Qualify assessment for extrusion-based additive manufacturing with 3D scan and machine learning
Xiaoyu Li, Mengna Zhang, Mingxia Zhou, Jing Wang, Weixin Zhu, Chuan Wu, Xiao Zhang
Abstract
Additive manufacturing (AM) technology has been widely used in the aerospace, automotive, and healthcare industries over the last few decades. However, the dependability of AM process in producing high-quality manufactured goods remains an open and challenging task. To address this challenge, this study presents a data-driven predictive model based on multiple machine learning algorithms for Fused Deposition Modeling (FDM) process. Temperature and vibration data from multiple sensors (thermocouples, infrared thermometers, and accelerometers) are collected and divided to extract the time-frequency feature accordingly. Then, the sensing data and process parameters are fused to predict dimensional deviations between the printed model and the original one. To quantify the importance of different process parameters and sensing data, a sensitivity analysis experiment is performed on the training data. Experimental evaluation has shown that the prediction model based on the Residual Attention neural network outperforms other machine learning models, such as 1D Convolutional Neural Network and Convolutional Neural & Long Short-Term Memory Network. The proposed robust model can improve the quality assurance of additive manufacturing in practice after optimization.