As-Printed Shape Prediction for Faster Additive Manufacturing Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Additive manufacturing (AM) processes often result in significant geometric deviations between nominal designs and fabricated parts, leading to issues like porosity, surface roughness, and performance degradation due to residual stresses and fatigue mechanisms.
Innovation Solution
The method involves using a combination of high-fidelity physics models and neural networks to predict the shape of as-printed parts in additive manufacturing. A high-fidelity physics model simulates the deposition process, while a neural network serves as a surrogate model, providing faster and more efficient predictions of the part shape, thereby accounting for manufacturing uncertainties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-fidelity physics models are used to simulate the deposition process, then prediction accuracy is improved, but computational time increases significantly
Solution Approach 1:
A neural network surrogate model is trained to replicate the behavior of high-fidelity physics models. The neural network learns the mapping from process parameters to as-printed shapes by training on datasets generated from physics-based simulations or experimental data. Once trained, the neural network can predict part shapes in seconds while maintaining accuracy comparable to the high-fidelity models, effectively creating a fast copy of the complex simulation system.
Solution Approach 2:
The neural network is trained in advance using comprehensive datasets that cover the expected range of process parameters and outcomes. This preliminary training phase allows the model to internalize complex relationships between deposition parameters and resulting geometries. During actual AM process planning, the pre-trained model can quickly predict outcomes without requiring real-time high-fidelity simulations, thus resolving the time-accuracy tradeoff.
2Adaptability or versatility
If traditional trial-and-error methods are used in AM process planning, then design flexibility is maintained, but productivity decreases due to multiple iterations
Solution Approach 1:
The system implements a feedback loop where the neural network predicts the as-printed shape based on proposed process parameters, compares it with the target geometry, and allows iterative optimization of parameters before actual manufacturing. This virtual feedback mechanism enables designers to explore multiple design variations and parameter settings quickly, maintaining design flexibility while eliminating the need for physical trial-and-error iterations, thus significantly improving productivity.
3Device complexity
If geometric deviations are not accounted for in AM processes, then manufacturing complexity is reduced, but manufacturing precision deteriorates due to porosity, surface roughness, and residual stresses
Solution Approach 1:
The neural network model is used in the planning stage to predict geometric deviations and as-printed shapes before manufacturing begins. By anticipating issues like porosity, surface roughness, and dimensional inaccuracies in the virtual domain, engineers can adjust process parameters, support structures, and toolpaths proactively to compensate for expected deviations. This preliminary prediction and compensation approach maintains simple manufacturing processes while achieving high manufacturing precision.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
A computer representation of a printable product part and a plan for the printable product part to be deposited using an additive manufacturing process are received (110). The printable product part comprises an accumulation of material deposited by the additive manufacturing process. The plan comprises a tool-path representation of the printable product part and process parameters. A plurality of as-printed shapes of the printable product part are determined after it has been deposited according to the plan (120). Geometric differences between any of the plurality of as-printed shapes with the computer representation of the product part are determined (130).