CN114611562A - Gas turbine moving blade fracture fault identification method - Google Patents

Gas turbine moving blade fracture fault identification method Download PDF

Info

Publication number
CN114611562A
CN114611562A CN202210403016.1A CN202210403016A CN114611562A CN 114611562 A CN114611562 A CN 114611562A CN 202210403016 A CN202210403016 A CN 202210403016A CN 114611562 A CN114611562 A CN 114611562A
Authority
CN
China
Prior art keywords
fault
blade
gas turbine
frequency amplitude
extreme point
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210403016.1A
Other languages
Chinese (zh)
Inventor
王钟
龚建波
黄恩亮
郭磊
张坤
杜宇飞
李丹
杨光伟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Institute of Engineering Thermophysics of CAS
Original Assignee
Institute of Engineering Thermophysics of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Institute of Engineering Thermophysics of CAS filed Critical Institute of Engineering Thermophysics of CAS
Priority to CN202210403016.1A priority Critical patent/CN114611562A/en
Publication of CN114611562A publication Critical patent/CN114611562A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
    • G01H17/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves, not provided for in the preceding groups
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Signal Processing (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention discloses a method for identifying a fracture fault of a moving blade of a gas turbine, and belongs to the field of equipment fault diagnosis. Aiming at the current situation that the high-frequency vibration acceleration effective value which is widely adopted at present is difficult to realize the monitoring and identification of the blade state of the gas turbine, the fault sensitive characteristics which are monitored, identified and extracted in the engineering are provided by taking the static blade passing frequency amplitude, the movable blade passing frequency amplitude and the rotor power frequency amplitude, the accurate identification of the fracture fault of the movable blade of the gas turbine is realized by the huge advantage of the convolution wavelet transformation algorithm when singular signals are processed, and the interference of similar vibration characteristics caused by factors such as airflow fluctuation, rotor unbalance (non-blade fracture) and working condition adjustment can be effectively eliminated.

Description

Gas turbine moving blade fracture fault identification method
Technical Field
The invention belongs to the field of fault diagnosis, relates to a method for identifying a fracture fault of a moving blade of a gas turbine, and particularly relates to a method for identifying a fracture fault of a moving blade of a gas turbine based on signal singularity detection.
Background
The gas turbine is not only widely applied to the industrial fields of energy, electricity, pipeline transportation and the like, but also serves as a representative novel turbine type power machine, becomes a core power source of a new generation of warplanes and ship propulsion devices, and has wide development and application prospects. The blade is a core part for realizing power-energy conversion, and becomes a sensitive part with frequent faults in the whole combustion engine due to the long-term action of complex loads such as alternating, large-amplitude centrifugal force, thermal stress, aerodynamic force and the like, wherein cracks and fractures of the blade are main fault forms. Once a blade fracture fault occurs, if fault early warning and diagnosis cannot be realized before a fracture transient state or fracture, the performance efficiency of the gas turbine is influenced slightly, and when the fault is serious, the blade falling off at a high speed can damage other blades or puncture a casing, and even a serious accident that the combustion engine is damaged is caused.
Vibration analysis is the most common method for realizing fault diagnosis of mechanical parts, and at present, domestic combustion engines gradually adopt a monitoring means based on a high-frequency vibration acceleration effective value, but because vibration components in acceleration signals of casings are complex, the following limitations still exist when the states of blades of the combustion engines are monitored: (1) the effective acceleration value is a total value of vibration of each component in the combustion engine transmitted to the casing, blade vibration is not a dominant vibration component, and even if the blade vibration strength fluctuates due to a fracture fault, the total value is difficult to change remarkably, namely the sensitivity of the monitoring parameter to the blade fracture fault is poor; (2) fault information of key components such as bearings and rotors in the combustion engine is transmitted to an acceleration sensor at a casing, and targeted component state monitoring can be carried out only by successfully extracting fault vibration components of the components, namely fault mechanisms are not fully considered by the monitoring parameters, so that the monitoring parameters are difficult to be used for blade fracture fault identification as a reverse problem and to be distinguished from other faults. Therefore, the currently widely adopted effective value of the high-frequency vibration acceleration cannot realize the monitoring and identification of the blade state.
