機床聲音信號特征的統(tǒng)計分析.pdf

1、南京航空航天大學碩士學位論文機床聲音信號特征的統(tǒng)計分析姓名：韓貞榮申請學位級別：碩士專業(yè)：機械制造及其自動化指導教師：左敦穩(wěn)20070601機床聲音信號特征的統(tǒng)計分析 II ABSTRACT In the paper, the feature of machine sound signal is analyzed and investigated by using the statistical method, and it is t

2、o get the feasible feature vectors. By collecting and preprocessing machine sound, extracting its feature, investigating the sound feature statistical distribution, some important results are obtained in initial stage.

3、 The main works include: 1. By generalizing the present state in the domestic and the foreign, the necessity of this investigation is purposed. And some normal methods about DSP and statistical are presented. 2. Th

4、e unload sounds of two types machines are collected by the sound signal collecting device. There are 1026 samples of the 38 models, 175 machines in plain and DNC lathe and in plain and DNC milling machines. And resampl

5、e is developed in the type of C618, C6132A1, N091, N092 and XS5040. 3. The features of the resample machine sound are abstracted and analyzed. It is shown that the single feature cannot describe the different machines i

6、n time field (the mean, variance and the maximum of autocorrelation) and frequency field (from the first main frequency to eighth and the first main hump to eighth), and the accurate recognition ratio is below 30%. While

7、 assembling the single feature effectively, the ratio can get above 90%. 4. The distribution rule of single feature in the same type machine is analyzed by the statistical method. It is found that the single feature of t

8、he different type machines has different distribution functions and parameters, and the single feature of the same type machines is approximate. 5. By the Factor Analysis Method, it is shown that the single feature has l

9、ittle contribution to the machine features from observing the Scree Plot, and they only show the features in certain aspect, such as the magnitude of the signal in time field and the frequency components and its weight

10、in frequency field. 6. The multi-feature vectors are analyzed by using the Fisher Discrimination Method and Three Layer BP Neural Network Technology. It is found that the recognition ratio of the machine type and the ro

11、tational speed is increased following the number of feature. And the type is recognized in 88.3% by assembling the time field and the 1st main frequency to 8th; the speed is in 92% by the whole 19 single feature vector

12、s. 7. For the need of the signal sampling and management, a simple feature base management system is developed by the Access software. Also, a visual interface is designed by the GUI tool of MATLAB, and its function is