A Fault Diagnosis Method for Mine Drainage Pumps

Abstract

The present invention provides a method for fault diagnosis of mine drainage pumps, including collecting vibration signals, motor temperature, drainage port pressure, and suction port pressure of mine drainage pumps under faults, performing noise reduction processing, extracting characteristic values, training, testing, and optimizing the RBFNN; finally, substituting the real-time vibration, motor temperature, drainage port pressure, and suction port pressure data of the mine drainage pump to be diagnosed into the optimized RBFNN to obtain the real-time fault diagnosis result of the mine drainage pump. The trained RBFNN of the present invention can accurately diagnose rotor imbalance faults, cavitation faults, impeller faults, foundation loosening faults, inner ring faults of bearings, bearing rotor faults, outer ring faults of bearings, and comprehensive bearing faults of mine drainage pumps according to the real-time vibration, motor temperature, drainage port pressure, and suction port pressure data collected in real time, so it has the beneficial effects of being real-time and accurate.

Description

Technical Field

[0001] The present invention relates to the field of fault diagnosis of mine drainage pumps, and specifically to a method for fault diagnosis of mine drainage pumps. Background Technique

[0002] Mine drainage pumps are an important part of coal mine safety production, and they shoulder the heavy responsibility of discharging water inflow and working water to the ground. Therefore, it is necessary to diagnose common faults of drainage equipment to guide maintenance personnel to quickly repair faulty components and avoid accidents. However, there are many reasons for inducing faults in mine drainage pumps, and the fault manifestations are also diverse. It is a strongly non-linear complex system, which makes it difficult to accurately diagnose the faults of drainage pumps. Therefore, domestic and foreign scholars have conducted in-depth research on the theoretical methods and technologies for fault diagnosis of mine drainage pumps.

[0003] Patent CN202010677913.2 provides a method for fault diagnosis of water pumps based on multi-source data fusion analysis. This method uses the principle of normal distribution and the Rada criterion to statistically analyze the state data during the normal operation period of the pump equipment to obtain the actual thresholds of each state variable, and then normalizes the monitored state variables of the pump equipment according to the thresholds, and uses weighted fusion to obtain the health index of the water pump. The state of the pump equipment is pre-alarmed according to the change range of the health index. However, although this method is simple, it is prone to false alarms, has low reliability, and has problems such as being unable to judge more complex faults. Summary of the Invention

[0004] To solve the above problems, the present invention provides a method for fault diagnosis of mine drainage pumps. The RBFNN signal input of the present invention fuses the vibration, motor temperature, drainage port pressure, and suction port pressure signals of the mine drainage pump, and performs noise reduction and feature extraction. The trained RBFNN can accurately diagnose the rotor imbalance fault, cavitation fault, impeller fault, foundation loosening fault, inner bearing ring fault, bearing rotor fault, outer bearing ring fault, and comprehensive bearing fault of the mine drainage pump according to the real-time vibration, motor temperature, drainage port pressure, and suction port pressure data of the mine drainage pump collected in real time. Therefore, it has the beneficial effects of being real-time and accurate.

[0005] To achieve the above technical objectives, the technical solution of the present invention is: a method for fault diagnosis of mine drainage pumps, including the following steps:

[0006] Step 1: Collect the vibration signal, motor temperature, drainage port pressure, and suction port pressure of the mine drainage pump under fault;

[0007] Step 2: Perform noise reduction processing on the vibration signal in Step 1;

[0008] Step 3: Extract the eigenvalue of the signal in Step 2;

[0009] Step 4: Combine the motor temperature, drain port pressure, and suction port pressure with the eigenvalue in Step 3 to form a vector table, substitute it into the radial basis function neural network (RBFNN), train, test, and optimize it to generate a complete and mature RBFNN.

[0010] Step 5: Substitute the real-time vibration, motor temperature, drain port pressure, and suction port pressure data of the mine drainage pump to be diagnosed into the optimized RBFNN to obtain the real-time fault diagnosis result of the mine drainage pump.

