【作者】 王彦岩；

【导师】 宋宝玉； 杨建国；

【作者基本信息】 哈尔滨工业大学 ， 车辆工程， 2010， 博士

【摘要】 近年来,随着石油资源的日益紧缺以及由此引发的燃油价格大幅提高,越来越多的发动机制造商开始采用更高的压缩比来达到提高燃油经济性的目的;同时,排放法规严格要求使用无铅汽油。这都导致汽油机发生爆震的可能性增大。爆震是限制火花点火发动机功率提高和经济性改善的一个重要因素,汽油机绝对不允许在强烈爆震的情况下工作,但当轻微爆震时,燃烧接近定容燃烧,汽油机的动力性与经济性均有所提高。现代汽油机大多都有爆震控制装置,但控制的原理、方法和精度一直在不断改进之中。研究爆震信号的特征提取方法,强度判定和诊断方法,并通过现代电子技术,实现基于爆震信号的实时、精确的点火闭环控制,使汽油机能够始终在轻微爆震附近(爆震边缘)工作,对进一步提高汽油机的性能、节省能源、减少排放具有重要的意义。在对汽油机进行爆震检测的基础上,研究了短时傅立叶变换(STFT)、平滑伪Wigner-Ville分布(SPWD)和离散小波变换(DWT)用于汽油机爆震特征提取的可行性和效果。发现:STFT和SPWD两种方法只有在适当滤掉信号的低频成分的前提下,才能从汽油机气缸压力信号和缸盖振动信号中提取出爆震特征;并且STFT和SPWD方法计算量大,耗时长;利用DWT方法可以直接从气缸压力信号和缸盖振动信号中有效地提取出轻微爆震特征,与前两种方法相比,速度更快,具有明显的优越性;而且,在只安装一只加速度传感器的情况下,利用DWT方法能够提取出4个气缸的爆震边缘特征。本文对20个时域参数进行了分析研究,找出了其中适合描述小波变换特征域信号爆震特征的指标。针对其它指标存在的随工况变化比较敏感的缺陷,提出了一种相对能量指标,该指标将小波子带信号和原始振动信号关联起来,对工况变化具有较好的适应性,适合用于爆震判定。提出了用多个特征值组合起来作为爆震判定指标判定爆震的方法,以弥补原来单个指标判定爆震方法存在的不足。在爆震诊断方法的研究中,将模糊聚类分析、神经网络方法应用到爆震诊断中,提出小波结合模糊C-均值聚类的爆震诊断方法和小波结合模糊神经网络的爆震诊断方法,给出了两种方法的原理,并分别用具体算例进行了检验,发现这两种方法都能够完成对汽油机燃烧模式的区分,并能诊断出爆震边缘燃烧。这两种方法相比,小波结合模糊C-均值聚类的爆震诊断方法更适合于用硬件实现。为改进现有汽油机爆震控制策略在实时性方面存在的不足,本文提出一种开环与闭环相结合,在一个工作循环内完成从信号检测到点火提前角调整的爆震控制策略。研究了用神经网络代替传统的点火MAP图来确定基本点火提前角的方法,建立了小波结合模糊C-均值聚类算法的爆震运算模型,通过仿真运算验证了模型的正确性。研究了爆震实时控制的实现方法,开发了爆震实时控制的两个关键部分:爆震运算单元和数据通讯单元。在FPGA硬件上实现了小波结合模糊C-均值的爆震运算算法,并对应用CAN总线实现FPGA与ECU之间的数据通讯的技术进行了研究。最后通过硬件试验验证了爆震运算单元和数据通讯单元的有效性。更多还原

【Abstract】 Recently,with the increasing scarcity of oil resources and substantial increase in fuel prices, more and more engine manufacturers began to use a higher compression ratio to achieve the purpose of improving fuel economy. At the same time, the strict emissions regulations require the use of lead-free gasoline. These would increase the possibility of the gasoline engine knock. Knock is an important factor which has a strong impact on the power and fuel enconomy of spar-ignition engine. The engine can never be allowed to work under conditions of a strong knock. However, when slight knock occurs, the engine is close to isometric burning, so the economy and power of the engine are enhanced. Many of the modern gasoline engines are equipted with knock control system, but the control principles, methods and accuracy have been constantly improving. To research the methods of knock signal feature extraction, intensity determination and diagnosis, and to achieve the real-time and accurate closed-loop ignition control based on the knock signal by modern electronic technology are of great significance to improving the gasoline engine performance, saving energy, and reducing emission of the engine.The article separately studied the feasibility and effectiveness of gasoline engine knock feature extraction methods, such as the Short-time Fourier Transform(STFT), smoothed pseudo Wigner-Ville distribution(SPWD) and the discrete wavelet transform(DWT). The study found that: when adopting the STFT and SPWD method, only under the condition of filtering out the low-frequency components of the signal can we extract the knock feature from the engine cylinder pressure signal and the cylinder head vibration signal; STFT and SPWD method also need a large amount of calculation and time consuming; DWT method can extract the slight knock feature well from the cylinder pressure signal and the cylinder head vibration signal; And compared with the previous two methods, DWT method is faster and has obvious superiority; Moreover, by adopting DWT method the slight knock feature of four cylinders can be extracted out under the condition of installing only one acceleration senor on the first cylinder head.Twenty time-domain parameters are analyzed and the indexes suited to describe knock features of wavelet transform domain signal are found out. A relative energy index is proposed to overcome the defect that other indexes are sensitive to the changes of working conditions. The relative energy index associated wavelet sub-band signal with the original vibration signals, has better adaptability to work condition changes. It is suitable for knock judgement. The method of using multiple eigenvalues as knock judgement indexes is proposed to compensate for deficiency of the single index in knock judgement.In the study of knock diagnosis methods, fuzzy cluster analysis and neural method are applied to knock diagnosis. The knock diagnosis methods of wavelet with fuzzy C-means clustering and wavelet with fuzzy neural network are put forward. The principles of the two methods also be discussed. The two methods are verified effective in distinguishing engine combustion modes by some simulation tests, and slight knock combustion is diagnosed by the two methods. Comparing the two methods, wavelet with fuzzy C-means method is more suitable for implementing by hardware in knock diagnosis.In order to conquer the shortcomings in real-time performance of the existing gasoline engine knock control strategy, a new knock control strategy combined the open-loop control with closed-loop control is put forward. Using this control strategy can complete the works including signal detection and ignition advance angle adjustment in one engine work cycle. The mehod of using neural network instead of traditional experimental MAP to determine the basic ignition advance Angle is studied. The simulation model of knock operation which using wavelet combined with fuzzy C-means clustering algorithm is established. The validity of the model has been proved by the simulation computation.Finally, the implement method of knock control is studied. Knock computing unit and data communication unit, which are two key parts of the knock control are developed. The wavelet combined with fuzzy C-means algorithm is realized in the FPGA hardware. Controller area network bus technology is applied to the data communicaiton betweeen FPGA and ECU. The effectiveness of the two units are verified by the hardware test.更多还原