Fault detection method and system for turbine pump of liquid rocket engine
The vibration signal of the liquid rocket engine turbo pump is decomposed through the MODWT and EEMD-HHT models, eliminate zero drift, and identify inherent modal functions, which realizes accurate detection of turbo pump failures, improving data quality and diagnostic effects.
Patent Information
- Application Number
- CN202211700788.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-28
- Publication Date
- 2025-07-01
AI Technical Summary
The vibration signal of the turbo pump of the liquid rocket engine is affected by environmental noise, temperature and electromagnetic interference, resulting in zero drift phenomenon and affecting the accuracy of fault detection.
The maximum discrete wavelet transform (MODWT) is used to eliminate the vibration trend, and the integrated empirical modal decomposition (EEMD) and Hilbert-yellow transform (HHT) models are used to decompose the vibration signals, identify the inherent modal function (IMF) to detect faults.
Improve the acceleration data quality of vibration signals, ensuring the accuracy of subsequent health management and part-level fault diagnosis.
Smart Images
Figure CN120231759A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the fields of liquid rocket engine test measurement and signal processing technology, and specifically to a fault detection method and system for a liquid rocket engine turbine pump. Background Technique
[0002] The field of liquid rocket engine technology is extensive and the working environment is harsh. A large number of ground and high-altitude simulation tests are required from development to final application. The rocket engine turbine pump is one of the most important and harshest working condition components in a liquid rocket engine. Abnormal operation of the turbine pump is the main source of engine failure. In on-site test measurements, since the measured vibration signals are mainly obtained by sensors and acquisition devices, the vibration signals not only contain the power frequency information of the turbine pump operation, but also are mixed with vibration aliasing and noise excited by high-speed rotation. Therefore, it will inevitably be affected by on-site environmental noise, temperature, electromagnetic interference, etc., resulting in non-linear output-zero drift phenomenon of the sensor vibration signals.
[0003] In rocket ground experiments, piezoelectric sensors are usually used to collect vibration signals. The harsh environmental conditions at the test site will interfere with and generate noise in the measurement signals due to series-mode interference, common-mode interference, electrical noise, temperature, vibration, etc. Therefore, a fault detection method for a liquid rocket engine turbine pump needs to be set up to achieve rapid detection of abnormal vibrations. Summary of the Invention
[0004] Aiming at the problems existing in the prior art, the present invention provides a fault detection method and system for a liquid rocket engine turbine pump, which is applied to the fault detection of a liquid rocket engine turbine pump, improves the quality of acceleration data measured in experiments and analyzes vibration characteristics, and ensures subsequent health management and component-level fault diagnosis.
[0005] The present invention is realized through the following technical solutions:
[0006] A fault detection method for a liquid rocket engine turbine pump includes the following steps:
[0007] Step 1, determine the zero drift data of the vibration acceleration signal of the turbine pump;
[0008] Step 2, adopt the principle of maximum discrete wavelet transform to determine the vibration trend of the vibration acceleration signal, eliminate the vibration trend and return the zero drift data within the zero reference line to obtain the reconstructed vibration acceleration data;
[0009] Step 3, according to the principle of ensemble empirical mode decomposition, perform EEMD decomposition on the reconstructed vibration acceleration signal obtained in Step 2 to obtain the intrinsic mode function of the reconstructed vibration acceleration data;
[0010] Step 4: According to the Hilbert principle, construct an EEMD-HHT model to transform the intrinsic mode function IMF in Step 3, so as to find the mutation position and complete the fault detection of the turbo pump.
[0011] Preferably, in Step 1, the vibration acceleration signal is subjected to frequency-domain conversion, with the zero line as the reference line, and the data drifting away from the reference line is zero-drift data.
[0012] Preferably, the method of frequency-domain conversion is as follows:
[0013]
[0014] where f is the sampling frequency and ω is the angular frequency.
[0015] Preferably, the method for determining the signal vibration trend in Step 2 is as follows:
[0016] Establish a MODWT model, use the MODWT model to decompose the reconstructed vibration acceleration data within the window, obtain the decomposition layer with the largest proportion of relevant energy, and use this decomposition layer as the signal vibration trend.
