At present, with the development of China's aerospace science and technology, especially the space development technology, the spacecraft space simulation experiment is gradually improving the requirements of experimental equipment. Ultra-low temperature and high vacuum are important test environments for the simulation experiments. Among them, the vacuum pump is one of the core equipment of the space simulator. Whether the vacuum pump can work normally will determine whether the space environment simulator can complete the vacuum thermal environment experiment of the spacecraft normally and effectively. Secondly, China has a large number of aerospace bases, as well as metallurgical industries, and the number of vacuum pumps is huge. Therefore, whether it is from the perspective of equipment safety or social and economic benefits, it is of great significance to detect the failure of the vacuum pump.

In the traditional mechanical fault diagnosis technology, Fourier transform is the most commonly used frequency domain signal processing method, but due to its own limitations, it is slightly weak in the face of nonlinear and time-frequency variation laws. The sampling step size of the wavelet transform changes with the frequency, which is consistent with the requirement that the high-frequency signal has high time resolution and the low-frequency signal has high frequency resolution. Therefore, it is more suitable for processing signals. Requirements for time domain and frequency domain.

Singular Value Decomposition (SVD) is a method for extracting signal features effectively. The singular values ​​obtained by SVD represent the intrinsic properties of the data, and its stability and invariance are better. Studies have shown that by performing signal reconstruction after SVD on the signal, the noise in the signal can be effectively removed, leaving useful information. By constructing the attractor trajectory matrix of the signal and performing SVD on it, by calculating the appropriate singular value for signal reconstruction, the random part of the signal can be eliminated, and the useful part of the signal is retained to the maximum to achieve signal denoising. .

Support Vector Machine (SVM) is a machine learning method widely used in pattern recognition. The basic theoretical principle is statistical theory. SVM has a strong advantage in dealing with high-dimensional, nonlinear, and small samples. Therefore, this paper selects SVM for fault mode identification.

In this paper, combined with SVD and wavelet packet transform, the fault feature extraction of the vacuum pump is realized, and the extracted feature vector is input into the SVM to realize the fault identification of the vacuum pump.

1 singular value decomposition (SVD)

1.1 SVD principle

For the acquired time signal x(n), the length is N, n=1, 2, 3, 4, ..., N, and the phase space is reconstructed, and the sampling interval is Ï„, then the reconstructed attractor The trajectory matrix A is [7]:

1.2 Research on Signal Denoising Method Based on SVD

For the acquired time series x(n), the wanted signal and noise are mixed. According to the study, if the signal is a smooth signal, then the rank r of the attractor trajectory matrix

Regarding the selection of the separation order k, in order to retain the useful signal as much as possible, the contribution rate of the singular value can be selected, and the contribution rate ρ is defined as follows:

It is generally believed that the useful information of the original signal can be basically retained when the contribution rate is greater than or equal to 0.9.

2 wavelet packet decomposition (WPD)

Compared with wavelet decomposition, the wavelet packet can further decompose the high frequency band decomposed by the upper layer, which can improve the time-frequency resolution of the signal and has higher application value [10].

The WPD algorithm is:

From the Parseval formula, the square of the wavelet packet coefficients Cj,k of x(n) has an energy dimension, so it is feasible to use the energy spectrum obtained by WPD to characterize the energy distribution of the signal.

3 experimental system and fault feature extraction

3.1 Collection of experimental data

The entire acquisition platform consists of a host computer, NI's acquisition card 6366, preamplifier and a sensor. The capture card has a sampling rate of up to 2 MS/s and supports 8-channel simultaneous acquisition. The sensor uses PAC's R3α with a center frequency of 29 kHz.

The experiment collected the vibration signal of the vacuum pump under normal operation and overload condition. The sampling rate was 100 kHz, and each group collected 5,000 points. 130 sets of data were collected, the first 60 sets were used as training samples of SVM, and the last 70 sets of data were used as check samples of SVM model. The experiment was carried out using a 110-molecular pump unit produced by Zhongke Science and Technology Co., Ltd., using R3α from PAC, and finally selecting the appropriate experimental samples for analysis.

3.2 Signal feature extraction

After the acquisition system, the vibration signal x(n) of the vacuum pump is obtained, and FIG. 1 is the original image of the collected overload signal.

According to the aforementioned, the original overload signal x(n) is subjected to singular value decomposition denoising. The autocorrelation function of x(n) is first calculated to obtain the delay step Ï„ of the attractor trajectory. After calculation, Ï„ is 6. The signal is subjected to singular value decomposition according to the determined delay step size, and the singular value decomposition is as shown in FIG. 2. The embedded dimension is chosen to be 200, and the length of x(n) is 5,000. Selecting the singular value according to the contribution rate, this paper retains 90% of the singular value. After calculation, for the test signal, the first 142 are reserved, and the last 58 are zeroed and the signal is reconstructed, so that the denoising is obtained. Test signal. The normal and overload signals of the degassed vacuum pump are subjected to 7-layer WPD using db11 wavelet, and the first 8 frequency bands of energy concentration are selected by the decomposition and reconstruction of the wavelet packet, as shown in FIG. 3 to FIG. Wherein, the ordinate represents the amplitude, and s70, s71...s77 represent the first, second, ..., 8th bands of the seventh layer, respectively.

For the 8 effective frequency bands obtained, find their energy separately:

In this way, an 8-dimensional vector [E0, E1, E2, E3, E4, E5, E6, E7] composed of band energy can be obtained, and the obtained wavelet packet energy spectrum is shown in Fig. 7.

4 pattern recognition

Support Vector Machine (SVM) was first proposed by Vapnik, and scholars often use it to solve linear regression and pattern recognition problems. The solution to the problem of SVM is to find an appropriate hyperplane as the classification surface, so that the isolation edge between the samples to be distinguished is maximized [12].

The test signal is first denoised by SVD, and then through 7 layers of WPD, an 8-dimensional vector composed of the energy of the 8th band of the 7th layer is obtained as an input of the SVM. The signal output under normal operation is 1 and the output under fault conditions is -1. The test results are shown in Figure 8.

It can be seen that the SVM corrects the fault and the normal signal with a correct rate of 98.57%. This shows that it is feasible to use the singular value denoising and the energy vector extracted by WPD as the characteristic information of the fault. The SVM is trained with the training samples, and then the test samples are used for testing. The obtained results are in agreement with the actual ones, so the fault identification with SVM has strong reliability.

5 Conclusion

This paper combines SVD, WPD and SVM to identify the fault of the vacuum pump. SVD can better remove unwanted noise in the signal, and then extract and extract the wavelet packet as the input vector of SVM. It has very high accuracy and can accurately and efficiently identify the fault of the vacuum pump. Therefore, the vacuum pump fault diagnosis method based on SVD, WPD and SVM is effective and feasible.

Valve-regulated Sealed Lead Acid Battery

Long Time Life Battery,Uninterruptible Power System,Thin Plate Technology Battery,Sealed Lead Acid Battery

Wolong Electric Group Zhejiang Dengta Power Source Co.,Ltd , https://www.wldtbattery.com