[0011] Furthermore, in Step 1, the faults of the mine drainage pump include bearing faults, rotor imbalance faults, cavitation faults, impeller faults, and foundation loosening faults. Among them, bearing faults include inner bearing faults, bearing rotor faults, outer bearing faults, and comprehensive bearing faults.

[0012] Furthermore, in Step 2, minimum entropy deconvolution (MED) is used for noise reduction. The specific method includes the following steps:

[0013] Step 2.1: Initialize the 0th-order filter r 0 (n) so that all its elements are 1.

[0014] Step 2.2: Calculate the noise-reduced signal s(n), where s(n) = r -1 (n) * y(n), and y(n) is the vibration data before noise reduction.

[0015] Step 2.3: Calculate the vector b. The calculation formula of the vector b is:

[0016] Step 2.4: Calculate the autocorrelation matrix A of y(n).

[0017] Step 2.5: Reset the filter r(n). The calculation formula is: r i (n) = A -1 b i ;

[0018] Step 2.6: Determine whether the energy error of the filter coefficients of r i (n) and r i-1 (n) is within 1%, and whether the number of loops i is greater than the set value. If both are no, then i = i + 1 and continue to loop. If one of the above judgment conditions is yes, then jump out of the loop and calculate the noise-reduced vibration signal s(n).

[0019] Furthermore, the method for extracting eigenvalues in Step 3 includes the following steps:

[0020] Step 3.1: Decompose the vibration signal s(n) in Step 2.6 into 32 groups of decomposition signals from low frequency to high frequency through five-layer wavelet packet decomposition;

[0021] Step 3.2: Use kurtosis, skewness, and energy ratio as the eigenvalue of the decomposition signal in Step 3.1;

[0022] The calculation formula of kurtosis is:

[0023]

[0024] The calculation formula of skewness is:

[0025]

[0026] The calculation formula of energy ratio is:

[0027]

[0028] where d r (k) is the wavelet packet decomposition signal, E is the mean value of the data in the brackets, u is the signal mean value, σ is the signal root mean square, and E N is the total energy of the signal.

[0029] Furthermore, in Step 3.1, the denoised vibration signal is respectively passed through a high-pass filter and a low-pass filter to obtain a high-frequency signal and a low-frequency signal, and then the obtained high-frequency signal and low-frequency signal are respectively passed through a high-pass filter and a low-pass filter, and this cycle is repeated five times to obtain 32 groups of decomposition signals from low frequency to high frequency;

[0030] Furthermore, the vector table expression in Step 4 is:

[0031] L = [T, P o , P i , K v1 , …, K vm , B s1 , …, B sm , E p1 , …, E pm , where T is the motor temperature, P O is the drain port pressure, P i is the suction port pressure, K v is the kurtosis, B s is the skewness, and E p is the energy ratio.

[0032] Furthermore, the method for training, testing, and optimizing the RBFNN in Step 4 includes the following steps:

[0033] Step 4.1: Use n training data samples as the initial clustering center c j(1 ≤ j ≤ n);

[0034] Step 4.2: Calculate the distance between each object and the data center according to the mean value of each clustering object;

[0035] Step 4.3: Re-divide the corresponding objects according to the minimum distance, and recalculate and modify the mean value of each cluster;

[0036] Step 4.4: Loop Steps 4.2 and 4.3 until there is no transformation in each cluster. At this time, the clustering center is the center c of the radial basis function. j ;

[0037] Step 4.5: Solve the width σ of the radial basis function. σ is the average distance between the clustering center c j and the training samples;

[0038] Step 4.6: Solve the weight w of the output layer jk , and the weight of the output layer can be directly calculated by the least squares method according to the basis function center c j and the width σ of the radial basis function. The calculation formula is:

[0039]

[0040] In the formula: is the expected output, y k (l) is the actual output of the number of loop times; η is the learning rate; solving the weight w of the output layer jk can indicate that the training of the RBFNN neural network is completed. The final output of the neural network can be calculated according to the formula: is calculated.