[0017] Preferably, the expression of the MODWT model is as follows:
[0018]
[0019]
[0020]
[0021]
[0022] In the formula, are the approximation coefficients and detail coefficients of MODWT at different resolutions, g n , h n are the coefficients of the low-pass filter and high-pass filter, and N is the length of the discrete signal.
[0023] Preferably, the method for determining the intrinsic mode function in Step 3 is specifically as follows:
[0024] S3.1: Determine the time series function x(t) of the reconstructed vibration acceleration signal, and identify all the extreme points of the time series function x(t);
[0025] S3.2: Use the spline interpolation method to fit the maximum and minimum points of the time series function respectively, obtain the upper envelope line and lower envelope line corresponding to the order function, determine the average curve of the upper and lower envelope lines, and determine the intrinsic mode function pre-component according to the time series function and the average curve;
[0026] S3.3. Use the pre-component of the intrinsic mode function to replace the time series function, take the upper envelope curve and the lower envelope curve as the initial values for the next iteration, and repeat step S3.2 until the pre-component of the intrinsic mode function meets the set conditions, then obtain the first intrinsic mode function component and the remaining reconstructed vibration acceleration signal;
[0027] S3.4. Repeat step S3.1 to continue the EEMD decomposition of the remaining reconstructed vibration acceleration signal r1(t) until the value of r n (t) is less than the pre-set value or is a monotonic signal, and then end to obtain multiple intrinsic mode function components to form the intrinsic mode function IMF.
[0028] Preferably, in step 4, transform the intrinsic mode function IMF, obtain the Hilbert time-frequency spectrogram in the time-frequency plane, and determine the mutation position according to the time-frequency spectrogram.
[0029] A system for a fault detection method of a liquid rocket engine turbopump includes
[0030] A zero-drift data module for determining the zero-drift data of the vibration acceleration signal of the turbopump;
[0031] A reconstruction module for using the principle of maximum discrete wavelet transform to determine the vibration trend of the vibration acceleration signal, eliminating the vibration trend and regressing the zero-drift data within the zero reference line to obtain the reconstructed vibration acceleration data;
[0032] A decomposition module for decomposing the reconstructed vibration acceleration signal by EEMD according to the principle of ensemble empirical mode decomposition to obtain the intrinsic mode function of the reconstructed vibration acceleration data;
[0033] A detection module for constructing an EEMD-HHT model to transform the intrinsic mode function IMF according to the Hilbert principle, so as to find the mutation position to complete the fault detection of the turbopump.
[0034] Compared with the prior art, the present invention has the following beneficial technical effects:
[0035] A fault detection method for a liquid rocket engine turbopump provided by the present invention is applied to the fault detection of a liquid rocket engine turbopump. The maximum discrete wavelet (MODWT) is used to solve the zero drift problem. The zero drift problem is comprehensively solved by seeking trends and eliminating trends. Then, for the vibration acceleration signal after trend elimination, the ensemble empirical mode decomposition (EEMD) is used to perform modal decomposition on the vibration signals in three directions of the turbopump. Furthermore, the obtained intrinsic mode functions (IMFs) are subjected to Hilbert-Huang transform to construct an EEMD-HHT analysis model, analyze the sudden change characteristics of the signal energy of the turbopump, and apply it to the fault detection of the liquid rocket engine turbopump, improving the quality of the measured acceleration data in the experiment and analyzing the vibration characteristics to ensure subsequent health management and component-level fault diagnosis. Description of the Drawings
[0036] Figure 1 It is a flowchart of the fault detection method for the liquid rocket engine turbopump of the present invention;
[0037] Figure 2 It is a flowchart of seeking trends of the maximum discrete wavelet (MODWT) of the present invention;
[0038] Figure 3 It is a flowchart of the ensemble empirical mode decomposition (EEMD) of the present invention;
[0039] Figure 4 It is a graph of seeking trends of the MODWT of the vibration acceleration signal of the present invention;
[0040] Figure 5 It is a comparison graph of the vibration acceleration signal after eliminating trends and reconstructing by the MODWT and the original vibration acceleration signal of the present invention;
[0041] Figure 6 It is a result graph of the ensemble empirical mode decomposition (EEMD) of the present invention;
[0042] Figure 7 It is an instantaneous power spectrum graph of the vibration acceleration signal of the present invention. Detailed Embodiments
[0043] The following further describes the present invention in detail with reference to the drawings, which is an explanation rather than a limitation of the present invention.
