The Realization Of Stall Detection For Axial Flow Compressors Based On Deterministic Learning Theory On The Computing Cluster

Turbofan engine known as the aircraft's heart is the general propulsion device which is commonly used in both military and civilian aircraft.Axial flow compressor is one of the core equipment of aeroengine,and its performance has important influence on the overall performance of aeroengine.Rotating stall and surge are two types of instabilities in the axial compressor,which may cause the axial compressor performance deterioration,evenly the engine stalled and engine off in flight.These pose a serious threat to flight safety,and it is necessary to avoid rotating stall and surge in flight.Rotating stall is usually considered as the sign of surge,so it has important significance for the performance of the axial flow compressor to detect stall inception in time.Based on its strong points and importance both in research and practical application,stall inception detection will be further studied based on the deterministic learning theory in this dissertation.The main contribution and innovation of this dissertation are summarized as follows:1.Based on parallel computing algorithm,the existing rotating stall detection system is reconstructed to solve the large-scale computation encountered in the research process.The main bottlenecks encountered in the stall detection of an axial compressor based on deterministic learning are analyzed and two types of factors that cause the large-scale computation are summarized as follows: Firstly,to construct large-scale patterns library with abundant information,large-scale computation is required;secondly,large-scale calculation is also required for stall inception detection with the multi-mode patterns.Therefore,in this paper existing stall detection system is analysed to study the inherent parallelism of stall detection.The task-level parallelism is adopted for a large number of patterns training,and vector operation is adopted for large-scale computation in an pattern training.By adopting a variety of parallelization methods,the existing stall detection system is reconstructed to run on various computing platforms,such as desktop computers,clusters and even computing clouds.2.Using the PCT and MDCS in the MATLAB,an CPU and GPU computer cluster is constructed.Firstly,the cluster of hardware platform which consists of six computers with the same performance is constructed by the use of local area network.Secondly,the licensed MATLAB,including PCT and MDCS,is installed on each computer to create a unified software support platform.Using the visual cluster management tool provided by MDCS,an accessible cluster object consists of six computers is constructed.Finally,to get access to the cluster object,MATLB job manger is installed on the computer treated as the primary node,and four computing units are turned on in each computer.The cluster can provide computing services for multiple users in the local area network.3.Based on the data from a low-speed axial flow compressor test rig of Beijing University of Aeronautics and Astronautics,an experiment is conducted to verify the effectiveness of the reconstructed system on computer cluster.Firstly,based on the computer cluster,reconstructed stall detection system can perform a number of training modes and complete the stall detection under the 90 patterns in a short time.Then,the performance contrast between different scale pattern library and different scale cluster is obtained.The results show that the reconstructed stall detection system on cluster platform can greatly improve the efficiency of the system;At the same time,an excellent analysis platform for stall inception detection in the axial flow compressor is built,which promotes the research of projects.