CN107956708A - A kind of potential cavitation fault detection method of pump based on quick spectrum kurtosis analysis - Google Patents

Abstract

The invention discloses a pump potential cavitation fault detection method based on fast spectral kurtosis analysis, which includes: step 1, collecting vibration acceleration signals for noise reduction, as signals to be processed; step 2, determining according to the data volume of the signals Decomposition order of signal processing; step 3, select the optimal carrier frequency and bandwidth according to the calculation results of the fast spectral kurtosis algorithm; step 4, perform Fourier transform on the selected carrier frequency and the signal within the bandwidth to obtain the spectrum package In step 5, compare the original signal time-domain diagram, the signal time-domain diagram processed by fast spectral kurtosis filtering, and the spectrum envelope diagram after Fourier transform of the selected area, and analyze the time and frequency characteristics of the cavitation fault signal . The method of the invention can detect more cavitation instantaneous signals, make the information in the time domain and frequency domain more clearly, and can clearly distinguish the normal state and the cavitation state of the pump.

Description

A pump potential cavitation fault detection method based on fast spectral kurtosis analysis

technical field

The invention belongs to the field of signal processing, in particular to a method for analyzing the real-time state of a pump and detecting its potential cavitation fault based on fast spectrum kurtosis.

Background technique

High-performance centrifugal pumps are widely used and in great demand in today's society. Due to working under complex conditions such as high pressure and high speed, cavitation failures of centrifugal pumps occur frequently, resulting in increased vibration frequency, increased noise, and blade corrosion, which seriously restricts pump performance and life. The traditional detection method is not sensitive to the detection of signal changes such as pump flow, head, vibration and noise in the initial stage of pump cavitation; but when the cavitation signal changes significantly, the cavitation fault has rapidly developed to a quite serious level.

The acoustic signal of cavitation bubbles has a large bandwidth span, strong instantaneousness, and high difficulty in processing; the vibration signal caused by cavitation is often strongly modulated by the blade rotation.

At present, the fault signal detection methods commonly used in the field of signal processing mainly include short-time Fourier transform and wavelet transform. The short-time Fourier transform is the most commonly used time-frequency analysis method, which represents the signal characteristics at a certain moment through a section of the signal in the time window. In the short-time Fourier transform process, the length of the window determines the time resolution and frequency resolution of the spectrogram. The longer the window length, the longer the intercepted signal, and the longer the signal, the higher the frequency resolution after Fourier transform. , the worse the time resolution; on the contrary, the shorter the window length, the shorter the intercepted signal, the worse the frequency resolution, the better the time resolution, that is to say, in the short-time Fourier transform, the time resolution and frequency resolution Rates can not have both, should be based on specific needs to choose. The short-time Fourier transform is severely affected by the time domain and frequency domain resolution, which limits its role. Moreover, for the vibration acceleration signal generated by cavitation, the short-time Fourier transform cannot analyze clear information.

The practicability of the wavelet transform is obviously stronger than that of the short-time Fourier transform. It inherits and develops the idea of localization of the short-time Fourier transform. -Frequency" window is an ideal tool for signal time-frequency analysis and processing. Discrete wavelet transform is often used in industrial production practice. However, there are still some disadvantages that the selection of wavelet basis is not unique and the combination of wavelet parameters is not robust. At the same time, there is also a combination of support vector machine and artificial neural network method to improve wavelet decomposition transformation, using pressure fluctuation signal and cavitation field distribution image, it is very effective for transient cavitation feature extraction, but the algorithm complexity is high and the parameter setting is still Empirical interpretation is required. In addition, the cavitation diagnosis method based on multi-dimensional analysis (impeller-guide vane-load-acoustic signal spectrum) is complementary to multi-scale analysis such as wavelet transform, and has successfully developed a high-sensitivity and high-reliability hydraulic turbine cavitation Monitoring and diagnostic systems, but this multi-dimensional analysis method fails to obtain high-resolution dynamic spectrum texture, which is not conducive to characterizing key transition processes from sheet to cloud cavitation.

