单独提取滚动轴承振动信号的时域或频域特征进行故障诊断，是目前常用的轴承诊断方法，诊断精度有待提高。以时域和频域的多维振动特征参量为指标，以历史诊断正确率作为特征参量权值，分别对滚动轴承的无故障和经常出现的滚珠故障、内环故障和外环故障工况进行特征提取和故障识别。多维时频域振动特征是单维特征依据诊断精度权重的集合。运用BP神经网络分别对信号的时域特征（TDF）、IMF能量矩（IEM）、小波包能量矩（WPEM），以及多维时频域特征进行智能故障判别。实验验证用多维时频域振动特征参量综合诊断的方法进行滚动轴承故障诊断，比单维特征的诊断结果精确且效率较高，该方法可以在滚动轴承故障诊断领域展开应用。

Extracting the time-domain or the frequency-domain features of vibration signals for analysis is a conventional method for rolling bearings fault diagnosis. But the effects of this diagnosis method need to be improved. In this paper, taking the multi-dimensional vibration characteristic parameters in time-domain and frequency-domain as the indexes and the correctness rate of historical diagnosis as the parametric weight, the features of fault-free rolling bearings and the features of rolling bearings with ball fault, inner and outer race faults are extracted and the faults are identified. It shows that the multi-dimensional vibration characteristic in time-frequency domains is the assemblage of single features. BP neural network is used for intelligent fault classification of signals according to the time-domain feature (TDF) parameters, IMF energy moment (IEM), wavelet package energy moment (WPEM) and multi-dimensional features respectively. Results of the diagnoses are compared one a

先前研究者普遍认为,类别推理学习条件下可以同时表征诊断性信息和非诊断性信息,而类别分类学习条件下中只能表征诊断性信息,不能表征非诊断性信息。而最近又有研究者发现部分呈现条件下的类别分类学习可以表征非诊断性信息。本研究通过两个实验系统比较了全部呈现和部分呈现条件下类别分类学习的结果,进一步探讨了分类学习条件下信息的表征情况,并进一步探讨了部分呈现条件下的分类学习能够表征非诊断性信息的原因。实验1发现全部呈现6个特征、缺失1个特征(即部分呈现5个特征)、缺失2个特征(即部分呈现4个特征)3种条件下都能表征诊断性信息,但只有部分呈现条件下能表征非诊断性信息。实验2发现全部呈现7个特征、缺失2个特征(即部分呈现5个特征)、全部呈现5个特征3种条件下都能表征诊断性信息,但只有部分呈现条件下能表征非诊断性信息。总的实验结果表明:全部呈现条件下的分类学习只能表征诊断性信息,而部分呈现条件下的分类学习能够同时表征诊断性信息和非诊断性信息,并且部分呈现条件下表征非诊断性信息的原因是被试进行了推理学习,而非注意广度的变化。

Previous researches have showed that category learning by inference way can represent diagnostic information and nondiagnostic information, but learning by classifying way only can represent diagnostic information such as exemplar features information. However, recently studies show that learning partial exemplars by classifying also can represent nondiagnostic information ( Taylor & Ross, 2009). Taylor & Ross (2009) offered an explanation of selective attention that there are comparably loose of attention resources in partial condition than entire condition. They left out the possibility that the subject might inference the missing features in the partial condition. In the real world, exemplars often appear with occluded features, but in laboratory research, they are almost always presented in their entirety. Two experiments were conducted to explore how partial classification leads to nondiagnostic features learning. Experiment 1 replicated the Taylor & Ross (2009) finding that learn

为了提高汽轮机故障诊断的精确性，文章运用转子振动实验台来模拟汽轮机转子的振动信号，对运行中的三种故障振动信号进行采集，然后运用局部特征尺度分解方法对汽轮机振动信号时间序列进行特征提取，组成特征向量。利用极限学习机作为故障诊断分类器，结果表明，局部特征尺度分解特征提取和极限学习机的诊断模型能够准确地对汽轮机故障进行诊断，具有很高的实际应用意义。

In order to improve the accuracy of fault diagnosis of steam turbine, this paper was used rotor vibration test bench to simulate the vibration signals of the steam turbine rotor, the operation of three kinds of fault vibration signal acquisition, and then using local characteristic scale decomposition approach to feature extraction steam turbine vibration signal time series of feature vector. Extreme learning machine was taken as a classifier of fault diagnosis. The results showed that the local characteristics of the scale decomposition of feature extraction and extreme learning machine diagnosis model can accurately the fault diagnosis for steam turbine, which had high practical application significance.

特征信息提取作为电力电子电路故障诊断的重要环节，直接影响到诊断结果的有效性。阐述了目前常用的各种故障诊断提取方法，结合实际应用对各个方法进行了优缺点评价。在此基础上，提出了未来特征提取方法的新思路，即构造能够发挥各自优点，实现功能互补的特征提取方法，为进一步提高故障诊断的准确性提供参考。

Feature information extraction, which directly influences the validity of diagnosis result, is an important link in power electronic circuit fault diagnosis. This paper summarized some methods of fault feature extraction in common use and evaluated the advantages and disad-vantages of each method with the combination of practical application. On the basis of this, this paper proposed a new idea for future feature ex-traction methods, which constructed the feature extraction method to bring each advantage into play and to realize the complementary function, to provide reference for further improving the accuracy of fault diagnosis.

为了提高模拟电路故障诊断准确率,提出一种联合选择特征选和分类器参数模型的模拟电路故障诊断方法 (Feature-Classifier)。将模拟电路故障特征子集和分类器参数编码成为粒子,然后粒子根据目标函数通过信息交流和互相协作找到最优特征子集和分类器参数,并根据最优特征子集对样本进行约简;分类器根据最优参数对约简后样本进行训练建立模拟电路故障诊断模型,并通过仿真实例对性能进行测试。结果表明,相对于其他模拟电路故障诊断方法,Feature-Classifier能够较快找到最优特征子集与分类器参数,不仅提高了模拟电路故障诊断准确率,并加快了故障诊断速度。

In order to improve fault diagnosis rate of analog circuits, this paper proposes a fault diagnosis method based on jointly selection features and classifier parameters model. The features and classifier parameters are encoded as a particle, and then, the optimal features and classifier parameters are obtained by the particle swarm optimization algorithm according to objection function, and the samples are reduced on the optimal features. Finally, the samples are input into the classifier to train and build the fault diagnosis model of analog circuits with the optimal parameters, and the simulation experiments are carried out to test the performance of the model. The results show that the proposed method can select the optimal features and classifier parameters quickly to improve the fault diagnosis rate of analog circuits and fasten the speed of the fault diagnosis compared with other methods.