[0041] As a preferred embodiment of the present invention, the motor temperature, drain port pressure, and suction port pressure in Steps 1 and 4 can be set to default.

[0042] The beneficial effects of the present invention are as follows:

[0043] The RBFNN signal input of the present invention fuses the vibration, motor temperature, drain port pressure, and suction port pressure signals of the mine drainage pump, and performs noise reduction and feature extraction. The trained RBFNN can accurately diagnose the rotor imbalance fault, cavitation fault, impeller fault, foundation loosening fault, bearing inner ring fault, bearing rotor fault, bearing outer ring fault, and bearing comprehensive fault of the mine drainage pump according to the real-time vibration, motor temperature, drain port pressure, and suction port pressure data of the mine drainage pump collected in real time. Therefore, it has the beneficial effects of being real-time and accurate. Description of the Drawings

[0044] Figure 1 is the MED noise reduction flow chart of the present invention;

[0045] Figure 2 This is the flow chart of the five - layer wavelet packet decomposition principle of the present invention;

[0046] Figure 3 This is the method for fault diagnosis of mine drainage pumps in the embodiments of the present invention. Specific embodiments

[0047] Next, the technical solution of the present invention will be described clearly and completely.

[0048] A method for fault diagnosis of mine drainage pumps includes the following steps:

[0049] Step 1: Collect the vibration signals, motor temperature, drainage port pressure, and suction port pressure of the mine drainage pump under faults, N for each;

[0050] Step 2: Denoise the vibration signal in Step 1 to obtain s(n);

[0051] Step 3: Extract the eigenvalues of the signal in Step 2;

[0052] Step 4: Combine the motor temperature, drainage port pressure, suction port pressure with the eigenvalues in Step 3 to form a vector table L, substitute it into the radial basis function neural network (RBFNN), train, test, and optimize it to generate a complete and mature RBFNN;

[0053] Step 5: Substitute the real - time vibration, motor temperature, drainage port pressure, and suction port pressure data of the mine drainage pump to be diagnosed into the optimized RBFNN to obtain the real - time fault diagnosis result of the mine drainage pump.

[0054] Furthermore, in Step 1, the faults of the mine drainage pump include bearing faults, rotor imbalance faults, cavitation faults, impeller faults, and foundation loosening faults. Among them, bearing faults include inner - ring bearing faults, bearing rotor faults, outer - ring bearing faults, and comprehensive bearing faults. Preferably, in order to collect non - destructive data of the drainage pump, the drainage pump faults are simulated by building a water pump drainage test bench.

[0055] Furthermore, as Figure 1 shown, in Step 2, minimum entropy deconvolution (MED) is used for denoising. The specific method includes the following steps:

[0056] Step 2.1: Initialize the 0 - order filter r 0 (n), and make all its elements equal to 1;

[0057] Step 2.2: Calculate the denoised signal s(n), where s(n)=r -1 (n)*y(n), and y(n) is the vibration data before denoising;

[0058] Step 2.3: Calculate vector b. The calculation formula for vector b is as follows:

[0059]

[0060] Step 2.4: Calculate the autocorrelation matrix A of y(n);

[0061] Step 2.5: Reset the filter r(n). The calculation formula is: r i (n) = A -1 b i ;

[0062] Step 2.6: Determine whether the energy error ε of the filter coefficients of r i (n) and r i-1 (n) is within 1%, and whether the loop count i is greater than the set value. If both are no, then i = i + 1 and continue the loop. If one of the above determination conditions is yes, then jump out of the loop and calculate the denoised vibration signal s(n).

