中文摘要：

为了从发动机缸盖振动信号中提取出完备的、高质量的状态特征，并选择出最优特征子集进行分类，建立了缸盖振动信号集成特征提取模型，提出了一种基于样本分散度的最优特征子集选择算法。集成特征提取模型选取多个完整工作循环数据处理，用提升小波包对其进行快速变换，消除波动影响和噪声，求取所构造特征集的各特征值，得到包含完备发动机状态信息的特征向量；最优特征子集优化选择算法，建立了基于样本分散度的特征选择模型，解决了冗余分类信息的消除问题，结合分类器选择出最优特征子集，使其规模与分类效果综合最优，用欧氏距离分类法和支持向量基分类器进行测试，所有40个特征输入分类器正确率分别为67．86％和70％，优化选择后特征个数分别降低为6个和5个，而分类正确率提高到了90．71％和90％。

英文摘要：

In order to extract self-contained status features with high quality from the vibration signal of an engine cylinder head and select an optimum feature subset as the classifier, an integrated feature extraction model and an optimum feature subset selection arithmetic based on sample decentralization degree are proposed. In the integrated feature extraction model, data in a series of work cycles are processed; the lifting wavelet transform with fast speed is adopted; the fluctuation and noise are eliminated; feature values of every feature in a eonstructed feature set are computed; and a feature vector including general status information is extracted. In the optimum feature subset selection arithmetic, a feature selection model based on sample decentralization degree is presented; and together with the classifier, an optimum feature subset cousidering both size and classification effect is selected. Euelid distance and support vector machine classifier is used in testing phase. For 40 feature classifiers, the success rate of elassification for Euclid distance is 67.86%, and that for support vector machine is 70%. The number of features used in classfieation is reduced to 6 or 5 after optimization. When 6 features are used, the success rate is 90.71% ; when 5 features are used, it is 90%.