Abstract:Deep learning has been used in many areas, such as feature detections in images and the game of go. This paper presents a study that attempts to use the deep learning method to predict turbomachinery performance. Three different deep neural networks are built and trained to predict the pressure distributions of turbine airfoils. The performance of a library of turbine airfoils were firstly predicted using methods based on Euler equations, which were then used to train and validate the deep learning neural networks. The results show that network with four layers of convolutional neural network and two layers of fully connected neural network provides the best predictions. For the best neural network architecture, the pressure prediction on more than 99% locations are better than 3% and 90% locations are better than 1%.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to predict the pressure distribution of turbine blades using deep - learning methods. Specifically, the research aims to accurately predict the static - pressure distribution on the surface of turbine blades by constructing and training different deep neural network (DNN) models. This problem is of great significance in the design and optimization of aero - engines and other turbomachinery. ### Research Background Traditional prediction of turbine - blade pressure distribution usually depends on computational fluid dynamics (CFD) simulations, but this method has high computational costs and is time - consuming. With the development of deep - learning technology, especially the successful application of convolutional neural networks (CNN) in fields such as image recognition, researchers hope to explore the potential of deep - learning in turbine - blade performance prediction to improve prediction efficiency and accuracy. ### Specific Problem Description 1. **Data Preparation**: In order to train the deep neural network, researchers first need a data set containing a large number of turbine blades and their pressure distributions. To this end, they generated a turbine - blade library containing 63,450 different geometries based on a benchmark 2D turbine blade, and used the MISES solver to obtain the pressure distribution of each blade. 2. **Model Selection and Training**: Researchers constructed three different deep neural - network architectures: - A two - layer fully - connected neural network (nn2) - A two - layer convolutional neural network plus a two - layer fully - connected neural network (cnn2_nn2) - A four - layer convolutional neural network plus a two - layer fully - connected neural network (cnn4_nn2) These models are respectively used to predict the pressure coefficient \( C_p \) at different positions of the turbine blade and evaluate their prediction performance. 3. **Performance Evaluation**: By comparing the prediction results of deep - learning models with CFD simulation results, the prediction accuracy of each model is evaluated. Special attention is paid to the performance of the model at different axial positions and the influence of different network architectures on prediction accuracy. ### Main Objectives - **Improve Prediction Accuracy**: Find the deep neural - network architecture that can most accurately predict the pressure distribution of turbine blades. - **Reduce Computational Costs**: Compared with traditional CFD methods, deep - learning models can significantly reduce the computation time while ensuring a certain level of accuracy. - **Explore New Methods**: Introduce deep - learning technology into the turbomachinery field and inspire more related research. ### Conclusion The research shows that deep neural networks with convolutional layers (such as cnn4_nn2) perform excellently in predicting the pressure distribution of turbine blades. The prediction error at more than 99% of the positions is less than 3%, and the error at 90% of the positions is less than 1%. This indicates that the deep - learning method has broad application prospects in turbomachinery performance prediction.

Pressure Predictions of Turbine Blades with Deep Learning

Deep-Learning Strategy Based on Convolutional Neural Network for Wall Heat Flux Prediction

Real-time power prediction approach for turbine using deep learning techniques

A deep learning framework for aerodynamic pressure prediction on general three-dimensional configurations

Deep learning method for fast prediction of film cooling performance

Deep learning-based predictive modelling of transonic flow over an aerofoil

Dual-convolutional neural network based aerodynamic prediction and multi-objective optimization of a compact turbine rotor

Deep Learning Based Fast Prediction and Optimization of Aerodynamic Performance for a Propeller with Gurney Flap

Aerodynamic Prediction on the Off-Design Performance of a S-CO2 Turbine Based on Deep Learning

Implicit Neural Representation For Accurate CFD Flow Field Prediction

Application of Machine Learning to Forced Response Predictions of an Industrial Axial Compressor Rotor Blade

Deep learning modelling of manufacturing and build variations on multi-stage axial compressors aerodynamics

Prediction of transonic flow over supercritical airfoils using geometric-encoding and deep-learning strategies

Flow Field Reconstruction of Compressor Blade Cascade Based on Deep Learning Methods

Classification and prediction of gas turbine gas path degradation based on deep neural networks

Novel Pressure-Based Optimization Method Using Deep Learning Techniques

Fast prediction of aerodynamic noise induced by the flow around a cylinder based on deep neural network

Deep Learning-Based Modeling Method for Probabilistic LCF Life Prediction of Turbine Blisk

Wind Power Prediction and Pattern Feature Based on Deep Learning Method

Fast Aerodynamic Prediction of Airfoil with Trailing Edge Flap Based on Multi-Task Deep Learning

Bidirectional prediction between wake velocity and surface pressure using deep learning techniques