Heat Sink Thermal Analysis Using Physics-Constrained Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current thermal analysis methods for heat sinks face challenges in achieving both high speed and accuracy, with physical models being time-consuming and deviating from actual values, while machine learning models are fast but less reliable in areas without training data, failing to ensure results align with physics.
Innovation Solution
A machine learning program that uses a thermal analysis device with a machine learning model trained on shape and heat distribution information, employing a loss function that constrains temperature relationships across the heat sink, integrating a physics-based loss function to ensure accuracy and speed through a neural network, allowing for high-speed and accurate thermal analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physical models are used for thermal analysis, then accuracy of thermal analysis results is improved, but analysis time increases significantly
Solution Approach 1:
The patent pre-calculates thermal analysis results for various heat sink shapes and stores them in a database before actual design work begins. This preliminary computation allows the system to quickly retrieve and compare pre-computed results during the design phase, avoiding time-consuming real-time physical model calculations while maintaining accuracy.
Solution Approach 2:
The patent creates a database that copies and stores pre-computed thermal analysis results from physical models for various heat sink configurations. Instead of running physical model calculations during design iteration, the system copies and compares results from this pre-built database, significantly reducing analysis time while preserving the accuracy benefits of physical models.
2Productivity
If machine learning models are used for thermal analysis, then analysis speed is improved, but reliability decreases in areas without training data
Solution Approach 1:
The patent introduces a database as an intermediary layer between the physical model results and the machine learning model. The system first queries the database for matching heat sink shapes, and only when no match is found does it fall back to the machine learning model. This intermediary approach ensures reliability by prioritizing exact matches from physical models while maintaining the speed advantage of machine learning for novel configurations.
3Productivity
If machine learning models are used for thermal analysis, then analysis speed is improved, but alignment with physical principles deteriorates
Solution Approach 1:
The patent pre-computes thermal analysis results using physically accurate models and stores them in a database before the actual design process begins. This preliminary action captures the physical principles in advance, allowing the system to quickly retrieve results that respect physical laws without performing time-consuming physical calculations during design iteration.
Solution Approach 2:
The database acts as an intermediary that bridges the gap between machine learning speed and physical accuracy. When the system queries the database with heat sink shape parameters, it retrieves results that were pre-calculated using physical models, ensuring alignment with physical principles while maintaining fast analysis speed.
Data Source
AI summary
A non-transitory computer-readable recording medium storing a machine learning program for causing a computer to execute a process, the process includes obtaining training data that includes shape information of a heat sink that serves as an explanatory variable and heat distribution information of the heat sink that serves as an objective variable, and executing, based on the training data, machine learning of a machine learning model according to a loss function that includes an expression that constrains a temperature relationship of a plurality of positions in the heat sink.