摘要
排气阀是柴油机的重要部件之一,其故障诊断一直受到研究者的关注,传统的学习机器在小样本学习时不具有良好的泛化能力,其现场效果与实验室精度差距较大。建立在统计学习理论基础之上的支持向量机具有和样本数相适应的最优泛化能力。利用支持向量机适合处理高维数据以及具有良好泛化能力的特点,建立了排气阀故障诊断模型,将排气阀振动信号经过小波包分解后提取的特征指标在小样本时进行支持向量机学习,通过不同核函数的支持向量机和其它智能方法准确率的比较证明:支持向量机较其它智能方法有较大的优越性;准确率对核函数有一定的敏感性;在常用的3种核函数中,线性核的诊断准确率达到了100%,是柴油机排气阀智能故障诊断支持向量机的最佳核函数。
Exhaust valve is one of the important parts of diesel engines, whose fault diagnosis attracts the researchers. However, the traditional intelligent methods lack better generation abilities, especially trained through a few samples. The infield-accuracy does not meet practical requirement. Support Vector Machine (SVM), which based on Statistic Learning Theory, has the adaptive generation ability. In this paper, the utilization of SVM' s with good generation ability was adopted, and SVM was trained with the Wavelet Packets decomposed coefficients as the input index. Comparing with accuracy of different kernel' s functions, the results showed that the SVM is superior to other intelligent fault diagnosis methods. The diagnosis accuracy is sensitive to the kernel function. The linear kernel' s accuracy is 100%, which is the best kernel for exhaust valve fault diagnosis of diesel engine among commonly used kernel SVMs.
出处
《内燃机学报》
EI
CAS
CSCD
北大核心
2006年第5期465-469,共5页
Transactions of Csice
关键词
支持向量机
排气阀
故障诊断
Support vector machine
Exhaust valve
Fault diagnosis