Abstract: During the process of new transmission products quality inspecting, the original fault feature vector dimension was too high to cause the detection in a low efficiency rate and low accuracy rate. To solve this problem, a feature extraction method based on order analysis theory together with a feature selection and classification method based on Genetic Algorithm-Back Propagation(GA-BP)algorithm were presented. The transmission　s order domain feature was extracted by using the order analysis theory, and it was composed with the domain's feature to form a feature vector set. The sum of the penalty factor and the distance ratio of inner-class and between-class was taken as the objective function value, the features of feature vector set were selected by Genetic Algorithm(GA), and the feature subset was obtained. Back-propagation neural network algorithmare was applied to classify the fault. The effectiveness of proposed algorithm was validated by experiments.

Key words: order analysis, feature extraction, feature selection, fault diagnosis, transmission

摘要： 针对汽车变速箱原始故障特征向量维数过高导致的检测效率低、准确率低的问题，提出一种基于阶次分析理论的特征提取方法和基于遗传算法—反向传播神经网络的特征选择与分类方法。首先运用阶次分析理论提取变速器的阶次域特征，与时域特征共同组成特征向量集；然后将类内类间距离比与惩罚系数之和作为目标函数值，利用遗传搜索策略对特征向量集进行特征选择，得到特征子集；最后用反向传播神经网络算法进行故障分类，得到检测结果，并通过实验验证了所提出方法的有效性。

关键词: 阶次分析, 特征提取, 特征选择, 故障诊断