Hydrogen compressor fault detection method based on variational cycle impact mode decomposition

Abstract

The invention discloses a hydrogen compressor fault detection method based on variational cycle impact mode decomposition, which comprises the following steps: firstly, establishing a working condition table according to field operation conditions, then acquiring surface vibration signal data of an exhaust valve of a cylinder when a compressor works normally, and then calculating variational cycle impact mode decomposition hyper-parameters; extracting a single-cylinder impact signal by using a variational cycle impact mode decomposition method; then piston vibration harmonic amplitude accumulative sums under different working conditions are extracted, and fault thresholds under the different working conditions are calculated; and finally, performing fault diagnosis on the acquired unknown vibration signal according to the fault threshold.

Description

Technical Field

[0001] The present invention belongs to the technical field of mechanical fault diagnosis, and more specifically, relates to a hydrogen compressor fault detection method based on variational cyclic impact mode decomposition. Background Art

[0002] In the hydrogen energy industry chain, hydrogen refueling stations are key infrastructures, and the operation and maintenance costs of their core equipment, hydrogen compressors, are as high as 35% of the total construction costs. Their operating status not only directly affects the economic benefits of hydrogen refueling stations, but also affects the safety level of the entire system. However, the regular maintenance model currently adopted by the industry is facing major technical challenges: under the condition of multi-cylinder parallel operation, the vibration signals between the cylinders have complex mutual coupling phenomena in the time domain, and are also affected by multi-source vibration interference such as mechanical shock and fluid pulsation, forming complex multi-source impact signals. These factors have led to obvious limitations in traditional time-frequency analysis methods in fault diagnosis, which are specifically manifested in inherent defects such as feature aliasing and insufficient signal-to-noise ratio, which seriously restrict the accuracy and reliability of equipment status monitoring.

[0003] At present, the existing fault diagnosis technology for reciprocating equipment mainly relies on frequency domain analysis methods such as empirical mode decomposition. Although such methods can decompose signals into multiple intrinsic mode functions, they also expose many significant shortcomings. On the one hand, there is an endpoint effect. Due to the discontinuity of the signal boundary, errors are easily generated in the construction process of the envelope, which affects the decomposition quality, especially when dealing with short-term signals or signals with drastic edge changes. This problem is particularly prominent; on the other hand, when facing multi-source impact signals generated when multiple cylinders work in parallel, serious modal aliasing will occur. Signals with different frequency components may be mixed in the same intrinsic mode function, resulting in a decrease in resolution and the inability to effectively separate the vibration signals of each cylinder, thus not having the ability to decouple multi-source impact signals. In view of the above problems, it is of great significance to develop a fault detection method that can effectively decompose multi-source impact signals, accurately extract fault features and realize early warning, which is of great significance to improve the operating reliability and safety of liquid-driven reciprocating hydrogen compressors. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a hydrogen compressor fault detection method based on variational cyclic impact mode decomposition, which realizes fault detection by collecting vibration signals of the compressor in a healthy state and setting thresholds using signal processing methods.

[0005] To achieve the above-mentioned object of the invention, the present invention provides a hydrogen compressor fault detection method based on variational cyclic impact mode decomposition, characterized in that it comprises the following steps:

[0006] (1) Collect the vibration signals at the surface of the cylinder exhaust valve when the hydrogen compressor is operating normally under different working conditions. Denote the vibration signal at the surface of the cylinder exhaust valve when operating normally under the k-th working condition as where represents the sampling value of the i-th sampling point under the k-th working condition, N is the number of sampling points, k = 1, 2, …, η, and η represents the number of operating conditions of the hydrogen compressor;

[0007] (2) Divide the vibration signal X k equally into n segments. Each segment of the vibration sub-signal is denoted as l = 1, 2, …, n;

[0008] (3) Traverse each segment of the vibration sub-signal in turn. Then, use the cyclic modulation spectrum to calculate the hyperparameters of the variational cyclic impact mode decomposition, including: the number of impacts M, the number of decompositions K, and the quadratic penalty factor α;

[0009] (4) Combine the hyperparameters of the variational cyclic impact mode decomposition and use the variational cyclic impact mode decomposition method to extract the cyclic impact signal of the cylinder;

[0010] (5) Perform an envelope spectrum on the cyclic impact signal, extract the amplitude A at the piston reciprocating frequency f f and the amplitudes A at its first S harmonics f,s , s = 1, 2, …, S; then sum up all the amplitudes and calculate the mean value, so as to extract the cumulative sum index of the piston vibration harmonic amplitudes

[0011]

[0012] (6) Obtain the fault diagnosis thresholds of the hydrogen compressor under different working conditions;

