Predicting Thermal Behavior in 3D Printers Using Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The end part quality in 3D printing, particularly in multi jet fusion 3D printers, is directly related to the voxel level thermal behavior in the build bed, which existing technologies fail to accurately predict and control, leading to suboptimal printing results.
Innovation Solution
A deep neural network (DNN) is trained to predict thermal behavior in 3D printers using automatically generated datasets from machine instructions and sensed thermal data, allowing for the generation of thermal distribution maps before printing to improve part quality by modifying the printing process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If existing thermal prediction technologies are used in 3D printers, then the printing process can proceed, but the thermal behavior prediction accuracy is insufficient leading to suboptimal part quality
Solution Approach 1:
The patent replaces traditional thermal modeling approaches with a neural network-based predictive system. The neural network learns thermal behavior patterns from training data consisting of layer sequences and thermal maps, substituting conventional thermal analysis methods with an intelligent system that achieves higher prediction accuracy for voxel-level thermal behavior in the build bed
Solution Approach 2:
The patent creates virtual thermal maps through neural network prediction that replicate actual thermal behavior. By training the network on real thermal data and layer sequences, it generates accurate predictions of thermal distribution that mirror physical thermal processes, enabling virtual optimization of printing parameters before actual printing
2Manufacturing precision
If thermal distribution is optimized to improve part quality, then manufacturing precision improves, but the complexity of the printing process increases
Solution Approach 1:
The patent performs thermal prediction and optimization before the actual printing process. By using the trained neural network to predict thermal behavior for planned layer sequences, the system allows modification of printing parameters in advance, preventing quality issues before they occur rather than requiring complex real-time adjustments during printing
Data Source
AI summary
A system includes a machine readable storage medium storing instructions and a processor. The processor is to execute instructions to receive contone agent maps of a three-dimensional (3D) part and sensed thermal maps from the 3D printing of the 3D part on a 3D printer. The processor is to execute instructions to generate layer sequences including the contone agent maps and the sensed thermal map for each layer of the 3D part. The processor is to execute instructions to select training samples from the layer sequences having temperature intensity variations within each layer or between neighboring layers. The processor is to execute instructions to train a neural network using the training samples to generate a model to predict thermal behavior in the 3D printer.