Considering that a gas turbine compressor generally has an arrangement mode of 'a (n-1) th stage stationary blade, an nth stage movable blade and an nth stage stationary blade', if a certain rotor blade in the nth stage movable blade is broken, the dynamic response of the whole stage rotor blade where the broken movable blade is located and the whole stage stator blade adjacent to the broken movable blade downstream can be significantly influenced. Specifically, the wake flow exciting force and the unbalanced centrifugal inertia force of the stator blade borne by the nth stage of rotor blade can generate corresponding changes along with the change of the acting area and the unbalance amount of the airflow, so that the exciting frequencies of the stator blade and the rotor blade, namely the stator blade passing frequency and the rotor power frequency, have obvious step-type sudden increase characteristics at the moment of failure; for the n-th stage stationary blade, because the front end excitation flow field is affected by the fracture fault, the wake flow excitation force of the moving blade is changed along with the fracture fault, so that the excitation frequency, namely the moving blade passing frequency, and the frequency amplitude shows a step-type sudden drop.
If the jump-type mutation characteristics of the characteristic parameters (meeting the definition of the first-class singular signals in engineering) are utilized to realize the effective capture of the fracture transient, the accurate detection of the characteristic mutation points needs to be researched and mastered. The existing mutation point detection algorithm comprises a Fisher optimal segmentation method, a heuristic segmentation algorithm, a sliding t/F detection method and the like. However, the above linear and stationary detection algorithm is not suitable for analyzing the non-stationary state data of the gas turbine. The wavelet transform is a time-frequency analysis method, has outstanding spatial locality, and can quickly and effectively extract fault transient information from non-stationary signals; meanwhile, the method also has the multi-resolution analysis capability, and has remarkable advantages when being applied to the gas turbine actual measurement vibration signals of dynamic transformation. The above two characteristics make it a powerful tool for detecting signal discontinuities.
Disclosure of Invention
In view of the above, aiming at the special step-type mutation characteristic of the fault sensitive characteristic parameter, the invention provides a method for identifying the fracture fault of the moving blade of the gas turbine based on signal singularity detection by combining a convolution wavelet transform algorithm, and the method can realize the effective identification of the fracture fault of the moving blade of the gas turbine.
The technical scheme adopted by the invention for realizing the technical purpose is as follows:
a method for identifying a fracture fault of a moving blade of a gas turbine based on signal singularity detection is characterized by comprising the following steps of:
SS 1: acquiring real-time vibration signal data x (t) of the gas turbine under a certain stable working condition, and taking the real-time vibration signal data x (t) as the input of a fracture fault identification model of a moving blade of the gas turbine;
SS 2: carrying out fault feature extraction on the real-time vibration signal data x (t) acquired in the step SS1, and respectively constructing fault sensitive feature vectors BPF taking the static blade passing frequency amplitude, the moving blade passing frequency amplitude and the rotor power frequency amplitude as target elementsS(t)、BPFR(t) and N (t), wherein the calculation formulas of the passing frequency of the static blade and the moving blade are respectively as follows:
the passing frequency of the stator blade is equal to the number of the (n-1) th stage stator blades multiplied by the power frequency of the rotor,
the passing frequency of the movable blades is equal to the n-th-stage movable blade number multiplied by the rotor power frequency;
SS 3: selecting a real function theta (t) with low-pass property as a smooth function, and constructing a wavelet function by taking a first derivative of the theta (t) as a wavelet base
Figure BDA0003600765110000031
Wavelet function described by a smoothing function theta (t) at a scale s
Figure BDA0003600765110000032
Can be expressed as:
Figure BDA0003600765110000033
in the formula, thetas(t) is a smoothing function at the scale s and satisfies:
Figure BDA0003600765110000034
SS 4: wavelet function constructed using step SS3
Figure BDA0003600765110000035
For the fault-sensitive feature vector BPF extracted in the step SS2S(t)、BPFR(t) and N (t) performing convolution wavelet transform to construct wavelet transform coefficient vectors Wf corresponding to the wavelet transform coefficientsS(t)、WfR(t) and WfN(t) of (d). Wherein, the convolution type wavelet transform satisfies the relation:
Figure BDA0003600765110000036
in the formula, Wf (s, t) is a wavelet transform coefficient; f (t) is the original signal to be detected; f (t) thetas(t) is a smoothing operator, representing the signal of the original signal after being "polished" by a smoothing function at a scale s;
in practice, the signal discontinuity detection is performed by smoothing the original signal at different scales and then detecting the original signal discontinuity by the first derivative of the "polished" signal. It can be easily found that the wavelet transform coefficients Wf (s, t) are just related to the "polished" post-signal f (t) θsThe first derivative of (t) is proportional. Therefore, the instantaneous mutation of the original signal can be accurately detected by using the local extreme point of the wavelet transform coefficient.
