摘要
在不同的燃烧状况下同时测量缸盖表面振动信号和缸内压力信号,对平均处理后的信号进行频域分析,发现缸内压力信号中相对于基频前50阶的谐波分量包含了所关心的主要信息。根据频域分析得到的复数谱的对称特性建立了训练样本,并对建立的BP和RBF神经网络进行训练。训练的结果表明RBF神经网络可以在更短的训练时间内,获得更小的均方误差。利用不同的神经网络进行了缸内压力信号的识别,识别的结果表明,RBF神经网络识别的精度高于BP神经网络。
Cylinder pressure and vibration signal are measured when a diesel engine runs under different conditions.Frequency domain analysis of the averaged signal shows that the cylinder pressure signal can contain the primary interesting information only if it preserves the first 50 orders harmonic components of the crankshaft rotation.Trained samples are established according to the symmetric feature of the complex spectrum obtained from frequency domain analysis and they are used to train the BP and RBF neural networks.Training results show that the RBF neural network can receive smaller mean square error(MSE) within shorter time.The cylinder pressure is recognized using different neural networks.The recognition results show that the RBF neural network has better precision than that of the BP one.
出处
《振动与冲击》
EI
CSCD
北大核心
2008年第2期133-136,共4页
Journal of Vibration and Shock
关键词
缸内压力
振动信号
频域特征
神经网络
cylinder pressure,vibration signal,frequency charateristic,neural network