[0044] Refer to Figure 1 , a fault detection method for a liquid rocket engine turbopump, includes the following steps:
[0045] Step 1, obtain the acceleration signal of the turbopump, and determine the zero drift data of the acceleration data with the zero line as the reference line.
[0046] Specifically, a piezoelectric sensor is used to record vibration acceleration signals. The recorded vibration acceleration signals are subjected to time-frequency domain conversion. Taking the zero line as the reference line, abnormal data segments drifting away from the reference line are determined to obtain zero-drift data.
[0047] The method of frequency domain conversion is as follows:
[0048]
[0049] Among them, f is the sampling frequency and ω is the angular frequency.
[0050] Step 2: According to the established wavelet model, determine the vibration trend of the vibration acceleration signal, eliminate the vibration trend, and make the zero-drift data return within the zero reference line to obtain a reconstructed vibration acceleration signal.
[0051] Specifically, refer to Figure 2 , based on the principle of the maximum discrete wavelet transform, establish a MODWT model, set the sampling frequency and wavelet packet function, use the MODWT model to decompose the reconstructed vibration acceleration data within the window, find the difference vector between adjacent two sampling points, determine the maximum value in the difference vector and the average value of several differences near the maximum value, set a threshold and compare each difference in the difference vector with the selected threshold one by one to judge whether the difference exceeds the selected threshold, calculate the proportion of the relevant energy of each level of decomposition, and find the decomposition layer with the maximum relevant energy as the vibration trend. Eliminate the vibration trend to pull the zero-drift data back within the zero reference line to obtain a reconstructed vibration acceleration signal.
[0052] The mathematical description formula of the MODWT model is as follows:
[0053]
[0054]
[0055]
[0056]
[0057] In the formula, are the approximation coefficients and detail coefficients of MODWT at different resolutions, g n , h n are the coefficients of the low-pass filter and high-pass filter, and N is the length of the discrete signal.
[0058] Step 3: According to the principle of ensemble empirical mode decomposition, decompose the reconstructed vibration acceleration signal obtained in Step 2 by EEMD to obtain the intrinsic mode functions of the reconstructed vibration acceleration data, specifically as follows:
[0059] S3.1. Add different white noise sequences to the acceleration signal described in step 1 to determine the time series function x(t) of the reconstructed vibration acceleration signal, and identify all extreme points of each time series function x(t).
[0060] S3.2. Use the cubic spline interpolation method to fit the maximum and minimum points of the time series function x(t) respectively, and obtain the upper envelope u(t) and the lower envelope v(t) corresponding to the time series function x(t).
[0061] S3.3. When v(t) ≤ x(t) ≤ u(t), establish the average curve m(t) of the upper and lower envelopes, and subtract the average curve m(t) from the time series function x(t) to decompose and obtain the pre-component h1(t) of the intrinsic mode function.
[0062]
[0063] h1(t) = x(t) - m(t) (6)
[0064] In the formula, m(t) is the average curve, and h1(t) is the intrinsic mode function.
[0065] S3.4. Use the pre-component h1(t) of the intrinsic mode function to replace the time series function x(t), take the upper envelope curve u1(t) and the lower envelope curve v1(t) as the initial values for the next iteration, and repeat step S3.3;
[0066]
[0067] h k (t) = h k-1 (t) - k-1 (t) (8)
[0068] Add different white noise to the time series function for each decomposition until the pre-component h k (t) of the intrinsic mode function meets the two necessary conditions of the intrinsic mode function. Then, perform an average operation on the obtained set of pre-components of the intrinsic mode function to obtain the first intrinsic mode function component C1(t) and the remaining reconstructed vibration acceleration signal r1(t);
[0069] C1(t) = h k (t) (9)
[0070] r1(t) = (t) - 1(t) (10)
[0071] The two necessary conditions are as follows: In the entire dataset, the sum of the number of local maxima and local minima must be equal to or at most differ by one from the number of zero crossings;
[0072] At any point in time, the mean value of the upper envelope defined by local maxima and the lower envelope defined by local minima is zero.