SS 5: judging wavelet transform coefficient vector WfS(t) whether each element in (t) is a local extreme point, if the detection result is the local extreme point, the step SS6 is entered, otherwise, a '0' is output at the position of the non-local extreme point element;
SS 6: detecting WfSEach local extreme point in (t) corresponds to a peak value PWf；
SS 7: judgment of PWfAnd (4) whether the peak value is a positive peak value (the positive peak value corresponds to a step-up type sudden change), outputting '1' at the positive peak element position when the condition is met, and otherwise outputting '0' at the non-positive peak element position. Thereby constructing a local extreme point peak value vector PS(t)；
SS 8: repeating the steps SS5 to SS7, and sequentially constructing local extreme point peak value vector PR(t) and PN(t) wherein a peak vector P is constructedRWhen (t), P should be judgedWfWhether the peak value is negative (the negative peak value corresponds to a step-down type sudden change); to construct a peak vector PN(t), still taking the positive peak value as a judgment condition;
SS 9: constructing a fault catastrophe detection vector P (t) in the following way:
P(t)＝PS(t)+PR(t)+PN(t)；
SS 10: judging whether each element in the fault catastrophe detection vector P (T) is equal to 3 or not, and if the condition is met, judging the position T of the element when P is equal to 3POutputting '1' and judging that the moving blade of the gas compressor of the gas turbine is at TPA fracture fault occurs at any moment; otherwise, outputting '0' at the position where the P ≠ 3 element, and considering that the blade fracture fault does not occur in the detection time period, thereby finally establishing a blade fracture fault identification vector.
The fault identification rules defined by the invention are as follows:
when all the three fault-sensitive characteristic parameters show specific step-type sudden changes at the same time, the fault of the gas turbine moving blade at the time is judged. As can be known from the construction principle of the fault mutation point detection vector, each element value in p (t) represents the number of sensitive features of a specific step type mutation in the current element position, so that whether a fault occurs or not and the fault occurrence time can be effectively judged by detecting the size and the position of each element in p (t).
Preferably, in step SS2, the characteristic frequency to be extracted mainly includes a stator blade passing frequency, a rotor blade passing frequency, and a rotor power frequency, and the frequency amplitudes of the three frequency components are used as target elements to perform fault characteristic extraction on the original vibration signal. Specifically, the original vibration signal is transformed into a frequency domain after Fourier transform, and then the maximum amplitude appearing within the range of +/-10 Hz with the characteristic frequency of theoretical calculation as the central frequency is taken as the target amplitude to be extracted.
Preferably, in step SS3, it is flatThe sliding function theta (t) selects a Gaussian function (Gauss) with low-pass property, and a wavelet function shown in the following formula is further constructed based on the Gaussian function (Gauss)
Figure BDA0003600765110000051
Figure BDA0003600765110000052
In the formula, the scale parameter s is 3.
Preferably, in step SS7, local extreme point peak vector PSThe (t) only contains the element "1" or "0", wherein the element "1" is the peak value P corresponding to the local extremum pointWfWhen the frequency amplitude value is greater than 0, outputting a result which shows that the static blade has step-up sudden change at the current element position through the frequency amplitude value; the element "0" is detected as a non-local extreme point or a peak P corresponding to a local extreme pointWfAnd outputting a result when the frequency amplitude is less than or equal to 0, wherein the result indicates that the static blade passing frequency amplitude does not generate specific step type mutation which meets the screening condition at the corresponding element position.
Preferably, in step SS8, local extreme point peak vector PRThe (t) only contains the element "1" or "0", wherein the element "1" is the peak value P corresponding to the local extremum pointWfThe output result when the frequency amplitude value is less than 0 indicates that the movable blade has step-down sudden change at the current element position through the frequency amplitude value, and the element 0 is detected as a non-local extreme point or a peak value P corresponding to the local extreme pointWfWhen the frequency amplitude is larger than or equal to 0, the output result indicates that the moving blade does not have specific step type mutation meeting the screening condition at the corresponding element position through the frequency amplitude; local extreme point peak vector PN(t) is also composed of element "1" or "0", where element "1" is the peak P corresponding to the local extremum pointWfWhen the output result is more than 0, the output result indicates that the rotor power frequency amplitude has step-up sudden change at the current element position, and the element '0' is detected as a non-local extreme point or a peak value P corresponding to a local extreme pointWfAnd outputting a result when the power frequency amplitude of the rotor is less than or equal to 0, wherein the result indicates that the specific step type mutation meeting the screening condition does not occur at the position of the corresponding element of the power frequency amplitude of the rotor.
Preferably, in step SS9, the fault discontinuity detection vector p (t) is mainly composed of elements "0" or "1" or "2" or "3", where the element value represents the number of sensitive features corresponding to the element position where a specific step type discontinuity occurs, and when the vane passing frequency amplitude, the blade passing frequency amplitude, and the rotor power frequency amplitude all show a specific step type discontinuity at the same time (i.e., output element "3"), it is determined that the gas turbine blade has a fracture fault at that time.