[0063] Furthermore, the method for extracting eigenvalues in step 3 includes the following steps:

[0064] Step 3.1: As Figure 2 shown, decompose the vibration signal s(n) in step 2.6 into 32 groups of decomposition signals from low frequency to high frequency through five-layer wavelet packet decomposition;

[0065] Step 3.2: Take kurtosis, skewness, and energy ratio as the eigenvalues of the decomposition signals in step 3.1;

[0066] The calculation formula for kurtosis is:

[0067]

[0068] The calculation formula for skewness is:

[0069]

[0070] The calculation formula for energy ratio is:

[0071]

[0072] where d r (k) is the wavelet packet decomposition signal, E is the mean of the data in the parentheses, u is the signal mean, σ is the signal root mean square, and E N is the total energy of the signal.

[0073] Further, in step 3.1, the denoised vibration signal is respectively passed through a high-pass filter and a low-pass filter to obtain a high-frequency signal and a low-frequency signal, and then the obtained high-frequency signal and low-frequency signal are respectively passed through a high-pass filter and a low-pass filter, and this is cycled five times to obtain 32 groups of decomposition signals from low frequency to high frequency;

[0074] Further, the vector table expression in step 4 is:

[0075] L = [T, P o , P i , K v1 , …, K vm , B s1 , …, B sm , E p1 , …, E pm , where T is the motor temperature, P O is the drain port pressure, P i is the suction port pressure, K v is the kurtosis, B s is the skewness, E p is the energy ratio.

[0076] Further, the method for training, testing and optimizing the RBFNN in step 4 includes the following steps:

[0077] Step 4.1: Take n training data samples as the initial clustering centers c j (1 ≤ j ≤ n);

[0078] Step 4.2: Calculate the distance between each object and the data center according to the mean of each clustering object;

[0079] Step 4.3: Re-divide the corresponding objects according to the minimum distance, and recalculate and modify the mean of each cluster;

[0080] Step 4.4: Loop steps 4.2 and 4.3 until each cluster does not change. At this time, the clustering center is the center c of the radial basis function j ;

[0081] Step 4.5: Solve the width σ of the radial basis function. σ is the average distance between the clustering center c j and the training samples;

[0082] Step 4.6: Solve the weight w of the output layer jk , and the weight of the output layer can be directly calculated by the least squares method according to the center c of the basis function j and the width σ of the radial basis function. The calculation formula is:

[0083]

[0084] In the formula: is the expected output, y k (l) is the actual output of the number of loop times; η is the learning rate; where there are basis function widths σ and σ j , since there are n training data samples, the previous basis function width σ refers to the basis function width in the data of the training data samples, while σ in the formula j emphasizes the basis function widths in different data samples.

[0085] Solve for the weight w of the output layer jk It can be indicated that the training of the RBFNN neural network is completed, and the final neural network output can be calculated according to the formula: Calculated. In the formula: w jk is the jth weight of the kth output.

[0086] As a preferred embodiment of the present invention, as Figure 3 shown, the motor temperature, drain port pressure, and suction port pressure settings in steps 1 and 4 are default, then the feature vector table L = [0, 0, 0, K v1 , …, K vm , B s1 , …, B sm , E p1 , …, E pm is a vector composed only of kurtosis, skewness, and energy ratio.

Claims

1. A method for fault diagnosis of mine drainage pumps, characterized in that, it includes the following steps: Step 1: Collect the vibration signals, motor temperature, drainage port pressure, and suction port pressure of the mine drainage pump under fault conditions; Step 2: Denoise the vibration signals in Step 1; Step 3: Extract the characteristic values of the signals in Step 2; Step 4: Combine the motor temperature, drainage port pressure, and suction port pressure with the characteristic values in Step 3 to form a vector table, substitute it into the radial basis function neural network (RBFNN), train, test, and optimize it to generate a complete and mature RBFNN; Step 5: Substitute the real-time vibration, motor temperature, drainage port pressure, and suction port pressure data of the mine drainage pump to be diagnosed into the optimized RBFNN to obtain the real-time fault diagnosis result of the mine drainage pump; The method for training, testing, and optimizing the RBFNN in Step 4 includes the following steps: Step 4.1: Take n training data samples as the initial clustering centers c j (1 ≤ j ≤ n); Step 4.2: Calculate the distance between each object and the data center according to the mean value of each clustering object; Step 4.3: Re-divide the corresponding objects according to the minimum distance, and re-calculate and modify the mean value of each cluster; Step 4.4: Loop steps 4.2 and 4.3 until each cluster does not change, and the cluster centers at this time are the centers c of the radial basis functions j ; Step 4.5: Solve for the width σ of the radial basis function, where σ is the average distance between the cluster center c j and the training samples; Step 4.6: Solve the weights w of the output layer jk , and the weights of the output layer can be directly calculated by the least squares method according to the centers c of the basis functions j and the widths σ of the radial basis functions. The calculation formula is as follows: Wherein: is the expected output, y k (l) is the actual output of the number of loop times; η is the learning rate; the weight value w of the output layer is solved jk It can be indicated that the training of the RBFNN neural network is completed, and the final output of the neural network can be calculated according to the formula: is calculated.