Contents of the invention

The invention provides a pump potential cavitation fault detection method based on fast spectral kurtosis analysis, which can detect more instantaneous signals, make the information in the time domain and frequency domain clearer, and can clearly distinguish the pump The normal state and cavitation state not only have a relatively large frequency detection range, but also are easy to operate.

A pump potential cavitation fault detection method based on fast spectral kurtosis analysis, comprising the following steps:

Step 1, noise reduction is performed on the collected vibration acceleration signal as the signal to be processed in the experiment;

Step 2, according to the data size of the signal to be processed in the test, determine the decomposition order of the fast spectral kurtosis algorithm with the highest degree of fitting as the decomposition order of the signal processing;

Step 3, according to the calculation result of the fast spectral kurtosis algorithm, select the carrier frequency and the corresponding bandwidth that can maximize the signal kurtosis;

Step 4, performing Fourier transform on the selected carrier frequency and the signal after the corresponding bandwidth to obtain the spectrum envelope;

Step 5: Analyze the time and frequency characteristics of the fault signal according to the time domain diagram of the original signal, the time domain diagram of the signal processed by fast spectral kurtosis, and the spectrum envelope diagram after Fourier transform of the selected area.

In step 1, the described denoising method is, in the processing program, use the pre-whitening processing noise to carry out denoising processing on the signal, and the pre-whitening processing is in the MATLAB software:

x=x-mean(x);

Na=100;

a=lpc(x,Na);

x = fftfilt(a,x);

x=x(Na+1:end);

where x is the processed signal.

The specific process of step 2 is:

Step 2-1, in MATLAB software, according to the actual data volume, set a preliminary decomposition order;

Step 2-2, at this order, observe the carrier frequency and bandwidth obtained by the fast spectral kurtosis algorithm, and observe the spectrum envelope after Fourier transform in this frequency range;

In step 2-3, the decomposition order is determined according to the peak characteristics of the spectrum envelope diagram.

The principle of the decomposition order is to adjust according to the peak density of the spectral envelope diagram of the processing result. If the peak density is too low, the decomposition order will be reduced; otherwise, it will be increased.

In step 3, the fast spectral kurtosis algorithm in MATLAB software is a function with the original signal, decomposition order (nlevel) and sampling frequency as independent variables.

In step 5, the process of analyzing the time and frequency characteristics of the fault signal is specifically:

Step 5-1, determine whether there is a cavitation fault according to whether there is an obvious shock signal on the time domain diagram of the signal processed by the fast spectral kurtosis, and determine the time of the cavitation fault signal according to the position of the shock signal on the time domain diagram;

Step 5-2: Determine the frequency characteristics of the cavitation fault signal according to the shaft frequency and leaf frequency information on the frequency envelope after Fourier transform.

The invention provides a fast spectral kurtosis spectrum texture analysis method, which processes the vibration signal of the pump through the fast spectral kurtosis function. The present invention selects the optimal carrier frequency and bandwidth that can contain the most instantaneous information for signal filtering, obtains frequency domain information by Fourier transforming the time domain information, and then detects the state of the pump, and detects the specific cavitation Analysis of failure frequency.

The method of the invention greatly improves the signal enhancement capability, can enhance the cavitation signal from the rotating blade frequency, and can clearly distinguish the relevant data of the pump, and also has an obvious distinction between the normal state and the cavitation state.

Description of drawings

Fig. 1 is the schematic flow chart of the method for pump real-time state monitoring and potential cavitation fault detection based on fast spectrum kurtosis spectrum texture analysis in the present invention;

Fig. 2 is a schematic diagram of the analysis and processing results under the rated state using fast spectral kurtosis;

Figure 3a is the time domain diagram of the original signal in the rated state;

Figure 3b is a time-domain diagram of the signal processed by the fast spectral kurtosis filter in the rated state;

Figure 3c is the spectrum envelope diagram after Fourier transform of the selected area in the rated state;

Fig. 4 is a schematic diagram of the analysis and processing results of the pump cavitation state by using fast spectral kurtosis;

Figure 5a is the time domain diagram of the original signal in the pump cavitation state;

Figure 5b is a time-domain diagram of the signal processed by fast spectral kurtosis filtering in the pump cavitation state;

Fig. 5c is the spectrum envelope diagram after Fourier transform of the selected region under the cavitation state of the pump.