[0013] (6.1) Calculate the mean value μ k :

[0014]

[0015] (6.2) Calculate the standard deviation σ:

[0016]

[0017] (6.3) Determine the threshold range [μ k - 2σ k , μ k + 2σ k when the hydrogen compressor is operating normally under the k-th working condition;

[0018] (6.4) Traverse each operating condition of the hydrogen compressor in sequence, and determine the fault diagnosis threshold of the hydrogen compressor under different operating conditions according to steps (1) to (6.3);

[0019] (7) Conduct real-time fault detection on the hydrogen compressor;

[0020] Take a vibration signal with a known operating condition type and unknown health status, and then calculate the mean and standard deviation of the cumulative sum index of the piston vibration harmonic amplitude according to steps (1) to (6.2). Then, determine whether the corresponding mean and standard deviation are within the fault diagnosis threshold range under the corresponding operating condition. If so, it is determined that the hydrogen compressor is operating normally; otherwise, it is determined that the hydrogen compressor is operating with a fault.

[0021] The invention purpose of the present invention is realized as follows:

[0022] The hydrogen compressor fault detection method based on variational cyclic impact mode decomposition provided by the present invention first establishes an operating condition table according to the on-site operation situation, then obtains the vibration signal data on the surface of the cylinder exhaust valve when the compressor is operating normally, and then calculates the hyperparameters of the variational cyclic impact mode decomposition. The variational cyclic impact mode decomposition method is used to extract the single-cylinder impact signal; then, the cumulative sum of the piston vibration harmonic amplitudes under different operating conditions is extracted to calculate the fault thresholds under different operating conditions; finally, according to the fault thresholds, fault diagnosis is performed on the collected unknown vibration signals.

[0023] Meanwhile, the hydrogen compressor fault detection method based on variational cyclic impact mode decomposition provided by the present invention also has the following beneficial effects:

[0024] (1) The present invention combines the characteristics of the variational model and the cyclic vibration mode. By optimizing the iterative process, it can effectively decompose the multi-source impact signals generated during multi-cylinder operation into independent single-cylinder impact signals, reduce the interference components in the signals, and significantly improve the accuracy of fault diagnosis. Compared with other traditional methods, the variational cyclic impact mode decomposition can more accurately identify the single-cylinder impact signals, thereby effectively removing other irrelevant signals and improving the reliability of fault diagnosis.

[0025] (2) The present invention proposes the cumulative sum index of the piston vibration harmonic amplitude, which reflects the overall energy change of the cylinder vibration impact sequence at the cyclic frequency and its harmonic frequencies through envelope analysis; in addition, combined with the variational cyclic impact mode decomposition, this index can provide an effective quantitative basis for the health status assessment of the cylinder, showing broad application potential in the fault diagnosis of reciprocating equipment. Description of the Drawings

[0026] Figure 1 It is the flow chart of the hydrogen compressor fault detection method based on variational cyclic impact mode decomposition of the present invention;

[0027] Figure 2 It is a schematic diagram for extracting the cumulative sum of the harmonic amplitudes of piston vibration. Specific embodiments

[0028] The following describes the specific embodiments of the present invention in conjunction with the accompanying drawings so that those skilled in the art can better understand the present invention. It should be particularly noted that in the following description, when the detailed description of known functions and designs may obscure the main content of the present invention, these descriptions will be omitted here.

[0029] Embodiment

[0030] In this embodiment, as Figure 1 shown, a hydrogen compressor fault detection method based on variational cyclic impact mode decomposition of the present invention includes the following steps:

[0031] S1. Collect the vibration signals at the surface of the cylinder exhaust valve during the normal operation of the hydrogen compressor under different working conditions. Denote the vibration signal at the surface of the cylinder exhaust valve during the normal operation under the k-th working condition as where represents the sampling value at the i-th sampling point under the k-th working condition, N is the number of sampling points, k = 1, 2,..., η, and η represents the number of operating conditions of the hydrogen compressor;

[0032] In this embodiment, the hydrogen compressor usually operates under 8 different working conditions, as shown in Table 1 specifically. Among them, the first column is the working condition number, the second column is the motor rotation frequency, the third column is the intake pressure, and the fourth column is the secondary exhaust pressure;

[0033] Operating condition number Motor rotation frequency / Hz Inlet pressure / MPa Secondary exhaust pressure / MPa Operating condition 1 40 8 24 Operating condition 2 40 8 32 Operating condition 3 40 10 30 Operating condition 4 40 10 40 Operating condition 5 45 8 24 Operating condition 6 45 8 32 Operating condition 7 45 10 30 Operating condition 8 45 10 40