Compared with the prior art, the method for identifying the fracture fault of the moving blade of the gas turbine based on the signal singularity detection has the beneficial effects that: by combining the response rule of fault sensitive characteristic parameters with the great advantage of convolution wavelet transform in singular signal processing, the problem of weak sensitivity of the existing monitoring parameters can be well solved, and meanwhile, the interference of similar mutation phenomena caused by other factors (such as airflow fluctuation, rotor imbalance, working condition adjustment and other non-blade fracture) can be effectively eliminated, so that the fault identification accuracy rate of the gas turbine moving blade fracture is improved, and certain diagnosis reliability is ensured.
Drawings
FIG. 1 is a flow chart of the overall steps of the method of the present invention.
Fig. 2a to fig. 2c are schematic diagrams of fault-sensitive feature vectors extracted by the method of the present invention.
Fig. 3a to fig. 3c are schematic diagrams of wavelet transform coefficient vectors constructed by the method of the present invention, respectively.
Fig. 4a to 4c are schematic diagrams of local extreme point peak vectors output by the method of the present invention, respectively.
Fig. 5 is a schematic diagram of a fault catastrophe detection vector established by the method of the present invention.
Fig. 6 is a schematic diagram of the fault identification effect obtained by the method of the present invention.
Detailed Description
In order to make the implementation objects, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be described in more detail below with reference to the accompanying drawings in the embodiments of the present invention. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments, which are part of the present invention, are not all embodiments, and are intended to be illustrative of the present invention and should not be construed as limiting the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, taking a certain gas turbine as an example, the method for identifying the fracture failure of the moving blade of the gas turbine based on the signal singularity detection of the present invention is implemented mainly by the following steps:
step (1): when a certain gas turbine runs under a stable working condition (the 97 th sample point corresponds to the moment), serious mechanical faults occur, and after the gas turbine is stopped on site, the hole detection step by step finds that the ninth-stage moving blade of the low-pressure gas compressor has obvious fracture traces. The number of the broken stage movable blades is 42, the number of the front stage stationary blades is 48, and the rotating speed of the low-pressure compressor is 6940r/min when a fault occurs. Because the horizontal measuring point of the front casing is close to the fracture position and is more sensitive to the fault response of the low-pressure compressor, vibration data (153 sample points in total and sampling frequency of 51.2kHz) of the measuring point are input into the blade fracture fault identification model.
Step (2): carrying out fault feature extraction on the real-time vibration signal data acquired in the step (1), and respectively constructing fault sensitive feature vectors BPF with the static blade passing frequency amplitude, the movable blade passing frequency amplitude and the rotor power frequency amplitude as target elementsS(t)、BPFR(t) and N (t), wherein the calculation formulas of the passing frequency of the static blade and the moving blade are respectively as follows: the stator blade passing frequency is (n-1) th stage stator blade number × rotor power frequency, and the rotor blade passing frequency is (n-th stage rotor blade number × rotor power frequency).
Specifically, the characteristic frequencies to be extracted include 5552Hz, 4858Hz and 115.7Hz, which respectively correspond to the stator blade passing frequency, the rotor blade passing frequency and the rotor power frequency, and the frequency amplitudes of the three frequency components are used as target elements to perform fault characteristic extraction on the vibration signals of the horizontal measuring points of the front casing. Performing fourier transform on 0.64s data contained in each sample point, taking a theoretically calculated characteristic frequency as a maximum amplitude appearing in a range of the central frequency +/-10 Hz as a target amplitude to be extracted, and drawing a sample point-amplitude curve, as shown in fig. 2a, 2b and 2 c.
And (3): selecting a real function theta (t) with low-pass property as a smooth function, and constructing a wavelet function by taking a first derivative of the theta (t) as a wavelet base
Figure BDA0003600765110000071
Performing mutation point analysis, and describing wavelet function with smooth function theta (t) at scale s
Figure BDA0003600765110000081
Can be expressed as:
Figure BDA0003600765110000082
in the formula, thetas(t) is a smoothing function at the scale s and satisfies:
Figure BDA0003600765110000083
the invention selects Gaussian function (Gauss) with low-pass property as smooth function theta (t), and constructs wavelet function with first derivative of Gauss function
Figure BDA0003600765110000084
Mutation point analysis was performed. Wavelet function constructed based on Gauss function under scale s
Figure BDA0003600765110000085
Comprises the following steps:
Figure BDA0003600765110000086
in the formula, the scale parameter s is 3.
