基于确定学习理论的低速轴流压气机旋转失速检测—仿真与试验研究

发布时间：2018-05-06 11:42

本文选题：轴流压气机 + 旋转失速　；参考：《华南理工大学》2015年博士论文

【摘要】：旋转失速和喘振是压气机常见的气动失稳现象,会造成压气机中流动情况恶化,压比和效率下降,甚至会导致叶片断裂,结构损坏和空中停车,严重危及飞行安全。如果能及时可靠地避免旋转失速/喘振的发生,对于提高航空发动机寿命及其性能和保障人身安全具有重要意义。旋转失速一般被认为是喘振的先兆,因此,捕捉旋转失速信号显得更为重要。本文基于确定学习理论,研究轴流压气机内部不稳定流动的建模,提前检测旋转失速和喘振的发生,以扩大压气机稳定运行范围,达到改善压气机性能的目的。主要成果和创新点概述如下:1、本文开展了低速轴流压气机模态波型失速的在线试验研究,以北京航空航天大学航空发动机重点实验室的低速轴流压气机实验台为研究对象,基于确定学习理论及动态模式识别方法,实现模态波型失速的在线提前检测。首先,在压气机机匣壁面周向布置多个动态压力传感器,获取压气机失速前和失速先兆的动态压力信号,进行离线数据处理,对模态波型旋转失速初始扰动的内在系统动态近似准确建模,并把结果存储在常值径向基函数(RBF)神经网络(NN)中。其次,研究在线试验的传感器布局、数据处理和实时性计算等,实现基于Lab VIEW的旋转失速检测系统,利用微小振动故障检测方法,在不同转速情况下,提前0.3-1秒实现对旋转失速的实时在线提前检测。2、本文研究了低速轴流压气机进口畸变下的失速检测。进口畸变是航空发动机稳定边界缩小和稳定性下降的重要因素之一,会加剧压气机内部流场的不稳定现象,甚至会引起压气机喘振的发生。因此,对进口畸变的非定常流动的捕捉为进一步提高叶轮机械的性能和稳定性有着非常重要的意义。本论文基于确定学习理论实现在进口畸变情况下预测流动失稳的发生。实验在北航航空发动机重点实验室的一台低速轴流压气机实验台上进行,利用插板扰流器模拟进口畸变的发生。进口畸变会增加不稳定流动干扰,使微弱的失速先兆信号更难捕捉。首先,研究故障估计器参数设置对故障残差的影响,寻找最优故障估计器参数,以准确预测出微小振动故障的发生。其次,利用机匣壁周向布置高频响应传感器获得进口畸变条件下动态压力数据,根据提出的基于确定学习的失速检测方法实现对畸变条件下失速先兆的检测。实验结果表明提出的方法可以完成对进口畸变下失速的提前检测。3、本文针对具有传播速度较快的小尺度扰动-突尖型失速开展建模与检测研究。突尖型失速是小尺度局部扰动,比模态波型发展速度更快,是在轴流压气机中更常见的流动崩溃现象。由于突尖型失速先兆的局部特性和流量的急剧衰减,所以很难对其进行失速前的检测。因此捕捉旋转失速或者喘振发生前的突尖型失速对主动控制更有意义。本文分析高阶Moore-Greitzer模型(Mansoux模型),开展了突尖型失速的建模和快速检测研究。首先,基于MIT的Mansoux-C3模型仿真研究,分析其失速初始扰动类型;其次,研究通过改变RBF神经网络结构参数、寻找最优RBF神经网络结构等方法提高微小振动信号的持续激励水平,并进而提高确定学习性能,实现对突尖型旋转失速进行近似准确动力学建模的方法。再次,利用确定学习理论对突尖型失速的未知系统内部动态进行局部准确建模;最后,在主要系统动态近似准确建模的基础上,实现对突尖型失速的快速检测。本文分析和研究了模态波型失速、进口扰动以及突尖型失速,并进行了在线实验。提出的失速检测方法在低速轴流压气机旋转失速检测的仿真和试验研究中得到验证。
[Abstract]:Rotating stall and surge are the common aerodynamic instability of the compressor, which will cause the deterioration of the flow in the compressor, the pressure ratio and the decrease of efficiency, even the blade fracture, structural damage and air parking, which seriously endangers the flight safety. If the rotating speed / surge is avoided in time and reliably, the life of the aero engine can be improved and the life of the aero engine can be improved. Its performance and safety are of great significance. Rotating stall is generally considered to be the precursor of surge. Therefore, it is more important to capture the rotating stall signal. Based on the theory of learning, this paper studies the modeling of unsteady flow inside the axial compressor and detects the occurrence of rotating stall and surge ahead of time in order to increase the stability of the compressor. The main achievements and innovation points are summarized as follows: 1. In this paper, the on-line test of modal wave velocity of low speed axial compressor is carried out, and the research object of the low speed axial compressor test platform in the Key Laboratory of the Beihang University is to determine the learning theory and the motion. The state pattern recognition method is used to realize the on-line early detection of modal wave type stall. First, a number of dynamic pressure sensors are arranged in the circumferential direction of the compressor casing wall to obtain the dynamic pressure signal of the compressor stall and the stall precursors, and the off-line data processing is carried out. The internal system dynamics of the initial disturbance of the modal wave type rotating stall is approximately accurate. The results are stored in the constant value radial basis function (RBF) neural network (NN). Secondly, the sensor layout, data processing and real time calculation of the on-line test are studied. The rotating stall detection system based on Lab VIEW is realized, and