【作者】 吴震宇；

【导师】 袁惠群；

【作者基本信息】 东北大学 ， 机械设计及理论， 2010， 博士

【摘要】 内燃机是工农业生产中的主要动力机械之一,它在石油钻井、船舶、汽车、铁路、农业、工程建筑等方面获得了广泛应用,其运行状态的好坏,直接影响到整个机组的工作状态。因此,对其状态监测和故障诊断,确保系统正常、安全运行,处于最佳运行工况,提高设备的维修质量和效率是十分必要的,并且也具有重要的经济意义。内燃机故障诊断技术是利用内燃机的状态信息和历史状况,通过分析和处理来定量识别其实时技术状态并预测异常故障未来技术状态的一门建立在多学科基础上的综合技术。本研究试图从工程应用的角度出发,以内燃机为研究对象,将经验模式分解、蚁群算法、支持向量机、混沌、模糊聚类、非线性振动、随机振动以及动力可靠性等先进理论和技术相结合,对内燃机振动信号降噪、故障特征信息提取、状态识别、状态预测、动力学特性以及可靠性评价等多个方面的问题进行了系统的研究。主要研究工作如下：(1)针对经验模式分解(Empirical Mode Decomposition,简称EMD)中存在的包络线拟合问题,提出一种改进三次样条插值算法,仿真实验表明,该方法较好地解决了三次样条插值算法容易引起的过冲和欠冲现象,将其与能够有效抑制端点效应的包络极值延拓法相结合,可以提高EMD在信号特征提取中的合理性和准确性。(2)支持向量机(Support Vector Machine,简称SVM)中参数的选择将直接影响分类和预测准确率,为避免传统的网格搜索参数带来的时间消耗和搜索范围难于确定问题,将具有良好优化性能的蚁群优化技术应用到支持向量机惩罚函数和核函数参数的优化,提出了蚁群支持向量机方法。通过蚁群参数优化和网格搜索参数优化的仿真实验表明：蚁群参数优化算法比网格搜索算法能更快更优的搜索到SVM的主要参数。(3)针对内燃机故障振动信号的非线性、非平稳特征以及内燃机故障诊断中难以获得大量典型故障样本的实际情况,提出基于固有模态混沌特征和SVM的内燃机故障诊断方法。实验结果表明,与直接从原始信号中提取混沌特征相比,固有模态混沌特征更能够有效描述原始信号的非线性、非平稳的特征,与SVM相结合,解决了内燃机故障诊断中难以获得大量典型故障样本的实际困难。(4)针对内燃机故障振动信号的非平稳性,同时考虑到对于同一类型的故障而言,故障状态的变化往往是一个渐变的过程,很难从特征量上直观地分辨出故障的变化程度,因此提出一种将固有模态能量比和模糊聚类相结合的内燃机故障诊断方法。实验结果表明,将固有模态函数的能量比作为故障诊断的特征量,能够反映各频带内隐含的故障特征信息,与改进的模糊C-均值聚类算法相结合,同时解决了具有模糊性的故障类型的辨识问题。(5)为了解决神经网络预测方法易陷入局部最小值、精度与泛化不可调和的矛盾,将SVM方法应用到内燃机状态预测；通过引入相空间重构的方法,解决了用于建立SVM预测模型所需输入矩阵构建的不合理性；采用EMD方法对原始信号进行降噪处理,不仅可以克服小波变换对小波基选取的依赖性,更重要的是为相空间重构过程中准确计算振动信号时间序列嵌入维数和时间延迟奠定基础,实验结果表明,该方法综合了现有方法各自的优点,能够比较准确地预测内燃机振动信号的变化趋势,性能优于传统的分析方法,对内燃机状态预警具有一定的参考意义。(6)考虑了轴系的质量偏心、变转动惯量、活塞与气缸间的非线性于摩擦和非线性弹性恢复力等因素,利用拉格朗日方程建立了内燃机轴系弯扭耦合非线性动力学模型,分别应用数值积分方法和解析方法对轴系的非线性动力学特性进行了仿真。结果表明：质量偏心、变转动惯量和活塞与气缸间的非线性干摩擦对轴系的非线性振动行为都有不同程度的影响。所得结论为改善系统动力学特性,保证内燃机安全稳定运行提供了理论依据。(7)综合考虑内燃机轴系激振力矩复杂、随机性的特点,引入精确高效的虚拟激励法,对内燃机轴系在非平稳随机激励作用下的动力特性进行了系统的研究,并基于首次超越破坏准则数学模型,分析了轴系在随机载荷作用下的动力可靠性问题。仿真结果表明：发火间隔角、随机激励的强度和传递路径对轴系动力可靠性都有一定影响,所得结论从定性的角度为降低轴系振动、延长使用寿命提供了参考。更多还原

