Prescriptive Data-Analytical Modeling of Laser Powder Bed Fusion Processes for Accuracy Improvement

Abstract

Laser powder bed fusion (LPBF) has the ability to produce three-dimensional lightweight metal parts with complex shapes. Extensive investigations have been conducted to tackle build accuracy problems caused by shape complexity. For metal parts with stringent requirements, surface roughness, laser beam positioning error, and part location effects can all affect the shape accuracy of LPBF built products. This study develops a data-driven predictive approach as a promising solution for geometric accuracy improvement in LPBF processes. To address the shape complexity issue, a prescriptive modeling approach is adopted to minimize geometrical deviations of built products through compensating computer aided design models, as opposed to changing process parameters. It allows us to predict and control a wide range of shapes starting from a limited set of measurements on basic benchmark geometries. An error decomposition and compensation scheme is developed to decouple the influence from different error components and to reduce the shape deviations caused by part geometrical deviation, laser beam positioning error, and other location effects simultaneously via an integrated modeling and compensation framework. Experimentation and data collection are conducted to investigate error sources and to validate the developed modeling and accuracy control methods.

Efficient Distortion Prediction of Additively Manufactured Parts Using Bayesian Model Transfer Between Material Systems

“…An example of a formulation that could be used for generalizing our model to various types of geometries can be found in Refs. [50,52]. The latter article suggests that the distortion models learned in our case study can serve to capture "global" distortion features that can be used to facilitate distortion modeling of more complicated shapes.…”

Quality Monitoring and Control in Additive Manufacturing

“…As in all thermal processes, the solidification and cooling of the 3D printed volume bring volume contractions, which can cause thermal stresses that can be hardly predicted when complex geometries are produced. This is why some authors focused on quality prediction, i.e., predict the final geometry of AM products, starting from a limited number of test cases [42]- [45] , using a transfer learning paradigm [46]- [48] or combining simulation and real data in a multi-stage calibration framework [49] .…”

Pyramid Ensemble Convolutional Neural Network for Virtual Computed Tomography Image Prediction in a Selective Laser Melting Process

“…For example, Sun et al proposed a functional quantitative and qualitative model to predict two types of quality responses (i.e., number of voids and the surface roughness) via offline setting variables and in situ process variables [23]. Huang et al developed a series of models to predict product deviations based on engineering knowledge and experiments [24][25][26]. According to the predicted deviation to a specific computer aided design, the optimal compensation plan can be implemented to improve the product geometry accuracy in AM.…”