【作者】 杨文宏；

【导师】 王珂；

【作者基本信息】 吉林大学 ， 通信与信息系统， 2010， 博士

【摘要】 爆震检测是汽车电子控制系统中的重要环节,准确的爆震检测和实时控制可以改善发动机的动力性和经济性,但是振动信号的低信噪比使得轻微爆震识别比较困难。本文对爆震信号处理方法和轻微爆震识别进行了系统的研究。在研究功率谱密度估计爆震特征频率和小波变换爆震分析基础上,进一步研究应用高阶累积量进行爆震信号处理的方法,提出了基于三阶累积量的振动信号降噪及爆震特征提取算法,以及基于峰度的爆震强度计算及分级方法,为进一步提升爆震探测识别技术奠定理论技术基础。功率谱密度反映了信号的功率随频率的变化情况,据此提出了利用参数化功率谱估计算法准确估计爆震特征频率,并以此为依据设计数字带通滤波器。实验分析表明在爆震较强时,滤波方法是一种效果良好且简单易行的爆震检测方法,但是当爆震较弱或背景噪声较强时,滤波方法仍然残留了大量的噪声,不利于准确检测轻微爆震和进行临界爆震分析。爆震一般持续时间较短,而小波变换良好的时频局部化分析特征,使其适合用于爆震信号处理,因此通过分量分析,研究利用离散小波变换提取爆震特征的有效方法,并采用两种爆震强度评价指标分析比较了小波变换方法和滤波方法的性能。研究表明,如果从信号的峰值幅值来评价爆震,对于弱爆震检测,二者的性能基本相同,但是在中轻度爆震检测方面,小波变换方法优于滤波方法。为了提高轻微爆震检测的精度,本文提出了一种基于三阶累积量的振动信号降噪方法。该方法利用了三阶累积量抑制对称噪声的性质,结合检测函数和合理的阈值,能有效地消除振动噪声,并提取爆震特征。模拟实验分析和实测振动信号分析均表明,与其它方法相比,该方法明显地提升了信噪比,更适用于轻微爆震识别。针对爆震实时控制的需要,进一步研究了振动信号的四阶累积量特征,分析信号的峰度与爆震强度的相关性,并提出了基于峰度的爆震强度检测算法,定量地计算爆震强度。结果表明,该信号处理算法计算量小,简便快捷,在此基础上进一步研究了爆震强度分级方法,将更加有效提高爆震控制效率。更多还原

