大型风电机组齿轮箱早期故障诊断技术与系统研究

发布时间：2018-05-14 17:48

本文选题：大型风电机组齿轮箱 + 非线性动力学　；参考：《机械科学研究总院》2016年博士论文

【摘要】：近年来我国风电行业发展迅速,装机容量逐年递增。大型风机长期在野外工作,工况恶劣,很多早期机械故障很难被及时发现和治理,长时间运行演变为严重故障,甚至导致重大事故,严重影响风电企业的经济效益。在大型风机的多个关键部件中,齿轮箱是故障多发部件,其出现严重故障时,维修困难且维修成本极高,因此对大型风机齿轮箱进行早期故障诊断研究,以期及早发现齿轮箱的潜在故障,进行预知维护维修,对企业降低运行维护成本,提高经济效益具有重要意义。以风电场的主流机型即双馈式变桨变速机型的增速齿轮箱为主要研究对象,运用润滑油液金属磨粒在线检测与振动信号分析相结合的方法对齿轮箱早期故障诊断开展研究,首先通过在线检测润滑油液中的金属磨粒信息判断齿轮箱磨损程度及润滑油受污染程度,实现定性预判齿轮箱的早期故障,进而采用振动信号的分析方法深入分析故障原因及部位,实现齿轮箱的早期故障诊断。论文主要涉及如下内容：(1)大型风电机组齿轮箱非线性动力学研究。基于非线性动力学理论,在考虑时变啮合刚度条件下建立了行星齿轮传动的非线性动力学模型,得到了不同转速及负载工况下齿轮啮合和轴承支撑的正常、故障等条件下系统各个部件的时间曲线、频率、相图等。结果表明：齿轮啮合和轴承支撑正常、齿轮啮合故障和轴承支撑故障等条件下,输入轴转频对系统固有特征信号具有调制影响,导致系统响应频谱中各阶主频出现边频现象；齿轮啮合故障条件下,啮合频率2倍频或4倍频占主要能量；轴承支撑故障条件下系统支持刚度导致的频率出现左移现象,同时被转频调制现象明显。研究结果能够为开展信号特征提取提供分析数据和提供部分故障现象评价依据。(2)基于自相关系数谱阈值信号消噪方法及改进二阶自适应NLMS信号消噪方法的研究。大型风机由于工况恶劣,采集的振动信号中包含复杂的干扰噪声以及塔筒随机振动的低频噪声。为了最大程度的消除两种噪声成分,提出了基于自相关系数谱阈值信号的消噪方法用于消除随机干扰噪声,并以该方法为基础,进一步提出了分组自相关阈值去噪方法及阈值自动获取方法；提出了改进二阶自适应NLMS消噪方法用于消除齿轮箱振动信号中耦合的塔筒随机低频振动噪声成分。开展了仿真信号及实测信号验证分析,结果表明：该两种方法对于消除振动信号中的随机干扰噪声及耦合的塔筒随机低频噪声具有较好的预处理效果。(3)基于阶次重采样的希尔伯特变换解调自相关功率谱特征提取方法研究。首先,进行了转轴角度三次方程拟合的等角度重采样研究,分别通过仿真信号和实测信号验证了三次方程拟合法的有效性。然后,进行了等角度重采样信号的希尔伯特变换解调方法研究。最后运用理论仿真数据和试验数据,进行了角度重采样信号的平方计算解调、能量计算解调以及希尔伯特变换解调方法对比分析。结果表明：计算阶次重采样信号的希尔伯特变换解调后的自相关功率谱特征提取效果较其他两种方法有效,能够更为准确的进行信号特征提取。(4)基于波包阈值熵t-SNE流形学习故障分离方法研究。研究了基于小波包分解时域及频域的t-SNE故障辨识方法,通过实测数据验证了采用t-SNE降维处理后的流形结构清晰,特点突出,能够更好的用于辨识设备的故障状态。实验数据分析表明：该方法相比其他高维数据构造方法及流形学习方法具有更好的故障分离效果。(5)基于油液全液流在线磨粒检测的早期齿轮箱故障诊断方法研究。重点设计并研发了相应的磨粒检测传感器及检测仪器系统,提出了基于局部最大最小值的金属磨粒识别方法,进行了润滑油磨粒检测实验研究。结果表明：设计的金属磨粒检测系统能够检测到最小150微米的金属磨粒,达到了非常好的测量效果,仪器系统可用于大型风机齿轮箱早期不明显故障的预判,通过在线检测润滑油液中金属磨粒的尺寸、数量等信息及早发现齿轮箱的潜在故障。(6)远程风电机组传动系统早期状态监测诊断系统开发。为实现大型风机齿轮箱的远程早期故障诊断,设计了基于以太网的嵌入式数据采集系统,制定了基于TCP/IP的远程数据传输协议,基于微软的.net开发了B/S(浏览器/服务器)模式的远程监测及早期故障诊断系统软件。
[Abstract]:In recent years, the wind power industry in China has developed rapidly and the installed capacity is increasing year by year. Large fans are working in the field for a long time, and the working conditions are bad. Many early mechanical faults are difficult to be found and treated in time. Long time operation has evolved into serious faults, even leading to major accidents, which seriously affect the economic benefits of wind power enterprises. In the parts, the gear box is a fault multiple component. When it has serious failure, it is difficult to maintain and the maintenance cost is very high. Therefore, it is of great significance to study the early fault diagnosis of the large fan gear box, so as to predict the potential malfunction of the early present gear box and carry out the maintenance and maintenance. It is of great significance for the enterprise to reduce the cost of operation and maintenance and to improve the economic benefit. The main research object is the speed increasing gear box of the main type of the wind electric field, that is the double fed variable speed variable speed model. The study of the early fault diagnosis of the gear box is carried out by the method of on-line detection and vibration signal analysis of lubricating oil metal abrasive particles. First, the gear box is judged by the on-line detection of the metal abrasive information in the lubricating oil. The degree of wear and the degree of contamination of the lubricating oil can be used to determine the early fault of the gearbox, and then the analysis method of vibration signal is used to analyze the causes and parts of the fault. The main contents of this paper are as follows: (1) the nonlinear dynamics of the gearbox of the large wind turbine group. The nonlinear dynamic model of the planetary gear transmission is established under the condition of time-varying meshing stiffness. The time curve, frequency and phase diagram of the parts of the system under the conditions of gear meshing and bearing support under different speeds and load conditions are obtained. The results show that the gear meshing and bearing support are normal, Under the conditions of gear meshing failure and bearing support fault, the input shaft frequency has modulation influence on the inherent characteristic signal of the system, which leads to the occurrence of the main frequency in the response spectrum of the system; the meshing