[0073] S3.5. Repeat step S3.1 to continue the EEMD decomposition of the remaining reconstructed vibration acceleration signal r1(t) until the value of r n (t) is less than a pre-set value or is a monotonic signal, and then end. Obtain multiple intrinsic mode function components to form the intrinsic mode function IMF. The instantaneous frequency and instantaneous amplitude of the intrinsic mode function IMF are both functions of time. The original signal x(t) can be regarded as the sum of all intrinsic mode function components and the residue, which is expressed as follows:
[0074] x(t) = C1(t) + C2(t) + … + C n (t) + R n (t) (11)
[0075] where, C n (t) is the IMF component, and R n (t) is the sum of the residues.
[0076] Step 4. According to the Hilbert principle, construct an EEMD-HHT model to transform the intrinsic mode function IMF in step 3, so as to find the mutation position to complete the fault detection.
[0077] According to the Hilbert principle, construct an EEMD-HHT model to perform Hilbert transform on the intrinsic mode function in step 3, obtain the Hilbert time-frequency spectrogram in the time-frequency plane, and thus find the mutation position to complete the fault detection.
[0078] In the Hilbert-Huang transform theory, the instantaneous frequency is defined as: For any time series x(t), the Hilbert transform y(t) is always expressed as follows:
[0079]
[0080] In the formula, P is the Cauchy principal value. When x(t) and y(t) are a pair of conjugate complex numbers, the analytic signal z(t) is obtained and expressed as follows:
[0081] Z(t) = x(t) + iy(t) = a(t)e iθ(t) (13)
[0082] where:
[0083] The parameters of the Hilbert-Huang transform are: the sampling frequency f s = 25600HZ, and the frequency resolution f0 = f s / N. According to the Nyquist theorem, the maximum frequency f max = f s / 2, and the minimum frequency f min = f0; where N is the length of the data.
[0084] The above-mentioned fault detection method for the turbopump of a liquid rocket engine finds the intrinsic mode functions (IMFs) of various signals in the vibration signal and decomposes them, and obtains the Hilbert energy distribution spectrogram in the time-frequency domain through the Hilbert-Huang transform of the IMF components, establishing the correspondence between spectral analysis and the time domain, analyzing the power spectrum of the signal, extracting the time-frequency key information in the signal, and analyzing the faults of the turbopump of the liquid rocket engine;
[0085] The present invention is from the data-driven level. After determining that the vibration acceleration data measured in the experiment is abnormal, the MODWT algorithm is used to perform wavelet decomposition on the original data to find the vibration trend of the original signal and eliminate and reconstruct the signal. Then, the reconstructed data is subjected to ensemble empirical mode decomposition using the EEMD algorithm, and the obtained intrinsic mode functions are transformed using the HHT algorithm to obtain the change with time at different frequencies and the vibration mutation points, completing the fault detection.
[0086] Embodiment 1
[0087] A fault detection method for the turbopump of a liquid rocket engine includes the following steps:
[0088] Step 1: Use a piezoelectric acceleration sensor to collect vibration acceleration signals at a frequency of 25,600 HZ, and perform time-domain conversion according to the frequency period according to the formula and, with the zero line as the reference line, determine the abnormal data segment drifting away from the reference line to obtain the zero-drift data.
[0089] Step 2: Refer to Figure 2 , which is the flowchart of finding the trend of the maximum discrete wavelet of MODWT. According to the principle of the maximum discrete wavelet transform, establish a MODWT model, select the wavelet basis function as sym4, set the sampling frequency to 25,600 HZ and the decomposition level to 6 layers, determine the decomposition layer with the largest proportion of relevant energy, obtain the vibration trend of the signal and ignore it, pull the zero-drift phenomenon back within the zero reference line, and obtain the reconstructed vibration acceleration signal. Refer to Figure 4 , which is the MODWT trend-finding diagram of the vibration acceleration signal, Figure 5 is the comparison diagram of the vibration acceleration signal after eliminating the trend and reconstruction and the original vibration acceleration signal.