【Abstract】 Knock control is one of important parts for automobile electronic control system. Serious engine knock decreases power performance of the vehicle, increases pollution or damages mechanical parts. However, when the engine works with light knock, engine power and fuel economy will be improved. Therefore, it is important to detect knock and provide feedback information to control system, so that the engine can work with light knock. In this process, the accurate light knock identify is a necessary pre-condition.The mostly used methods for knock detection use accelerometers to measure cylinder vibration, since the accelerometer is costless and the measure system is easy to maintain. But the vibration signal is induced by engine knock indirectly, and the drawback is that the low signal-to-noise ratio of the vibration signal makes the light knock detection difficult. Therefore, this paper researched knock signal processing methods, proposed some new knock feature extraction, knock intensity determine and light knock detection methods. The simulated experiement and real vibration signals processing results verified the validity of these methods.Vibration signals were obtained with an accelerometer mounted on the engine cylinder head. Knock was induced through changing ignition advance angle. A lot of signals with different working conditions were recorded for analyzed.Fourier spectrum of the vibration signal was analyzed, the results indicated that knock lead to larger amplitude in a special band, which includes the knock characteristic frequency. But this is only a rough evaluation for engine knock with influence of noise, it cann’t be used to detect knock intensity accurately. In order to find out exact knock frequency, a method using power spectrum estimation was proposed in this paper. AR model was used to describe typical knock process, and model parameters were estimated with Burg and covariance algorithms. The experiment result showed that knock characteristic frequency can be estimated, but the results were different for different vibration signals, it illustrated that knock frequency was affected by the engine working condition. For the engine used in the experiment system, variations in the fundamental knock frequency can be as much as±600Hz.Through power spectrum estimation results of many signals with different engine working conditions, a band range of knock frequency was obtained. The band-pass filter was used to extract knock information. According to power spectrum estimation results, the center frequency and the cut-off frequency were decided. Because the IIR filter involves few memory units and it has high efficiency, so an IIR filter was chosen. In order to avoide distortion of knock component, Butterworth filter was applied, since the amplitude response within the appointed band is flat.The advantage of the filtering method is that parameters of the programmable filter can be adjusted conveniently when used for different engines. The results of real signals processing indicate that the performance of filtering method is good when knock is strong, however, when knock is light or noise is strong, the performance decrease, so filtering method is not sufficient to detect light knock accurately.Wavelet transform is an effective tool to detect abnormal signals because of its multi-resolution. Wavelet transform is suitable to identify knock, since knock always cause amplitude increasing abruptly. Application of wavelet transform in knock feature extraction was researched in this paper. When applied the mother wavelet function, because the duration of knock is short, wavelet functions with compact supports is beneficial to knock detection. Discrete wavelet transform was applied to analyze signals with different knock intensity. The experiment result indicated that the knock feature included in the detail component d 2 was obvious for strong knock, while the knock feature included in the detail component d 3 was obvious for light knock. Actually, the probability of engine working with strong knock is low, it means more to identify light knock, so the detail component d 3 should be used for further research.In order to compare filtering method and wavelet transform method, two kinds of knock intensity criteria were chosen. Calculated the peak and root mean square of the signal processed with filtering method and wavelet transform method. The experiment results indicate that the peak and root mean square have correlation when used for knock evaluation. When knock is light, the performance of them is the same. But if knock is strong, the root mean square of knock signal is more sensitive than the peak, because both of amplitude and lasting time increase with strong engine knock. Compared with two methods, if the criterion of root mean square was applied, two methods performed alike. But when the criterion of peak was applied, wavelet transform method is better than filtering method for moderate knock. However, the experiment result indicated that wavelet transform method didn’t act better than filtering method for light knock detection.In order to identify light knock accurately, knock signal processing method with higher order cumulants was proposed in this paper. The skewness of the vibration noise was analyzed, and the result showed that the skewnee was close to be zero, so the probability distribution near to be symmetrical. Because third-order cumulants are effective to remove noise with symmetrical distribution, a vibration signal denoising method with third-order cumulants was proposed. Based on the detection function proposed, the method can enhance the impact of the knock signal components by setting a reasonable threshold and suppress vibration noise.In order to select the appropriate threshold, analyzed the relationship between sets of knock signal and non-knock signal’s detection function value and the threshold. If the threshold is too small, more noise will be kept down; if the threshold value increases, although the noise is removed much more, some of useful sample data also be discarded because of the smaller amplitude, which corresponds the end part of the knock process. Balance these two factors, this paper choose a suitable threshold, which retain more than 90% of the knock-component, while more than 90% of the noise will be removed.Applied third-order cumulants based method to extract knock feature from real vibration signals, and compared it with filtering method, advantages of the proposed method were explained with two important aspects. Firstly, third-order cumulants based method can remove much more background noise, improve signal-to-noise ratio, so the character of the entire knock process was observed clearly. Secondly, the estimated knock energy in time domain was closed to the actual value. When the signal-to-noise ratio is low, the improvement is more obvious. The real vibration signals processing results also confirmed that knock feature extracted using third-order cumulants based method is better, which is also helpful to improve the performance of light knock detection.A knock detection method with less computation was proposed in this paper, so as to fulfill the real time knock control. Fourth-order cumulants of the vibration signals were analyzed, and the results indicate that probability distribution of the vibration signal is not Gaussian, but the super-Gaussian. With the determined engine operating condition, kurtosis of the vibration signal has special correlation with knock intensity. As the kurtosis contains the information of current knock intensity, this paper proposes a knock intensity determination method using kurtosis. Because kurtosis is the special diagonal slice of fourth-order cumulants, computation involved in the approach is simple. The experiment confirmed the method was effective, and it can provide numerical values for variety of knock intensity.After the knock intensity was calculated numerically, a method was proposed to classify a variety of knock conditions. With a determined engine rotation speed, six groups of vibration signals were collected when ignition advance angle was increased with the step of 2 o CA, and each group contains at least 100 cycles. The first group of signals corresponds to knock-free status; while the final group of signals corresponds to serious knock. Calculated knock intensity for each group of signals using the proposed method in this paper, and analyzed how the numerical values were affected by ignition advance angle. According to the experiment result, a series of thresholds were set up, in order to divide knock conditions into four grades: non-knock, light knock, moderate knock and severe knock. The ignition advance angle adjusting strategy can vary with different kinds of knock grades; so the method is helpful to improve the efficiency of the knock control system.更多还原