Machine Learning EMI Prediction with Simplified Circuit Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating electromagnetic interference (EMI) in electronic circuits require high-resolution discretization, leading to increased calculation costs for equivalent circuit analysis and training data generation, especially when modeling complex current distributions and edge interactions.
Innovation Solution
A machine learning program that simplifies circuit shapes by removing specific features like slits and slots, generates current distribution images, and uses electromagnetic field analysis to create an EMI prediction model, reducing calculation costs while maintaining accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution discretization is performed to accurately reproduce complex current distribution in plane patterns, then measurement precision of current distribution is improved, but calculation cost of equivalent circuit analysis increases
Solution Approach 1:
The patent divides the circuit board into multiple regions based on current distribution characteristics. High-resolution discretization is applied only to regions with complex current patterns (such as near slits and edge portions), while low-resolution discretization is used for regions with simple current patterns. This segmented approach maintains measurement precision where needed while reducing overall calculation cost.
Solution Approach 2:
The patent applies different discretization resolutions to different spatial locations based on their specific characteristics. Regions with slits, edge portions, or complex geometries receive high-resolution treatment, while uniform regions receive low-resolution treatment. This local quality principle ensures accurate current distribution measurement in critical areas without uniformly increasing calculation cost across the entire circuit board.
2Manufacturing precision
If high-resolution discretization is performed to model complex current distribution, then manufacturing precision of current modeling is improved, but calculation time increases
Solution Approach 1:
The patent segments the circuit board into multiple regions with different discretization resolutions. By identifying and isolating regions that require high modeling precision (such as areas with slits or edge portions), the patent applies high-resolution discretization only to those specific segments, thereby achieving accurate current modeling without the time penalty of uniform high-resolution discretization across the entire board.
Solution Approach 2:
The patent applies high-resolution discretization partially rather than excessively across the entire circuit board. By limiting high-resolution treatment to only those regions where it is necessary for accurate current modeling, the patent achieves the required manufacturing precision while avoiding the excessive calculation time that would result from applying high resolution uniformly throughout.
3Reliability
If high-resolution discretization is performed to accurately model plane patterns with slits and edge portions, then reliability of EMI estimation is improved, but cost of generating training data increases
Solution Approach 1:
The patent segments the circuit board into multiple regions based on their contribution to EMI characteristics. Regions with slits, edge portions, or complex geometries that significantly impact EMI are modeled with high resolution, while other regions use lower resolution. This segmentation enables reliable EMI estimation by focusing computational resources on EMI-critical regions, thereby reducing the overall cost of generating training data while maintaining estimation reliability.
Solution Approach 2:
The patent applies local quality by assigning different discretization resolutions to different spatial locations based on their EMI relevance. Areas with slits and edge portions that are known to generate or affect EMI are modeled with high precision, while regions with minimal EMI impact use lower resolution. This approach ensures reliable EMI estimation without the prohibitive cost of uniform high-resolution training data generation.
Data Source
AI summary
The present invention relates to a machine learning program for causing a computer to execute a process. In an example, the process executed by the computer when the program is executed includes: acquiring a current distribution image by an equivalent circuit simulation based on circuit information; acquiring a shape image that indicates a shape of a circuit based on the circuit information; acquiring an EMI value by electromagnetic field analysis based on the circuit information; and generating an EMI prediction model by machine learning based on training data that includes the current distribution image, the shape image, and the EMI value.