[0034] Table 1

[0035] In this embodiment, the sampling frequency is set to 25600, and the number of data points N collected under each working condition is 768000 points;

[0036] S2. Divide the vibration signal X k equally into n segments, and each segment of vibration sub-signal is denoted as l = 1, 2,..., n; in this embodiment, to increase the data length and ensure the accuracy of fault diagnosis, the vibration signal is divided into 4 segments, and each segment of signal is 30 s long;

[0037] S3. Traverse each segment of vibration sub-signal in turn, and then calculate the hyperparameters of variational cyclic impact mode decomposition by using cyclic modulation spectrum, including: the number of impacts M, the number of decompositions K, and the quadratic penalty factor α;

[0038] S3.1. Perform a short-time Fourier transform on the vibration sub-signal ​

[0039]

[0040] Among them, is the time-frequency representation of, w(t) represents the window function, ω represents the frequency, t represents the window center position, j is the imaginary unit, and τ is the integration variable; in this example, the Hanning window is selected as the window function;

[0041] S3.2. Take and take the absolute value to obtain the time-frequency spectrum;

[0042] S3.3. Perform discrete Fourier transform on each frequency slice in the time-frequency spectrum to obtain the cyclic modulation spectrum;

[0043] S3.4. Search for the frequency corresponding to the maximum peak in the cyclic modulation spectrum diagram and denote it as the piston reciprocating frequency f, and calculate the cycle period c:

[0044] c = 1 / f

[0045] S3.5. Calculate the impact number M:

[0046]

[0047] Among them, f s represents the sampling frequency, represents rounding down;

[0048] S3.6. Calculate the decomposition number K: By setting a threshold search for the number of peaks exceeding the product of the maximum peak and the threshold in the cyclic modulation spectrum diagram to determine the decomposition number K; the threshold should be set according to the specific device type and model. In this embodiment, the threshold is set to 0.1;

[0049] S3.7. Calculate the quadratic penalty factor α:

[0050]

[0051] Among them, δ represents the window lower limit. In this embodiment, the window lower limit δ of the Hanning window can control the size of the window width. To ensure that the impact components can be completely extracted and the influence of cross aliasing between components is minimized as much as possible, the window lower limit δ in this embodiment is set to 0.1;

[0052] S4. Combine the hyperparameters of the variational cyclic impact mode decomposition and use the variational cyclic impact mode decomposition method to extract the cyclic impact signal of the cylinder;

[0053] S4.1. Initialize the vibrator signal Frequency center of K cyclic impact components in Set to 0, we get:

[0054]

[0055] S4.2. Initialize the vibrator signal The time positions of K cyclic impact components in Thus, we get: Where, represents the initialization time position of the m-th impact in the p-th cyclic impact component;

[0056] S4.3. Define the number of iterations h and initialize h = 0;

[0057] S4.4. Extract the impacts after the h-th iteration

[0058]

[0059] Where, represents the m-th impact in the p-th cyclic impact component in the h-th iteration, represents the time position of the m-th impact in the p-th cyclic impact component in the h-th iteration, represents the frequency center position of the p-th cyclic impact component in the h-th iteration, λ represents the Lagrange multiplier,

[0060] S4.5. Update the frequency center after the h-th iteration;

[0061]

[0062] Where, represents the increment of the frequency center of the m-th impact in the p-th cyclic impact component in the h-th iteration;

[0063] S4.6. Update the time position after the h-th iteration;

[0064]

[0065] Where, represents the time position increment of the m-th impact in the p-th cyclic impact component in the h-th iteration;

[0066] S4.7. Update the Lagrange multiplier after the h-th iteration:

[0067]

[0068] Among them, τ represents a coefficient; the Lagrange multiplier ensures that the decomposed signal always satisfies the constraint conditions during the iteration process, and the coefficient τ controls the step size of the change of the Lagrange multiplier in each iteration process. In this example, the coefficient τ is set to 0.00001;

[0069] S4.8. Calculate the cyclic impact component after the h-th iteration

[0070]

[0071] Among them, represents the p-th in the h-th iteration;

[0072] S4.9. Determine whether the iteration convergence condition is satisfied:

[0073]

[0074] Among them, ε represents the error limit, which is used to ensure the accuracy of the solution process. In this embodiment, in order to simultaneously meet the accuracy and calculation speed, the error limit ε is set to 0.000001;

[0075] If the convergence condition is satisfied, output K cyclic impact components and enter step S4.10; if the convergence condition is not satisfied, increment the current iteration count h by 1, and then return to step S4.4;

[0076] S4.10. Perform an inverse short-time Fourier transform on each cyclic impact component to reconstruct the time-domain signal of each cyclic impact component, and then select the cyclic impact component corresponding to the maximum amplitude of the time-domain signal as the cyclic impact signal of the cylinder.