2. The method for fault diagnosis of mine drainage pumps according to claim 1, characterized in that, in Step 1, the faults of the mine drainage pump include bearing faults, rotor imbalance faults, cavitation faults, impeller faults, and foundation loosening faults, where bearing faults include inner ring faults of the bearing, bearing rotor faults, outer ring faults of the bearing, and comprehensive bearing faults.

3. The method for fault diagnosis of mine drainage pumps according to claim 1, characterized in that, in Step 2, minimum entropy deconvolution (MED) is used for denoising, and the specific method includes the following steps: Step 2.1: Initialize the 0th-order filter r 0 (n) such that all its elements are 1 Step 2.2: Calculate the denoised signal s(n), where s(n) = r -1 (n) * y(n), and y(n) is the vibration data before denoising; Step 2.3: Calculate vector b. The calculation formula for vector b is as follows: Step 2.4: Calculate the autocorrelation matrix A of y(n); Step 2.5: Reset the filter r(n), and the calculation formula is: r i (n) = A -1 b i ; Step 2.6: Determine whether the energy errors of the two filter coefficients r i (n) and r i-1 (n) are within 1%, and whether the number of loops i is greater than the set value. If both are no, then i = i + 1 and continue the loop. If one of the above judgment conditions is yes, then jump out of the loop and calculate the vibration signal s(n) after noise reduction.

4. The method for fault diagnosis of mine drainage pumps according to claim 3, characterized in that, the method for extracting characteristic values in Step 3 includes the following steps: Step 3.1: Decompose the vibration signal s(n) in Step 2.6 into 32 groups of decomposition signals from low frequency to high frequency through five-layer wavelet packet decomposition; Step 3.2: Use kurtosis, skewness, and energy ratio as the characteristic values of the decomposition signals in Step 3.1; The calculation formula for kurtosis is: The calculation formula for skewness is: The calculation formula for energy ratio is: where d r (k) is the wavelet packet decomposed signal, E is the mean of the data within the brackets, u is the signal mean, σ is the root mean square of the signal, and E N is the total energy of the signal.

5. The method for fault diagnosis of mine drainage pumps according to claim 4, characterized in that, in Step 3.1, the denoised vibration signal is respectively passed through a high-pass filter and a low-pass filter to obtain a high-frequency signal and a low-frequency signal, and then the obtained high-frequency signal and low-frequency signal are respectively passed through a high-pass filter and a low-pass filter, and this is cycled five times to obtain 32 groups of decomposition signals from low frequency to high frequency.

6. The method for fault diagnosis of mine drainage pumps according to claim 5, characterized in that, the expression of the vector table in Step 4 is: L = [T, P o , P i , K v1 , …, K vm , B s1 , …, B sm , E p1 , tI, E pm , where T is the motor temperature, P O is the drain port pressure, P i is the suction port pressure, K v is the kurtosis, B s is the skewness, E p is the energy proportion.

7. The method for fault diagnosis of mine drainage pumps according to claim 1, characterized in that, the motor temperature, drainage port pressure, and suction port pressure in Step 1 and Step 4 can be set as default.

Images