Generative Adversarial Network for Plane Cascade Flow Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting steady flow in axial flow compressors are plagued by low accuracy and poor reliability due to the complexity of the system and limited experimental data, making it difficult to model and simulate the flow field effectively.
Innovation Solution
A steady flow prediction method using a generative adversarial network is developed, which preprocesses simulation data from Computational Fluid Dynamics (CFD) experiments, constructs an Encoding-Forecasting network module, and employs a deep convolutional network module within a generative adversarial network framework to predict flow field images in a plane cascade, enhancing prediction accuracy and detail clarity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If mathematical and physical methods are used to establish equations for simulating the flow field, then the model can be constructed, but the accuracy is poor due to systematic uncertainty and complexity of internal evolution
Solution Approach 1:
The patent replaces traditional mathematical and physical modeling methods with a data-driven deep learning approach. Instead of establishing complex equations based on mechanical and physical principles, the invention uses convolutional neural networks to directly learn flow field patterns from sensor data, thereby avoiding the systematic uncertainties inherent in traditional modeling while improving prediction accuracy
Solution Approach 2:
The patent creates a virtual copy of the flow field using deep learning models. By training neural networks on historical sensor data, the system generates predicted flow field images that replicate the behavior of the actual flow field without requiring direct measurement or complex physical modeling, thus improving both reliability and accuracy
2Measurement precision
If sensors are deployed at fixed measuring points to collect data, then the state characteristics can be analyzed, but the measurement range is limited and cannot reflect the whole axial flow compressor flow field
Solution Approach 1:
The patent makes the flow field prediction system universally applicable across the entire axial flow compressor by using deep learning models trained on comprehensive sensor data. The trained model can predict flow field conditions at any location within the compressor, transforming limited sensor measurements into comprehensive flow field information that covers the entire system
Solution Approach 2:
The patent introduces a deep learning-based flow field prediction model as an intermediary between fixed sensor measurements and the complete flow field state. This intermediary process reconstructs the full flow field distribution from limited sensor data, enabling comprehensive flow field coverage without deploying sensors throughout the entire system
3Measurement precision
If CFD simulation experiments are conducted to obtain flow field image data, then comprehensive flow field information can be obtained, but the computational resources required are significant
Solution Approach 1:
The patent creates simplified copies of the flow field using lightweight deep learning models that have been trained on CFD simulation data. Once trained, these models can generate accurate flow field predictions without requiring repeated CFD simulations, dramatically reducing computational resource consumption while maintaining prediction accuracy
Solution Approach 2:
The patent performs comprehensive CFD simulations and model training in advance to create pre-trained deep learning models. These pre-trained models can then rapidly predict flow field conditions for new operating conditions without requiring additional CFD computations, thereby reducing real-time computational resource requirements
Data Source
AI summary
A steady flow prediction method in a plane cascade based on a generative adversarial network is provided. Firstly, CFD simulation experimental data in the plane cascade are preprocessed, and a test dataset and a training dataset are divided from the simulation experimental data. Then, an Encoding-Forecasting network module, a deep convolutional network module and a generative adversarial network prediction model are constructed successively. Finally, prediction is conducted on test set data: the test set data is preprocessed in the same manner, and data dimensions are adjusted according to input requirements of a saved optimal prediction model; and flow field images in the plane cascade at an inlet attack angle of 10° are obtained through the prediction model. The present invention can effectively avoid the problem of limited measurement range of sensors in an axial flow compressor, and the prediction result is highly consistent with the calculation result of CFD.