Neural Thermal Mapping for Voxel-Level Additive Manufacturing Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current additive manufacturing techniques face challenges in predicting transient thermal behavior at voxel-level resolution due to limited resolution of built-in thermal sensors and inadequate upscaling methods, which hinder online print tuning and control.
Innovation Solution
A deep neural network-based approach that utilizes low-resolution thermal sensing and contone maps to predict thermal images at voxel-level resolution, achieving enhancement by approximately 20 times the original resolution, enabling real-time in-situ voxel-level thermal image prediction and feedback control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If built-in thermal sensors are used for thermal imaging, then thermal sensing is integrated into the additive manufacturing device, but the resolution is limited and insufficient for voxel-level thermal prediction
Solution Approach 1:
A deconvolution neural network is introduced as an intermediary computational model that takes low-resolution thermal images from built-in sensors and contone maps as inputs, then processes them to generate high-resolution thermal predictions at voxel-level detail, effectively mediating between the limited sensor capability and the required prediction precision
Solution Approach 2:
The system creates a high-resolution thermal image copy or prediction based on the low-resolution actual thermal measurement, using the neural network to synthesize detailed thermal information that exceeds the original sensor resolution, enabling voxel-level thermal analysis without requiring high-resolution physical sensors
2Measurement precision
If conventional upscaling methods are used to increase thermal image resolution, then resolution enhancement is achieved, but the accuracy is insufficient for accurate thermal prediction
Solution Approach 1:
The deconvolution neural network serves as an intelligent intermediary that goes beyond simple interpolation by learning the complex relationship between low-resolution thermal measurements and high-resolution thermal distributions, incorporating contone map information to guide the upscaling process and ensure physically accurate thermal predictions
Solution Approach 2:
The system transforms the thermal image from low-resolution to high-resolution by changing the resolution parameter through neural network processing, while simultaneously changing the information content by integrating contone map data to ensure the upscaled image maintains accurate thermal characteristics rather than just increasing pixel count
3Measurement precision
If high-resolution thermal sensors are used, then voxel-level thermal imaging is achieved, but the device complexity and cost increase
Solution Approach 1:
Instead of physically implementing high-resolution thermal sensors, the system creates a computational copy of high-resolution thermal imaging capability through the deconvolution neural network, which processes outputs from simple built-in sensors to generate detailed thermal images, avoiding the complexity and cost of actual high-resolution sensor hardware
Solution Approach 2:
The patent replaces the mechanical/optical approach of using physically complex high-resolution thermal sensors with a computational/software-based approach using neural networks, substituting hardware complexity with intelligent algorithms that achieve the same functional outcome of high-resolution thermal imaging
Data Source
AI summary
Examples of methods for thermal mapping by an electronic device are described herein. In some examples, a map is obtained. In some examples, a first thermal image at a first resolution is obtained. In some examples, a neural network is used to determine a second thermal image at a second resolution based on the map and the first thermal image. The second resolution is greater than the first resolution in some examples.