[0090] Step 3: Refer to Figure 3 , the EEMD ensemble empirical mode decomposition flowchart, Figure 6It is the effect diagram of EEMD (Ensemble Empirical Mode Decomposition). According to the principle of ensemble empirical mode decomposition, the reconstructed vibration acceleration signal is decomposed by EEMD. All extreme points of the time series function x(t) are identified. The upper and lower envelope lines u(t) and v(t) are obtained by fitting the maximum and minimum points using the cubic spline interpolation method, satisfying v(t) ≤ x(t) ≤ u(t). The average curve m(t) of the upper and lower envelope lines is established. After subtracting the average curve m(t) from the time series function x(t), the pre-component h1(t) of the intrinsic mode function of the reconstructed signal vibration signal is obtained.
[0091] Use the pre-component h1(t) of the intrinsic mode function to replace the time series function x(t), and increase the iteration times of the upper envelope curve and the lower envelope curve by 1 to obtain u1(t) and v1(t) corresponding to the time series function h1(t). Repeat the above process until the pre-component h k (t) satisfies the two necessary conditions of IMF, and the first intrinsic mode function component is obtained.
[0092] Repeat the above process to continue the EEMD decomposition of the remaining reconstructed signal vibration signal r1(t) until r n (t) is less than the pre-set value or is a certain monotonic signal, and then end. Multiple intrinsic mode function components are obtained to form the intrinsic mode function IMF. The instantaneous frequency and instantaneous amplitude of the intrinsic mode function IMF are both functions of time. The original signal x(t) can be regarded as the sum of all intrinsic mode function components and the residue, which is expressed as follows:
[0093] x(t) = C1(t) + C2(t) + … + C n (t) + R n (t) (11)
[0094] Among them, C n (t) is the IMF component, and R n (t) is the sum of residues.
[0095] The parameters of the ensemble empirical mode decomposition are: the added noise level Nstd = 0.1, the number of added noise times NR = 100, the maximum number of intrinsic mode functions numIMF = 9, and the maximum iteration and screening times Maxsift = 10, to obtain IMF1 - IMF9 and the residual.
[0096] Step 4: Construct the EEMD - HHT model to perform the Hilbert transform on the intrinsic mode function, Figure 7 It is the instantaneous power spectrum diagram of the vibration acceleration signal. The parameters of the Hilbert - Huang transform are: the sampling frequency f s = 25600HZ, the frequency resolution f0 = f s / N. According to the Nyquist theorem, the maximum frequency f max = fs / 2, the minimum frequency f min = f0; where N is the length of the data. The model clearly shows the fluctuations on different time scales and the changes of different frequencies over time. Compared with the traditional empirical mode method, in the obtained power spectrum diagram, the mutation position is found to complete the fault detection.
[0097] The present invention also provides a system for a fault detection method of a liquid rocket engine turbopump, including a zero-drift data module, a reconstruction module, a decomposition module, and a detection module;
[0098] The zero-drift data module is used to determine the zero-drift data of the vibration acceleration signal of the turbopump;
[0099] The reconstruction module is used to determine the vibration trend of the vibration acceleration signal by using the principle of the maximum discrete wavelet transform, eliminate the vibration trend, and return the zero-drift data within the zero reference line to obtain the reconstructed vibration acceleration data;
[0100] The decomposition module is used to decompose the reconstructed vibration acceleration signal by EEMD according to the principle of the ensemble empirical mode decomposition to obtain the intrinsic mode function of the reconstructed vibration acceleration data;
[0101] The detection module is used to construct an EEMD-HHT model to transform the intrinsic mode function IMF according to the Hilbert principle, so as to find the mutation position to complete the fault detection of the turbopump.
[0102] The present invention proposes a fault detection method for a liquid rocket engine turbopump, which uses the maximum discrete wavelet (MODWT) to solve the zero-drift problem, comprehensively solves the zero-drift problem from finding the trend and eliminating the trend, and then uses the vibration acceleration signal after the trend elimination. The vibration signals in three directions of the turbopump are decomposed by EEMD, and then the obtained intrinsic mode function (IMF) is subjected to Hilbert-Huang transform to construct an EEMD-HHT analysis model to analyze the mutation characteristics of the signal energy of the turbopump, which is applied to the fault detection of the liquid rocket engine turbopump, improves the quality of the measured acceleration data in the experiment and analyzes the vibration characteristics, and ensures subsequent health management and component-level fault diagnosis.
[0103] The above content is only to illustrate the technical idea of the present invention, and the protection scope of the present invention cannot be limited thereby. Any changes made on the basis of the technical solution according to the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.