[0077] S5. Perform an envelope spectrum on the cyclic impact signal, and extract the amplitude A at the piston reciprocating frequency f f and the amplitudes A at its first S harmonics f,s , s = 1, 2,..., S; in this embodiment, S = 5 is taken;

[0078] Then sum up all the amplitudes and calculate the mean value, so as to extract the piston vibration harmonic amplitude sum index

[0079]

[0080] As Figure 2 shown, the amplitude at the piston reciprocating frequency f and the amplitudes at the first 5 harmonics are extracted from the envelope spectrum for calculating the piston vibration harmonic amplitude sum;

[0081] S6. Obtain the fault diagnosis threshold of the hydrogen compressor under different working conditions;

[0082] S6.1. Calculate the mean value μ k:

[0083]

[0084] S6.2. Calculate the standard deviation σ:

[0085]

[0086] S6.3. Determine the threshold range [μ k - 2σ k , μ k + 2σ k when the hydrogen compressor operates normally under the k-th working condition;

[0087] S6.4. Traverse each working condition of the hydrogen compressor in turn, and determine the fault diagnosis threshold of the hydrogen compressor under different working conditions according to steps S1 to S6.3;

[0088] S7. Perform real-time fault detection on the hydrogen compressor;

[0089] Take a vibration signal with a known working condition type and unknown health status, and then calculate the mean and standard deviation of the cumulative sum index of the piston vibration harmonic amplitude according to steps S1 to S6.2. Then, determine whether the corresponding mean and standard deviation are within the fault diagnosis threshold range under the corresponding working condition. If so, it is determined that the hydrogen compressor is operating normally; otherwise, it is determined that the hydrogen compressor is operating with a fault.

[0090] Example verification

[0091] In this embodiment, a comparative experiment was carried out using an LFC-D-470 / 50-200 / 450-PLN type liquid-driven reciprocating hydrogen compressor. Through machining, a circumferential wear band with a width of 0.5 mm and a depth of 1.2 mm was manufactured on the surface of the piston seal ring. This fault mode can cause the internal leakage rate in the cylinder to increase by 15% to 20%, which is a relatively common fault type in compressors (accounting for 43.7% of the on-site fault records). To verify the method of the present invention, normal and fault data sets were obtained, as shown in Table 2. Part of the healthy data was used as the diagnosis threshold, and both types of data were used for testing.

[0092] Table 2. Data set obtained in the laboratory

[0093] Fault type Duration / s Piston wear 120 Health data 120

[0094] This embodiment was tested on a device installed with the MATLAB R2019a environment.

[0095] Next, the effects of various fault diagnosis methods were counted, specifically including directly using the original data and analyzing it in combination with traditional indicators such as the cumulative sum of piston vibration harmonic amplitudes, RMS, and kurtosis value, as well as the results of fault diagnosis by combining the data processed by the variational cyclic impact modal decomposition method with RMS and kurtosis value. The statistical results are shown in Table 3.

[0096] Table 3. Comparison of the diagnosis results of the method of the present invention and other methods

[0097]

[0098] The results show that after the vibration signal is decomposed by the variational cyclic impact decomposition modal method, the diagnostic accuracy is significantly improved compared with directly using the original data. Among the selected indicators, the cumulative sum of piston vibration harmonic amplitudes shows significantly better diagnostic effects than traditional indicators such as RMS and kurtosis value. Especially when identifying piston faults under complex working conditions, it shows higher sensitivity and accuracy. This result fully proves the effectiveness of the present invention in detecting compressor faults.

[0099] Although the above describes the illustrative specific embodiments of the present invention for the convenience of those skilled in the art to understand the present invention, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions made using the concept of the present invention are within the scope of protection.

