Machine Learning CFD Data Reconstruction via Uniform Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
CFD simulations for IT devices face high processing and storage costs due to large data files, difficulty in sharing and collaboration due to specialized software requirements, and redundancy in data that is hard to compress, limiting accessibility for non-CFD personnel.
Innovation Solution
Utilizing a machine learning model with uniform spatial sampling and deep neural networks to convert CFD data into a compact format, enabling efficient data reduction and flexible analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If CFD simulation data is stored in traditional formats, then data completeness is maintained, but data file size becomes large and processing costs increase
Solution Approach 1:
The patent creates a compact representation (copy) of CFD data using machine learning models. Instead of storing complete raw CFD simulation data, the system trains a neural network model on the original data and stores only the model parameters and uniform sampling coordinates. This compact copy can reconstruct the original data with minimal loss, dramatically reducing storage requirements while maintaining data completeness for analysis purposes.
2Quantity of substance
If CFD data is compressed using traditional methods, then data size is reduced, but data accuracy and completeness are lost
Solution Approach 1:
The patent replaces traditional mechanical compression algorithms with a machine learning-based compression system. Instead of using lossy compression algorithms that discard data, the system uses a neural network model that learns the underlying patterns and relationships in CFD data. The model can reconstruct data at query points with high accuracy by leveraging learned physical relationships, maintaining data integrity while achieving compression.
3Adaptability or versatility
If CFD simulation results are shared with non-CFD personnel, then accessibility is improved, but specialized software requirements create barriers
Solution Approach 1:
The patent introduces a machine learning model as an intermediary between CFD simulation data and end users. The model acts as a translator that converts complex CFD data into a format accessible through simple coordinate queries. Non-CFD personnel can query the model using basic spatial coordinates without needing specialized CFD software or expertise, while the model handles the complex data reconstruction internally.
4Stability of the object's composition
If uniform spatial sampling is used for training, then data distribution is improved, but training data requirements increase
Solution Approach 1:
The patent applies uniform spatial sampling selectively at key locations rather than uniformly across the entire domain. The system identifies important regions and samples uniformly within those regions while using fewer samples in less critical areas. This partial application of uniform sampling maintains data distribution stability where needed while reducing overall training data requirements.
Data Source
AI summary
Embodiments of the present disclosure provide a method, an electronic device, and a computer program product for determining computational fluid dynamics (CFD) data. The method includes acquiring a spatial coordinate set associated with a target object. The method further includes determining a location coded set associated with the spatial coordinate set. The method further includes determining CFD data associated with the target object by using a machine learning model based on the location coded set, wherein a sample spatial coordinate set associated with training of the machine learning model includes a plurality of sample spatial coordinate subsets acquired through uniform spatial sampling. In this way, the cost of acquiring CFD data can be reduced and the speed of acquiring the CFD data can be improved.