Abstract: In order to extract entire status features with high quality from the vibration signal on diesel engine cylinder head the forecasting algorithm based on kalman filtering is proposed to analyze time history of the signal.. A characteristic subset is made up by decomposition of singular value of transfer matrix through Kalman transfer matrix including entire engine state information. Considering the distribution at different working condition, the samples of a characteristic subset is classified and tested by use of extreme learning machine. It is shown by practical application that this algorithm offers higher precision for engine fault diagnosis.

Key words: vibration and wave ; , cylinder head vibration signal , feature extraction , kalman transfer matrix , SVD

摘要： 为了从发动机缸盖振动信号中提取出全面的、高质量的状态特征，建立缸盖振动信号的时变参数模型，提出缸盖振动信号Kalman滤波预测算法，通过引入包含发动机状态信息的Kalman转移矩阵，得到转移矩阵的奇异值分布，构成特征子集。研究不同工况下特征子集的分布，选用极限学习机对特征样本进行分类和测试，实际应用结果表明，发动机故障诊断精度较高。

关键词: 振动与波, 缸盖振动信号, 特征提取, Kalman转移矩阵, 奇异值分解