Chen, Qian
(2022)
Multiscale Process Modeling of Residual Deformation and Defect Formation for Laser Powder Bed Fusion Additive Manufacturing.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
Laser powder bed fusion (L-PBF) additive manufacturing (AM) is capable of producing complex parts near net shape with good mechanical properties. However, undesired residual stress and distortion that lead to build failure and defects such as porosity are preventing broader applications of L-PBF. To realize the full potential of L-PBF, a multiscale modeling methodology is developed to predict residual deformation, melt pool, and porosity formation.
To predict the residual deformation and stress in L-PBF at part-scale, a multiscale process modeling framework based on inherent strain method is proposed. Inherent strain vectors are extracted from detailed multi-layer process simulation with high fidelity at micro-scale. Uniform but anisotropic strains are then applied to L-PBF part in a layer-by-layer fashion in a quasi-static equilibrium finite element analysis (FEA) to predict residual distortion and stress. Besides residual distortion and stress prediction at part scale, multiphysics modeling at powder scale is performed to study the melt pool variation and defect formation induced by process parameters, preheating temperature and spattering particles. Melt pool dynamics and porosity formation mechanisms associated with these factors are revealed through simulation and experiments.
Based on the proposed part-scale residual stress and distortion model, path planning method is developed to tailor the laser scanning path for a given geometry to prevent large residual deformation and building failures. Gradient based path planning for continuous and island scanning strategy is formulated and full sensitivity analysis for the formulated compliance- and stress-minimization problem is performed. The feasibility and effectiveness of this proposed path planning method is demonstrated experimentally using the AconityONE L-PBF system.
In addition, a data-driven framework utilizing machine learning is developed to predict the thermal history at part-scale for L-PBF. In this work, a sequential machine learning model including convolutional neural network (CNN) and recurrent neural network (RNN), long short-term memory unit, is proposed for real-time thermal history prediction. A 100x prediction speed improvement is achieved compared to the finite element analysis which makes the prediction faster than real fabrication process and real-time temperature profile available.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
16 January 2022 |
| Date Type: |
Publication |
| Defense Date: |
29 October 2021 |
| Approval Date: |
16 January 2022 |
| Submission Date: |
8 November 2021 |
| Access Restriction: |
No restriction; Release the ETD for access worldwide immediately. |
| Number of Pages: |
185 |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
Swanson School of Engineering > Mechanical Engineering |
| Degree: |
PhD - Doctor of Philosophy |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
Additive manufacturing; Computational mechanics |
| Date Deposited: |
16 Jan 2022 18:06 |
| Last Modified: |
16 Jan 2022 18:06 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/41922 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}