ML Thermal Imaging for Print-Resolution Additive Manufacturing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing additive manufacturing techniques face challenges in achieving thermal resolution at or near print resolution, leading to difficulties in predicting and managing thermal behavior during the manufacturing process, which affects the quality of 3D objects.
Innovation Solution
Utilizing machine learning models, particularly deep learning and neural networks, to enhance the resolution of thermal information by combining simulation results with geometrical data, allowing for real-time thermal management and agent distribution at voxel-level precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If simulation results are used for thermal prediction, then thermal information is obtained, but the resolution is insufficient to achieve print resolution
Solution Approach 1:
The patent creates a digital twin (virtual model) of the additive manufacturing process that replicates thermal behavior. This virtual copy allows high-resolution thermal prediction without requiring expensive physical measurements, resolving the contradiction between obtaining thermal information and achieving sufficient resolution.
Solution Approach 2:
The patent replaces physical thermal measurement systems with computational modeling and machine learning algorithms. This substitution enables high-resolution thermal prediction through software rather than hardware limitations, achieving print-resolution thermal data without the cost and complexity of high-resolution thermal sensors.
2Measurement precision
If high-resolution thermal simulation is performed, then thermal prediction accuracy improves, but computational cost increases significantly
Solution Approach 1:
The patent performs comprehensive thermal simulations and data collection during the offline training phase before actual manufacturing. This preliminary action creates pre-trained machine learning models that can then rapidly predict thermal behavior during production without requiring heavy computational resources in real-time, resolving the contradiction between accuracy and computational cost.
Solution Approach 2:
The patent creates a simplified computational model (digital twin) that captures essential thermal physics without the full computational complexity of detailed simulations. This copy maintains predictive accuracy while dramatically reducing computational overhead during actual manufacturing operations.
3Manufacturing precision
If thermal management is optimized at voxel-level precision, then manufacturing quality improves, but system complexity increases
Solution Approach 1:
The patent uses a virtual digital twin to model and optimize thermal processes at voxel-level precision without requiring complex physical measurement and control systems. The computational model handles the complexity of voxel-level thermal management, allowing high manufacturing quality while keeping the physical system relatively simple.
Solution Approach 2:
The patent replaces complex physical thermal measurement and control systems with machine learning-based predictive models. This substitution enables voxel-level thermal optimization through software intelligence rather than hardware complexity, maintaining high manufacturing quality while reducing system complexity.
Data Source
AI summary
Examples of methods for thermal image determination are described. In some examples, a method may include determining, using a first machine learning model, a first thermal image of a first layer of additive manufacturing. In some examples, the method may include determining, using a second machine learning model, a second thermal image of a second layer based on a simulated thermal image. The second thermal image may have a second resolution that is greater than a first resolution of the simulated thermal image in some examples. In some examples, the method may include determining, using a third machine learning model, a third thermal image based on the first thermal image and the second thermal image.