Detailed ways

Fast Spectral Kurtosis is a fourth-order spectral analysis tool. defined as where H(n,f) is the complex envelope of the signal x(n) at frequency f. <> is the operator for averaging. Fast spectral kurtosis can analyze unsteady processes very well, such as transient signals, and the kurtosis value of highly unsteady transient signals depends on the frequency resolution (Δf) of the estimator, and each transient phenomenon corresponds to a An optimal frequency band {f, Δf}. Therefore, in the actual analysis process, the optimal frequency and frequency resolution information should be found, so that within this interval, the kurtosis reaches the maximum value, that is, the relevant transient information can be found.

Under many conditions of the pump, such as cavitation and vane deformation, etc., the vibration of the pump will change suddenly, thus generating a large amount of transient information. In this way, the fast spectral kurtosis algorithm provides a good tool for detecting and diagnosing pump cavitation faults due to its good detection of transient information and good ability to resist noise.

In order to describe the present invention more specifically, the technical solution of the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

As shown in Figure 1, the pump potential fault detection method based on fast spectral kurtosis analysis includes the following steps:

S01, respectively collect the vibration signals of the pump with normal load and the pump with cavitation phenomenon through the vibration acceleration sensor, and import the data into the processing program.

In the processing program, use the method of pre-whitening to deal with the noise, and perform noise reduction on the signal. In MATLAB software, the pre-whitening statement is:

x=x-mean(x);

Na=100;

a=lpc(x,Na);

x = fftfilt(a,x);

x=x(Na+1:end);

where x is the processed signal.

S02. Calculate the signal obtained by the noise reduction process using a fast spectral kurtosis function, where the fast spectral kurtosis function is a function that takes the original signal, decomposition order (nlevel) and sampling frequency as independent variables.

According to the size of the data volume, select an appropriate calculation decomposition order. Here, the principle of the decomposition order is determined according to the peak density of the spectrum envelope diagram of the processing result. If the peak density is too low, the decomposition order is reduced; otherwise, it is increased. In this test data, the test collection frequency is 40960Hz, and the test data is basically between 640,000 and 650,000, so the sixth-order analysis is adopted.

S03, find the frequency with the largest spectral kurtosis and the corresponding frequency bandwidth in the analysis result graph of the fast spectral kurtosis. The results of fast spectral kurtosis analysis under normal load conditions are shown in Figure 2. The optimal carrier frequency is 2400 Hz, and the frequency bandwidth is the quotient of the sampling frequency and the order of 2 plus the first power, so the frequency bandwidth is 320 Hz. The results of fast spectral kurtosis analysis in the cavitation state are shown in Figure 4, the optimal carrier frequency is 19626.6667Hz, and the frequency bandwidth is 1706.6667.

S04, transform and analyze the signal with the largest spectral kurtosis in the time domain and frequency domain, use Fourier transform to obtain the envelope diagram of the frequency, and then detect and diagnose the state of the pump by analyzing the frequency. Among them, Fig. 3a, Fig. 3b and Fig. 3c are respectively the original signal time domain diagram under normal load state, the signal time domain diagram after fast spectral kurtosis filtering and the spectrum envelope diagram after Fourier transform; Fig. 5a, Fig. 5b and 5c are the time-domain diagram of the original signal in the cavitation state, the time-domain diagram of the signal processed by fast spectral kurtosis filtering, and the spectrum envelope diagram after Fourier transform, respectively.

