufacturing processes; Production management -- Data processing; Periodicals; 670.5
• DOI: 10.1016/j.jmapro.2022.04.010
• Languages: English
• ISSNs: 1526-6125

Abstract

Abstract: Laser Directed Energy Deposition (LDED) is a promising metal Additive Manufacturing (AM) technology capable of fabricating thin-walled parts to support some high-value applications. Accurate and efficient prediction on the molten pool width is critical to support in-situ control of LDED for part quality assurance. Nevertheless, owing to the intricate physical mechanisms of the process, it is challenging to designing an effective approach to accomplish the prediction target. To tackle the issue, in this research, a new data model-driven predictive approach, which is enabled by a hybrid machine learning model namely CNN-BiLSTM, is presented. High prediction accuracy and efficiency are achievable through innovative measures in the research, that is, (i) the CNN-BiLSTM model is designed and configured by addressing the characteristics of the LDED process; (ii) process parameters related to the deposition and heat accumulation phenomena during the LDED process are extensively considered to strengthen the prediction accuracy. Experiments for thin-walled part fabrication were conducted to validate and benchmark the approach. In average, 4.286% of the mean absolute percentage error (MAPE) was acquired, and the prediction time took by the approach was only 0.04% of that by a finite element analysis (FEA) approach. Compared to the LSTM model, the BiLSTM model and the CNN-LSTM model, MAPEs of the CNN-BiLSTM model were improved by 27.0%, 17.3% and 12.6%, respectively. It demonstrates that the approach is competent in producing good-quality thin-walled parts using the LDED process. Highlights: A hybrid machine learning-driven predictive approach for the LDED process is presented. Process parameters critical to the LDED process are extensively considered to strengthen the prediction accuracy. Experiments for thin-walled part fabrication were conducted to validate and benchmark the approach.