【作者】 庞茂；

【导师】 周晓军；

【作者基本信息】 浙江大学 ， 机械制造及其自动化， 2006， 博士

【摘要】 工业生产和科学技术的发展对车辆运行的可靠性、可用性、可维修性等提出了更高要求，小波、分形和混沌等现代非线性信号处理方法的出现为车辆检测技术提供了新的理论和思路，不同程度地满足了发展的需求。而寻求更好的非线性时间序列分析方法已成为信号处理领域中的前沿课题。论文结合驱动桥、主减速器及差速器等汽车零部件性能试验机课题，将信号小波降噪、分形和混沌等现代非线性理论应用到车辆产品检测和信号处理中，对设备运行时表现出的各种非线性特征进行了深入研究，为分析车辆零部件的振动特性开辟了途径。主要研究内容有： 考虑主减速器运转中主传动锥齿轮齿侧间隙、时变啮合刚度和齿轮副综合误差等齿轮传动中非线性因素的基础上，运用8自由度锥齿轮非线性动力学模型分析主减速器振动特性，仿真研究主减速器周期、拟周期、混沌等三种典型振动形式，并深入分析了不同参数对系统非线性的动态响应特性的影响和关联维数、最大Lyapunov指数等非线性特征量值对于主减速器振动特性的表征能力。 针对现场测量的信号信噪比低，对信号分析影响较大，将基于解析小波变换模极大值的信号消噪技术应用到主减速器振动信号降噪中。解析小波变换仅反映信号的正频率，其模振荡较之实小波变换要小，为此将由Hilbert变换构造的解析小波基引入小波极大模信号降噪中，汽车主减速器振动信号的实例分析证明，与实小波极大模降噪方法相比，该方法具有更好的降噪效果。 基于GP算法的关联维数计算方法简单，但计算量大，且难以实现自动化。论文从降低关联积分计算量、提高关联维数计算速度和无标度区间的自动识别两个方面进行了探索。提出运用关联积分曲线的二阶局部斜率实现无标度区间的自动识别，从而实现关联维数的自动计算，运用该方法对Lorenz系统的分析证明了其有效性。进而分析了不同工作状态下主减速器振动信号的关联维数和最大Lyapunov指数。研究结果表明，不同状态下主减速器信号的关联维数和最大Lyapunov指数有明显的可分性，两者都可以作为识别信号特征和程度的有效量化指标。此外，论文将多个传感器信号的非线性特征量运用遗传编程优化得到复合特征，结果表明与采用传统的统计特征量作为优化终端符相比，非线性特征终端符得到的信号复合特征能够更好更多还原

【Abstract】 With the development of industry and science technology, higher reliability, usability and maintainability of machinery are expected. Modern non-linear signal processing methods, such as wavelet, fractal and chaos etc, have provided more advanced and reliable theory for vehicle test and satisfy the requirement of industry development to some extent. It is a promising subject in signal processing field to seek better approach for nonlinear time series analysis. Based on the projects of automobile parts performance test bed (such as drive axis, main reducer and differential), this article applies wavelet de-noising, fractal and chaos to product test and signal processing of automobile transmission, studies the nonlinear features of automobile deeply and opens up a new route to signal processing of complex vehicle parts vibration accurately. Main contents as follows:Thinking of nonlinear factors in gear pairs system: gear backlash, meshing stiffness, gear resultant error, 8 degree of freedom nonlinear kinetics model of main reducer is built, Differential equation is computed by 4-order Runge-Kutta numerical integration method. Three typical vibrations of main reducer are simulated, and the influence of different system parameters on nonlinear dynamic characteristics and the ability of correlation dimension and largest Lyapunov exponent reflecting dynamic performance of gears transmission system are analyzed.Due to the fact that vibration characteristics of faulty machinery are complex and defect-related vibration signal is normally buried in the wideband noise, de-noising method based on analytic Wavelet Transform Modulus Maximum (WTMM) is introduced into the noise reduction of automobile vibration signals. Analytic wavelet transform (AWT) only reflects positive frequencies of signals and its modulus oscillation is weaker than real wavelet transform (RWT), so signal noise reduction and singularity detect based AWT can be more accurate than RWT for the signal with additive white noise. Analytic wavelet based constructed by Hilbert transform is applied to the noise reduction based on WTMM. Experiment with automobile main reducer results show that noise reduction using modulus maximum of analytic wavelet is better than that of real wavelet.The theory of correlation dimension computation based on GP is concise, but the computation burden is heavy, and scaling region recognition automatically is hard. Analyzing the influencing factor of correlation dimension, a method to scaling region recognition and correlation dimension computation automatically based on second local slope of correlation integral is presented. The effectiveness of this method was verified by the analysis of Lorenz attractor. Data, which are sampled in an automobile main reducer performance test bed, is analyzed by this method. Experiments results show correlation dimension and largest lyapunov exponent of different main reducers are different, they can be used as the quantification factor of recognizing signal features and level. Moreover, nonlinear features coming from several sensors are constructed compound signal feature by GP. Compound signal feature based nonlinear features can distinguish various working state more更多还原