From in-situ monitoring toward high-throughput process control: cost-driven decision-making framework for laser-based additive manufacturing

• Published in: Journal of manufacturing systems Vol. 51; pp. 29 - 41
• Main Authors: Jafari-Marandi, Ruholla, Khanzadeh, Mojtaba, Tian, Wenmeng, Smith, Brian, Bian, Linkan
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.04.2019
• Subjects: Additive manufacturing (AM); Artificial neural networks (ANN); Porosity prediction; Self-organizing error-drive neural networks (SOEDNN); Thermal history; Additive manufacturing (AM); Artificial neural networks (ANN); Thermal history; Self-organizing error-drive neural networks (SOEDNN); Porosity prediction
• ISSN: 0278-6125; 1878-6642
• DOI: 10.1016/j.jmsy.2019.02.005

•The paper presents an ANN-based cost-driven classification method for intelligent porosity detection.•The method accounts for spatial distribution of defects and recommends cost-wise smart corrective actions.•The method excels at porosity detection in regard to both accuracy and cost when compared with the recent state of the arts.•Significant cost savings are shown by using the presented data- and cost-driven decision making framework. Additive manufactured (AM) parts are subject to low repeatability compared to their traditional counterparts. The complex Process-Structure-Property relationship that governs the laser-based AM processes calls for advanced analytics approaches for quality control. Tremendous efforts have been dedicated to the in-situ monitoring of AM processes by leveraging the thermal history during fabrication. Melt pool image is regarded as one of the most informative process signatures for real-time porosity detection. Attempting to control/correct all microstructure defects during the AM fabrication may significantly reduce the process throughput, which has been a bottleneck of the AM technology for its wider industrial adoption. Without distinguishing different types of defects from one another, and without formally characterizing the cost/impact of microstructural defects on part property, an efficient in-process control strategy cannot be obtained. In this paper, a cost-driven decision-making framework is proposed to formulate costs of the spatial distribution of microstructural defects and the corresponding control actions, based on in-situ melt pool images. A case study based on thin wall fabrication using a Laser Engineered Net Shaping process is used to illustrate the effectiveness in both classification accuracy and misclassification cost. This work is expected to lay a theoretical foundation for the development of an efficient in-process control strategy, which aims to improve the mechanical properties of fabricated part while maintaining high process throughput.