基于支持向量机的涡轮泵故障检测算法研究

发布时间：2018-06-24 15:32

  本文选题：涡轮泵 + 故障检测　； 参考：《电子科技大学》2015年硕士论文

【摘要】：涡轮泵是液体火箭发动机的核心部件,具有高昂的研制成本。涡轮泵恶劣的工作环境导致其具有很高故障率。因此,研究涡轮泵故障检测技术,对降低涡轮泵故障损失具有重大意义。目前,基于涡轮泵壳体振动信号的故障检测算法是该领域的一个研究热点。首先,总结并阐述了液体火箭发动机故障检测技术的国内外研究现状和涡轮泵故障检测算法的相关理论。在此基础上,以某型号液体火箭发动机涡轮泵历史试车的壳体振动加速度信号为研究对象,分别研究了两种基于支持向量机的涡轮泵故障检测算法。第一个算法是基于时域特征和快速支持向量机的涡轮泵故障检测算法。该算法以样本步长信号的能量和能量变化绝对值作为时域特征。同时,为解决由于训练样本过多所导致的训练缓慢甚至无法训练的问题,引入快速支持向量机方法,从原始训练样本集中筛选边界训练样本集,确保了决策分类函数的准确性,而且大大缩短了训练时间。同时,提出了一种多指标加权报警策略,为该算法判断检测步长信号中是否含有故障提供了依据。第二个算法是基于频域特征和模糊分类支持向量机的涡轮泵故障检测算法。该算法以样本步长信号的频段幅值标准差作为频域特征,将一个样本步长信号的频谱分成若干频段,计算各个频段幅值标准差并构造成一个向量,作为一个训练(检测)样本。同时,引入模糊分类支持向量机的方法,通过构造故障隶属度函数,以实现对故障检测样本的故障隶属度的计算。并且,将故障隶属度用于剔除误分故障检测样本,从而实现算法对虚警风险的控制,提高算法的准确性。经实验验证,以上两种涡轮泵故障检测算法均满足算法验证的准确性、实时性和及时性要求,对于改善液体火箭发动机涡轮泵试车的安全性,以及减少涡轮泵故障损失,具有一定的效果和积极的意义。
[Abstract]:Turbine pump is the core component of liquid rocket engine and has high development cost. Turbine pump has a high failure rate due to its poor working environment. Therefore, it is of great significance to study turbine pump fault detection technology to reduce turbine pump fault loss. At present, fault detection algorithm based on vibration signal of turbine pump shell is a research hotspot in this field. Firstly, the research status of liquid rocket engine fault detection technology at home and abroad and the relevant theory of turbine pump fault detection algorithm are summarized and expounded. On this basis, two turbine pump fault detection algorithms based on support vector machine are studied based on the vibration acceleration signal of the turbine pump of a liquid rocket engine. The first algorithm is based on time domain feature and fast support vector machine (SVM). The time domain feature of the algorithm is the absolute value of the energy and energy variation of the sample step-size signal. At the same time, in order to solve the problem that the training is slow or unable to train due to too many training samples, the fast support vector machine (FSVM) method is introduced to screen the boundary training samples from the original training samples to ensure the accuracy of the decision classification function. And greatly shortened the training time. At the same time, a multi-index weighted alarm strategy is proposed, which provides a basis for the algorithm to judge whether there are faults in the detection step size signal. The second algorithm is based on frequency domain feature and fuzzy classification support vector machine. In this algorithm, the frequency band standard deviation of the sample step size signal is taken as the frequency domain feature, the spectrum of a sample step size signal is divided into several frequency bands, the amplitude standard deviation of each frequency band is calculated and a vector is constructed as a training (detection) sample. At the same time, the method of fuzzy classification support vector machine is introduced, and the fault membership function is constructed to calculate the fault membership of fault detection samples. Furthermore, the fault membership degree is used to eliminate the fault detection samples, so that the algorithm can control the false alarm risk and improve the accuracy of the algorithm. The experimental results show that the above two turbine pump fault detection algorithms meet the accuracy, real-time and timeliness requirements of the algorithm verification, which can improve the safety of the liquid rocket engine turbine pump test operation and reduce the turbine pump failure loss. Have certain effect and positive meaning.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：V463
，

本文编号：2061997