the micro vibration fault detection method is used to realize the rotational stall at 0.3- 1 second in advance at different speeds. In real-time online early detection.2, this paper studies the stall detection under the inlet distortion of the low-speed axial compressor. The inlet distortion is one of the important factors for the reduction of the stability boundary of the aeroengine and the decline of the stability of the aeroengine. It will aggravate the instability of the internal flow field of the compressor, and even cause the compressor surge. The capture of unsteady flow is of great significance to further improve the performance and stability of turbomachinery. This paper is based on the determination of learning theory to predict the occurrence of flow instability in the case of imported distortion. The plate spoiler simulated the occurrence of the inlet distortion. The inlet distortion will increase the unstable flow interference and make the weak stall signal more difficult to capture. First, the effect of the parameter setting of the fault estimator on the fault residuals is studied, and the parameters of the optimal fault estimator are found to accurately predict the occurrence of small and small vibration faults. Secondly, the circumference of the casing wall is used. A high frequency response sensor is arranged to obtain dynamic pressure data under the condition of imported distortion, and the detection of stall precursors under distortion conditions is realized based on the proposed method based on Determination of learning based stall detection. The experimental results show that the proposed method can complete the early detection.3 for the imported distortion stall, and this paper is aimed at the fast propagation speed. The modeling and detection of small scale disturbance - apex stall is carried out. The sudden tip type stall is a small scale local disturbance, which is faster than the modal wave type. It is a more common flow collapse in axial compressor. It is difficult to detect the stall before stall due to the sharp decline of local characteristics and flow of the sudden tip type stall. Therefore, it is more meaningful to capture the prop type stall before the rotating stall or the surge occurred. In this paper, the high order Moore-Greitzer model (Mansoux model) is used to develop the modeling and rapid detection of the sudden tip type stall. First, the Mansoux-C3 model simulation based on MIT is used to analyze the initial type of the stall initial disturbance. Secondly, the research is done. By changing the structure parameters of the RBF neural network and searching for the optimal RBF neural network structure, the continuous excitation level of the small vibration signals is improved, and the learning performance is improved, and the approximate accurate dynamic modeling method for the sudden sharp rotating stall is realized. Finally, on the basis of the approximate accurate modeling of the main system dynamic and approximate accurate modeling, the rapid detection of the sudden tip type stall is realized. In this paper, the modal wave type stall, the inlet disturbance and the sudden tip type stall are analyzed and studied, and the on-line experiment is carried out. The proposed method of stall detection is used in the rotating stall of the low speed axial compressor. The test is verified in the simulation and experimental research.

【学位授予单位】：华南理工大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：V233

【参考文献】