S05, using the processing result graph to analyze and compare the information of the pump under the normal load state and the cavitation state. The relevant information of the pump cannot be analyzed from the original signal, as shown in Figure 3a and Figure 5a. It can be found that in the vibration signal of the pump under normal conditions, the data processed by the fast spectral kurtosis filter is still relatively flat in the time domain, and no obvious impact can be seen in the time domain, as shown in Figure 3b. In the spectrum envelope after Fourier transform, as shown in Fig. 3c, the shaft frequency (24.61 Hz) and the related leaf frequency harmonic information (75 Hz, 100 Hz, etc.) can be clearly found. In this test, the speed of the pump is 1375 revolutions per minute, so its shaft frequency is 25Hz. In the cavitation state, after fast spectral kurtosis filtering, the shock signal can be clearly found in the time domain, as shown in Figure 5b. On the frequency envelope diagram after Fourier transform, a large amount of high-frequency information appears, and its range even reaches above 1600Hz, and the leaf frequency signal (172Hz) is particularly prominent, indicating that the air bubbles produced by the pump cavitation phenomenon affect the leaves. The frequencies are superimposed, and the bubbles generated by cavitation impact the blade with high frequency, as shown in Figure 5c. Therefore, it is confirmed that the spectral kurtosis analysis can detect and analyze the cavitation state of the pump.

This example uses the vibration acceleration data of the pump. In the rated normal state, there is obvious frequency information, but in the cavitation state, there is an obvious impact signal at a specific time point. On the frequency spectrum, the cavitation can also be found. There is a large amount of high-frequency information, which shows that the fast spectral kurtosis algorithm has a good effect on the vibration state monitoring and fault analysis of the pump.

The above-mentioned specific embodiments have described the technical solutions and beneficial effects of the present invention in detail. It should be understood that the above-mentioned are only the most preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, supplements and equivalent replacements made within the scope shall be included in the protection scope of the present invention.

Claims (6)

1. A pump potential cavitation fault detection method based on fast spectral kurtosis analysis, including: Step 1, noise reduction is performed on the collected vibration acceleration signal as the signal to be processed in the test; Step 2, according to the experimental signal to be processed, determine the decomposition order of the fast spectral kurtosis algorithm with the highest fitting degree as the decomposition order of signal processing; Step 3, according to the calculation result of the fast spectral kurtosis algorithm, select the carrier frequency and the corresponding bandwidth that can maximize the signal kurtosis; Step 4, performing Fourier transform on the selected carrier frequency and the signal after the corresponding bandwidth to obtain the spectrum envelope; Step 5: Analyze the time and frequency characteristics of the fault signal according to the time domain diagram of the original signal, the time domain diagram of the signal processed by fast spectral kurtosis, and the spectrum envelope diagram after Fourier transform of the selected area. 2. The pump potential cavitation fault detection method based on fast spectral kurtosis analysis according to claim 1, characterized in that, in step 1, the noise reduction method is, in the processing program, using pre-whitening to process noise , to denoise the signal. 3. the pump potential cavitation fault detection method based on fast spectral kurtosis analysis according to claim 1, is characterized in that, the specific process of step 2 is: Step 2-1, in MATLAB software, set a preliminary decomposition order according to the data size of the test signal to be processed; Step 2-2, at this order, observe the carrier frequency and bandwidth obtained by the fast spectral kurtosis algorithm, and observe the spectrum envelope after Fourier transform in this frequency range; In step 2-3, the decomposition order is determined according to the characteristics of the spectrum envelope diagram. 4. The pump potential cavitation fault detection method based on fast spectral kurtosis analysis according to claim 1 or 3, characterized in that, in step 2, the determination principle of the decomposition order is based on the data volume of the test signal to be processed and Processing results in spectral envelope peaks to adjust for density. 5. the pump potential cavitation fault detection method based on fast spectral kurtosis analysis according to claim 1, it is characterized in that, in step 3, described fast spectral kurtosis algorithm is one with original signal, decomposition in MATLAB The order and the sampling frequency are functions of the independent variables. 6. The pump potential cavitation fault detection method based on fast spectral kurtosis analysis according to claim 1, characterized in that, in step (5), the process of analyzing the time and frequency characteristics of the fault signal is specifically, Step 5-1, according to whether there is an obvious cavitation shock signal on the time domain diagram of the signal processed by the fast spectral kurtosis, determine whether there is a cavitation fault, and determine the location of the cavitation fault signal according to the position of the shock signal on the time domain diagram time; Step 5-2: Determine the frequency characteristics of the cavitation fault signal according to the shaft frequency and leaf frequency information on the frequency envelope after Fourier transform.