【Abstract】 Internal combustion engine(ICE) is one of the main power machinery in industrial and agricultural production, which is widely applied in oil drilling, marine, automotive, railway, agriculture, engineering and construction, etal. Its operation has a driect impact on working status of the entire unit. Therefore, its condition monitoring and fault diagnosis, to ensure the system operating normally, in the best conditions and improve the quality and efficiency of the maintenance of equipment is necessary, and also have important economic significance.Fault diagnosis technology of ICE is an integrated technology based on a multidisciplinary basis, and the real-time and future technical status can be identified and forecasted quantitatively by analyzing and processing status information and historical conditions of ICE. The research regards ICE as the research object from the engineering application perspective, and vibation signal de-noising, faults feature extraction, pattern recognition, condition prediction, dynamic characteristics and reliability evaluation are systematically studied based on integrating some advanced theories and techniques such as the empirical mode decomposition, ant colony algorithm, support vector machine, chaos, fuzzy clustering, nonlinear vibration, random vibration and dynamic reliability. The main research works are as follows:(1) For the envelope fitting problem existing in empirical mode decomposition(EMD), a modified cubic spline interpolation algorithm was proposed. Simulation experiments showed that the method can solve the cubic spline interpolation algorithm easily resulting in the phenomena of overshoot and undershoot, and which can improve reasonableness and accuracy of EMD in the signal feature extraction by combined with extremum envelope extension method which can effectively inhibit end effect.(2) Parameters of support vector machine(SVM) will directly affect the classification and prediction accuracy, in order to avoid the traditional grid search parameters bringing the time consuming and determining research scope difficultly, an ant colony optimization was used to select parameters of SVM; and a novel algorithm "ant colony optimization support vector machine"(ACO-SVM) was put forward, and compared with grid search parameter optimization. Simulation experiments showed that ant colony optimization algorithm can search the main parameters of SVM faster and better than the grid search algorithm.(3) For the non-linear and non-stationary characteristics of fault vibration signals of ICE and the difficulty to obtain a large number of fault samples in practice, a fault diagnosis scheme for ICE was proposed by using SVM and chaos characteristics based on intrinsic mode function(IMF). Experimental results showed that compared with extracting the chaos characteristics directly from the original signal, the chaos characteristics of IMF are more able to effectively describe the non-linear and non-stationary characteristics of the original signals, and the practical difficulty to obtain a large typical number of fault samples is solved by combined with SVM.(4) For the non-stationary characteristics of fault vibration signals of ICE, while considering the fault status changing was often a gradual process for the same fault type, it was difficult to visually distinguish the changing degree of fault by the features. A new method was presented for the fault diagnosis of ICE based on the energy ratios of IMF and fuzzy clustering. Experimental results showed that taking the energy ratios of IMF as the characteristics for the fault diagnosis of ICE, which can reflect the underlying fault features in different bands, and identification problem of fuzzy fault type is solved by combined with improved fuzzy C-means clustering algorithm at the same time.(5) In order to solve the neural network prediction method easily falling into local minimum and the contradictions between the precision and generalization, SVM was applied into the ICE condition prediction; the problem that the required input matrix for building SVM prediction model was unreasonable can be solved based on phase-space reconstruction method; The EMD using the original signal de-noising can not only overcome the dependence on the wavelet basis in wavelet transform, and which can lay the foundation for accurately calculating embedding dimension and time-delay of vibration signals in the process of phase-space reconstruction. Experimental results showed that the method combines the advantages of the existing methods, and is able to accurately predict the trends of ICE vibration signals, the performance is superior to the traditional analysis methods, it has a certain reference significance for status early-warning of ICE. (6) A coupled bending-torsional nonlinear dynamic model for crankshaft system of ICE was established by considering mass eccentricity, varied inertia and nonlinear dry friction as well as nonlinear elasticity between piston and cylinder. The nonlinear dynamic characteristics of the shafting were simulated by numerical integration method and analytical method respectively, The results showed that the mass eccentricity varied inertia and nonlinear dry friction between piston and cylinder have some effects on the nonlinear vibration of the shafting. The conclusions provide a theoretical basis for improving the dynamic characteristics of the system and ensuring safe and stable operation of ICE.(7) For the complicated and random characteristics of the exciting torque on the shafting of ICE, the dynamic characteristics of the shafting subjected to the non-stationary random excitation were studied by the accurate and efficient pseudo-excitation method, and the dynamic reliability of the shafting subjected to random loads was analyzed based on first excursion failure criterion model. The simulation results showed that fire angle, intensity and transfer paths of excitations have effects on the dynamic reliability of the shafting, and the conclusions provide a qualitative reference for reducing shafting vibration and prolonging service life.更多还原