frequency is 2 frequency doubling or 4 frequency doubling of the main energy under the gear meshing fault condition, and the system support stiffness is caused by the bearing support failure. The results can provide analysis data for signal feature extraction and provide some basis for evaluation of fault phenomena. (2) study on the method of denoising based on the threshold signal of autocorrelation coefficient spectrum threshold signal and the improvement of the two order adaptive NLMS signal denoising method. In order to eliminate two kinds of noise components, a denoising method based on the autocorrelation coefficient spectrum threshold signal is proposed to eliminate the random interference noise, and based on this method, the packet autocorrelation threshold is further proposed. An improved two order adaptive NLMS denoising method is proposed to eliminate the random low-frequency vibration noise components of the coupling of the gear box vibration signals. The simulation and measured signal verification and analysis are carried out. The results show that the two methods are used to eliminate the random interference noise in the vibration signals and the results. The coupled tower barrel random low frequency noise has good preprocessing effect. (3) study on the autocorrelation power spectrum feature extraction method of Hilbert transform demodulation based on order resampling. First, the equal angle resampling study of the three times equation fitting of the rotating axis angle is carried out, and the three equation fitting is verified by the imitation real signal and the measured signal respectively. Then, the Hilbert transform demodulation method of the equal angle resampling signal is studied. Finally, the theoretical simulation data and the experimental data are used to carry out the square calculation and demodulation of the angle resampling signal, the energy calculation and demodulation and the Hilbert transform demodulation method. The results show that the order resampling is calculated. The autocorrelation power spectrum feature extraction effect after the signal Hilbert transform demodulation is more effective than the other two methods. (4) study on the fault separation method based on the packet threshold entropy t-SNE manifold learning. The t-SNE fault identification method based on the time domain and frequency domain of wavelet packet decomposition is studied. The data verify that the manifold structure with t-SNE dimension reduction is clear and characteristic, and it can be used to identify the fault status of the equipment better. The experimental data analysis shows that the method has better fault separation effect compared with other high dimensional data construction methods and manifold learning methods. (5) based on the oil liquid full liquid flow on-line abrasive detection The fault diagnosis method of early gear box is studied. The corresponding abrasive detection sensor and detecting instrument system are designed and developed. The metal abrasive recognition method based on the local maximum and minimum value is put forward, and the experimental research of lubricating oil abrasive particle detection is carried out. The result shows that the design of the gold particle detection system can detect the minimum of 150 micro. The metal abrasive grain of rice has achieved very good measurement effect. The instrument system can be used to predict the early non obvious fault of the large fan gear box. By on-line measuring the size and quantity of metal abrasive in the lubricating oil, the information of the quantity and the potential fault of the early present gear box. (6) the early state monitoring and diagnosis system of the transmission system of the long distance wind turbine is opened. In order to realize the remote early fault diagnosis of large fan gear box, an embedded data acquisition system based on Ethernet is designed, and a remote data transmission protocol based on TCP/IP is developed. Based on the.Net of Microsoft, the remote monitoring and early fault diagnosis system software of B/S (Browser / server) mode is developed.

【学位授予单位】：机械科学研究总院
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TM315;TH132.41

【参考文献】