Claims

1. A hydrogen compressor fault detection method based on variational cyclic impact mode decomposition, characterized in that: The following steps are involved: (1) The vibration signal of the cylinder exhaust valve surface when the hydrogen compressor is working normally under different working conditions is collected. The vibration signal of the cylinder exhaust valve surface when the hydrogen compressor is working normally under the kth working condition is recorded as in, represents the sampling value of the i-th sampling point under the k-th working condition, N is the number of sampling points, k = 1, 2, ..., η, η represents the number of operating conditions of the hydrogen compressor; (2) The vibration signal X k It is divided into n segments, and the oscillator signal of each segment is recorded as l=1,2,…,n; (3) Traverse each oscillator signal in turn Then, the hyperparameters of the variational cyclic shock mode decomposition are calculated using the cyclic modulation spectrum, including the number of shocks M, the number of decompositions K, and the quadratic penalty factor α; (4) Combining the hyperparameters of variational cyclic shock decomposition, the cyclic shock signal of the cylinder is extracted using the variational cyclic shock modal decomposition method; (5) Make an envelope spectrum of the cyclic impact signal and extract the amplitude A at the piston reciprocating frequency f f And the amplitude A of the first S harmonics f,s , s=1,2,…,S; then all amplitudes are accumulated and averaged to extract the piston vibration harmonic amplitude accumulation index (6) Obtaining the fault diagnosis threshold of the hydrogen compressor under different working conditions; (6.1), calculate the mean μ k : (6.2), calculate the standard deviation σ: (6.3) Determine the threshold range [μ k -2σ k ,μ k +2σ k ]; (6.4) traverse each operating condition of the hydrogen compressor in turn, and determine the fault diagnosis threshold of the hydrogen compressor under different operating conditions according to steps (1) to (6.3); (7) Real-time fault detection of hydrogen compressor; Take a vibration signal with a known operating condition type and an unknown health status, and then calculate the mean and standard deviation of the piston vibration harmonic amplitude accumulation index according to steps (1) to (6.2), and then determine whether the corresponding mean and standard deviation are within the fault diagnosis threshold range under the corresponding operating condition. If so, it is determined that the operating status of the hydrogen compressor is normal, otherwise, it is determined that the operating status of the hydrogen compressor is faulty.

2. The hydrogen compressor fault detection method based on variational cyclic impact mode decomposition according to claim 1 is characterized in that: The method for calculating the hyperparameters of the variational cyclic impact mode decomposition using the cyclic modulation spectrum is: (2.1) for the oscillator signal Perform a short-time Fourier transform, in, for The time-frequency representation of w(t) is the window function, ω is the frequency, t is the window center position, j is the complex unit, and τ is the integral variable; (2.2) Take the absolute value to get the time spectrum; (2.3) Perform discrete Fourier transform on each frequency slice in the time-frequency spectrum to obtain the cyclic modulation spectrum; (2.4), search for the frequency corresponding to the maximum peak in the cyclic modulation spectrum and record it as the piston reciprocating frequency f, and calculate the cycle period c: c=1 / f (2.5) Calculate the impact number M: Among them, f s represents the sampling frequency, Indicates rounding down; (2.6) Calculate the decomposition number K: By setting the threshold Search for peaks exceeding the maximum value and threshold in the cyclic modulation spectrum The number of peaks of the product determines the decomposition number K; (2.7) Calculate the quadratic penalty factor α: Where δ represents the lower limit of the window.

3. The hydrogen compressor fault detection method based on variational cyclic impact mode decomposition according to claim 1 is characterized in that: The method for extracting the cyclic impact signal of the cylinder by using the variational cyclic impact modal decomposition method is: (3.1) Initialize the oscillator signal The frequency center of the K cyclic impulse components in Set it to 0, and we get: (3.2) Initialize the oscillator signal The time position of the K cyclic impact components in is Thus we get: in, represents the initialization time position of the mth impact in the pth cyclic impact component; (3.3) Define the number of iterations h, and initialize h = 0; (3.4) Extract the impact after the hth iteration in, represents the mth impulse in the pth cyclic impulse component in the hth iteration, represents the time position of the mth impulse in the pth cyclic impulse component in the hth iteration, represents the frequency center position of the p-th cyclic impulse component in the h-th iteration, λ represents the Lagrange multiplier, (3.5) Update the frequency center after the hth iteration; in, represents the increment of the frequency center of the mth impulse in the pth cyclic impulse component in the hth iteration; (3.6), Update the time position after the hth iteration; in, represents the time position increment of the mth impulse in the pth cyclic impulse component in the hth iteration; (3.7) Update the Lagrange multiplier after the hth iteration: Among them, τ represents the coefficient; (3.8) Calculate the cyclic impact component after the hth iteration in, represents the pth in the hth iteration; (3.9) Determine whether the iterative convergence condition is met: Among them, ε represents the error limit; If the convergence condition is met, K cyclic impact components are output and the process goes to step (3.10); if the convergence condition is not met, the current number of iterations h is increased by 1 and the process returns to step (3.4); (3.10) Perform inverse short-time Fourier transform on each cyclic impact component, reconstruct the time domain signal of each cyclic impact component, and then select the cyclic impact component corresponding to the maximum amplitude of the time domain signal as the cyclic impact signal of the cylinder.

Images