【作者】 姚建军；

【导师】 向阳；

【作者基本信息】 武汉理工大学 ， 轮机工程， 2006， 硕士

【摘要】 柴油机气缸压力的测量和分析对于改善内燃机的工作性能、进行柴油机最优点火和喷油时刻控制,排放控制以及状态监测和故障诊断都具有很重要的作用。压力获取比较常用的实验室方法是在气缸盖上打孔,加工测压通道,用压力传感器直接测量气缸压力。这种方法需要安装耐高温、耐冲击的压力传感器,而且传感器的安装会改变柴油机的性能,无法用于在线实时测量。因此,通过其它容易从外部测试的信号来识别气缸压力的非直接方法就有很实际的应用价值。本文在此背景下,研究了柴油机气缸压力恢复的振动识别方法。主要研究内容如下: 1.首先在4135柴油机试验台架上进行了测试试验,测试了柴油机第四缸压力信号、气缸盖的振动响应信号和上止点信号。分析了柴油机的缸盖系统的激励源和振动响应信号的特性,把缸盖系统看成一个多输入单输出的线性系统,对测试数据进行时域统计平均预处理,提高信噪比,通过时域加窗截取对于气缸压力的那一段振动响应信号,去除邻缸的干扰和其它不相关的激励力的影响。再设计一个有限冲击响应低通滤波器,消除高频噪声,求出系统的传递函数,恢复气缸压力。还论证了时域统计平均预处理方法的应用和测点的选取对提高缸内压力识别的精度的重要性,此外还比较了识别压力和实测压力在特征值上的误差,论证了传递函数方法的适用性。 2.用倒频谱方法识别出柴油机的气缸压力,给出最能反映柴油机燃烧分析中性能参数的识别数值进行比较,并分析倒频谱方法识别的特点。 3、分析了柴油机气缸盖系统激励和响应信号的非线性特性,阐述了基于BP神经网络的柴油机识别方法。对测试的振动响应信号进行时域统计平均和低通滤波后,训练BP神经网络,利用自适应梯度下降算法,自适应调节学习速率,提高网络精度,识别柴油机气缸压力。并分析了中间层神经元个数和输入层神经元的个数对压力识别的影响。更多还原

【Abstract】 Measuring and analyzing the cylinder pressure of the diesel engine is import to improve the diesel engine performance、 control the injecting timing 、 condition monitor and fault diagnosis. The ordinary library method to obtain the cylinder pressure is making a hole in the cylinder cover, measure directly using the pressure transducer. In this method the transducer endure high pressure and high temperature gas’ impact, it’s difficult to use in practice. The indirect recovery method using other easy testing signal may provide practical utility. On the basis of such background, this dissertation intends to study the vibration identification methods.1. At first test the fourth cylinder pressure signal、 vibration signal、 top dead center signal of the 4135 diesel engine. Then the characteristics of the excitation and vibration response of cylinder cover are analyzed. The cylinder cover system is regarded as a multi-input/single-output linear system, The experiment data are denoised by time statistic averaging, which could improve the signal noise ratio, then the measurement signal is split up into several sequences related to the combustion pressure using time domain windows and high frequency noise using a FIR low-pass filter, The transfer function of the system is used to recovery the cylinder pressure. Time statistic averaging and the right position of testing the vibration response signal is vital to improve the identification precision. Compare the raw pressure and the identification pressure at the position of eigenvalue, Demonstrate the applicability of transfer function method.2. Identify the cylinder pressure of the diesel engine using the cepstral analysis. And compare the raw pressure and the identification pressure at the position of eigenvalue which reflect the combustion performance, analyzing the characteristic of the cepstral method.3. The non-linear characteristic of the excitation and vibration response of cylinder cover are analyzed in this paper, expound a diesel engine cylinder pressure recovery method based on BP neural network. It was proposed to process the acceleration signal due to the combustion pressure with denosing method of timestatistic average, and remove high frequency noise using a lowpass filter, then train BP neural network using the adaptive grads algorithm, adaptive adjust the learning speed, improve the precision of the network, identify the cylinder pressure. Analyze the effect on cylinder pressure recovery caused by the middle layer neural network and the input layer neural network.更多还原