Computationally efficient distortion prediction in Powder Bed Fusion Additive Manufacturing
Abstract
This work addresses the analysis and development of computationally efficient distortion prediction numerical methodologies applicable to powder bed based selective laser melting (SLM) process. Initially, state of the art of simplified distortion modelling methodologies based on finite element (FE) models is introduced. Existing methodologies are described in terms of complexity and applicability to SLM process.
The methodology known as inherent shrinkage, previously developed for multipass welding processes, is applied to predict SLM process induced distortion in Inco 718 testing geometry (cantilever). An assessment about predictive capability of this simplified model based on correlation between numerical results and experimental measurements is performed. Experimental distortions are measured after cutting of base plate connected supports. Initially, the influence of meshing, layer activation and equivalent thermal loads is investigated in terms of prediction capability and computational cost. Subsequently, isotropic and non-isotropic thermal expansion coefficients (α) are considered in the FE-model definition. Results demonstrate that it is feasible to accurately predict the distortion induced by different scanning strategies (chessboard pattern, transversal stripes and longitudinal stripes) in short times. Current developments entail a cost-effective alternative for controlling and reducing distortions in SLM parts.
Keywords
Download Options
Introduction
In powder bed based additive manufacturing (AM) processes components are directly built up layer by layer. Both laser (SLM, DMLS, SLS…) and electron beam (EBM) heat sources are employed to locally melt thin layers of predeposited metallic powders which will shape the final part. Distortions are critical during AM of metallic parts, since they increase manufacturing costs, times and generate wastes and scraps. Currently, corrective measures are often not taken until production step and they are mainly based on trial an error approach. This experimental corrective approach lacks of efficiency, flexibility and cost-efficiency. Alternatively, preventive methods based on distortion prediction numerical methodologies can anticipate distortions even form the design stage and entail a powerful tool for part design and process optimization. However, this requires the development of new computationally efficient modelling strategies that can be applicable to real parts
In Selective Laser Melting (SLM), the material is locally and rapidly heated above its melting temperature and then allowed to solidify and cool to form a dense geometry. Extremely high heating and cooling rates are obtained due to the highly concentrated nature of heat source, i.e., laser beam, and quick processing rates (typical scanning rates around 1 m/s). This involves large thermal gradients. Thereby, residual stresses and distortions are generated due to the local inherent nature of the process. Moreover, post-manufacturing processes, such as cutting of support structures, lead to redistribution of internal stresses and strain modifying as-build distortions.
Conclusion
Inherent shrinkage simplified numerical methodology has been successfully applied to powder bed fusion SLM process. This new computationally efficient methodology is suitable for the fast and accurate prediction of real components and has been validated in real simple geometry (cantilever) by comparing model results with experimentally measured distortion.
This methodology requires the definition of a FE-model with a sequential activation of layers (“birth and death” or model change technique). A minimum number of layer activation steps is required in order to ensure good prediction. For the geometry studied in this work a good prediction was achieved for a ratio between real manufacturing layers and model layer activation steps of 8 (400 real layers, activated in 50 steps). Above this ratio, the predictive capability decreases, whereas the increase of this ratio does not improve the accuracy of the results, but increases computational cost.
Experimental distortion mainly depend on scanning strategy (chess-board, transversal stripes, longitudinal stripes). Despite the simplicity of the model, it was feasible to predict the distortion using different scanning strategies. Good correlation with chess-board scanning strategy was obtained by applying isotropic model (no differences in the 3D components of the thermal expansion coefficients), whereas, close matching with both longitudinal and transversal stripe scanning strategies was attained by orthotropic models (higher thermal expansion coefficient in longitudinal or transversal direction).
Current results support the idea that inherent shrinkage methodology can be a computationally efficient tool for fast distortion prediction in powder bed fusion AM processes. A next step towards the implementation of this methodology should be to transfer to more complex parts.