Claims
1. A fault detection method for a turbopump of a liquid rocket engine, characterized in that, It includes the following steps: Step 1: Determine the zero-drift data of the vibration acceleration signal of the turbo pump; Step 2: Adopt the principle of maximum discrete wavelet transform to determine the vibration trend of the vibration acceleration signal, eliminate the vibration trend to bring the zero-drift data back within the zero reference line, and obtain the reconstructed vibration acceleration data; Step 3: According to the principle of ensemble empirical mode decomposition, perform EEMD decomposition on the reconstructed vibration acceleration signal obtained in Step 2 to obtain the intrinsic mode function of the reconstructed vibration acceleration data; Step 4: According to the Hilbert principle, construct an EEMD-HHT model to transform the intrinsic mode function IMF in Step 3, so as to find the mutation position to complete the fault detection of the turbo pump.
2. The fault detection method of a liquid rocket engine turbopump according to claim 1, characterized in that In Step 1, perform frequency-domain conversion on the vibration acceleration signal. Taking the zero line as the reference line, the data drifting away from the reference line is the zero-drift data.
3. The fault detection method for a liquid rocket engine turbopump according to claim 2, characterized in that, The method of the frequency-domain conversion is as follows: where f is the sampling frequency and ω is the angular frequency.
4. A fault detection method for a turbopump of a liquid rocket engine according to claim 1, characterized in that The method for determining the signal vibration trend in Step 2 is as follows: Establish a MODWT model, and use the MODWT model to decompose the reconstructed vibration acceleration data within the window to obtain the decomposition layer with the largest proportion of relevant energy, and take this decomposition layer as the signal vibration trend.
5. A fault detection method for a liquid rocket engine turbopump according to claim 1, characterized in that The expression of the MODWT model is as follows: wherein, are the approximation coefficients and detail coefficients of MODWT at different resolutions, and g n , h n are the coefficients of the low-pass filter and the high-pass filter, and N is the length of the discrete signal.
6. A fault detection method for a turbopump of a liquid rocket engine according to claim 1, characterized in that, The method for determining the intrinsic mode function in Step 3 is specifically as follows: S3.1: Determine the time series function x(t) of the reconstructed vibration acceleration signal, and identify all extreme points of the time series function x(t); S3.2: Use the spline interpolation method to fit the maximum and minimum points of the time series function respectively to obtain the upper envelope line and the lower envelope line corresponding to the order function, determine the average curve of the upper and lower envelope lines, and obtain the pre-component of the intrinsic mode function according to the time series function and the average curve; S3.3: Use the pre-component of the intrinsic mode function to replace the time series function, take the upper envelope curve and the lower envelope curve as the initial values for the next iteration, and repeat Step S3.2 until the pre-component of the intrinsic mode function meets the set conditions, then obtain the first intrinsic mode function component and the remaining reconstructed vibration acceleration signal; S3.
4. Repeat step S3.1 to continue the EEMD decomposition of the remaining reconstructed vibration acceleration signal r1(t) until the value of r n (t) is less than a preset value or is a monotonic signal, and end to obtain multiple intrinsic mode function components to form the intrinsic mode function IMF.
7. A fault detection method for a turbopump of a liquid rocket engine according to claim 1, characterized in that, In Step 4, transform the intrinsic mode function IMF to obtain the Hilbert time-frequency spectrogram in the time-frequency plane, and determine the mutation position according to the time-frequency spectrogram.
8. A system for implementing the fault detection method of a liquid rocket engine turbopump according to any one of claims 1-7, characterized in that, It includes a zero-drift data module for determining the zero-drift data of the vibration acceleration signal of the turbo pump; a reconstruction module for adopting the principle of maximum discrete wavelet transform to determine the vibration trend of the vibration acceleration signal, eliminating the vibration trend to bring the zero-drift data back within the zero reference line, and obtaining the reconstructed vibration acceleration data; a decomposition module for performing EEMD decomposition on the reconstructed vibration acceleration signal according to the principle of ensemble empirical mode decomposition to obtain the intrinsic mode function of the reconstructed vibration acceleration data; a detection module for constructing an EEMD-HHT model to transform the intrinsic mode function IMF according to the Hilbert principle, so as to find the mutation position to complete the fault detection of the turbo pump.