And (4):wavelet function constructed by means of step SS3
Figure BDA0003600765110000087
Performing convolution wavelet transform on the fault-sensitive feature vector extracted in step SS2, calculating each wavelet transform coefficient according to the following formula, and drawing a sample point-wavelet transform coefficient curve, as shown in fig. 3a, 3b and 3 c.
Figure BDA0003600765110000088
In the formula, Wf (s, t) is a wavelet transform coefficient; f (t) is the original signal to be detected; f (t) thetas(t) is a smoothing operator, representing the signal of the original signal after being "polished" by a smoothing function at a scale s; in practice, the signal discontinuity detection is performed by smoothing the original signal at different scales and then detecting the original signal discontinuity by the first derivative of the "polished" signal. It can be easily found that the wavelet transform coefficients Wf (s, t) are just related to the "polished" post-signal f (t) θsThe first derivative of (t) is proportional. Therefore, the instantaneous mutation of the original signal can be accurately detected by using the local extreme point of the wavelet transform coefficient.
And (5): judging wavelet transform coefficient vector WfS(t) whether each element in (t) is a local extreme point, if the detection result is the local extreme point, the step SS6 is entered, otherwise, a '0' is output at the position of the non-local extreme point element;
and (6): detecting WfSEach local extreme point in (t) corresponds to a peak value PWf；
And (7): judgment of PWfAnd (4) whether the peak value is a positive peak value (the positive peak value corresponds to a step-up type sudden change), outputting '1' at the positive peak element position when the condition is met, and otherwise outputting '0' at the non-positive peak element position. Thereby constructing a local extreme point peak value vector PS(t) of (d). Local extreme point peak vector PSThe (t) only contains the element "1" or "0", wherein the element "1" is the peak value P corresponding to the local extremum pointWfOutput result when > 0, indicating that the stator blade has stepped through the frequency amplitude at the current element positionAn up-type mutation; the element "0" is detected as a non-local extreme point or a peak P corresponding to a local extreme pointWfAnd outputting a result when the frequency amplitude is less than or equal to 0, wherein the result indicates that the static blade passing frequency amplitude does not generate specific step type mutation which meets the screening condition at the corresponding element position.
And (8): constructing local extreme point peak value vector
Repeating the steps SS5 to SS7, and sequentially constructing local extreme point peak value vector PR(t) and PN(t) wherein a peak vector P is constructedRWhen (t), P should be judgedWfWhether the peak value is negative (the negative peak value corresponds to a step-down type sudden change); to construct a peak vector PNIn the case of (t), the positive peak is still used as the determination condition. Local extreme point peak vector PRThe (t) only contains the element "1" or "0", wherein the element "1" is the peak value P corresponding to the local extremum pointWfThe output result when the frequency amplitude value is less than 0 indicates that the movable blade has step-down sudden change at the current element position through the frequency amplitude value, and the element 0 is detected as a non-local extreme point or a peak value P corresponding to the local extreme pointWfWhen the frequency amplitude is larger than or equal to 0, the output result indicates that the moving blade does not have specific step type mutation meeting the screening condition at the corresponding element position through the frequency amplitude; local extreme point peak vector PN(t) is also composed of element "1" or "0", where element "1" is the peak P corresponding to the local extremum pointWfWhen the output result is more than 0, the output result indicates that the rotor power frequency amplitude has step-up sudden change at the current element position, and the element '0' is detected as a non-local extreme point or a peak value P corresponding to a local extreme pointWfAnd outputting a result when the power frequency amplitude of the rotor is less than or equal to 0, wherein the result indicates that the specific step type mutation meeting the screening condition does not occur at the position of the corresponding element of the power frequency amplitude of the rotor.
Specifically, the detection process from step (5) to step (8) is continuously executed, and each element in the wavelet transform coefficient vector is screened and assigned. After the convolution wavelet transform, a plurality of element positions in each wavelet transform coefficient vector detect local extreme points. Wherein, WfS(t) local positive peak points occur at 36 th, 42 th, 51 th, 61 th, 69 th, 85 th, 97 th, 111 th, 117 th and 141 th sample points; wfR(t) in18, 31, 50, 69, 97, 124 and 142 th sample points of (a) present local negative peak points; wfNThe 13 th, 19 th, 30 th, 40 th, 65 th, 78 th, 85 th, 97 th, 111 th, 125 th and 139 th sample points in (t) show local positive peak points. The local positive peak point or the negative peak point is considered to have a specific step type mutation meeting the screening condition, and outputs "1" at the position of the corresponding sample point, and outputs "0" at the position of the other non-peak point, and a sample point-model output value curve is drawn, as shown in fig. 4a, 4b and 4 c.
And (9): constructing a fault catastrophe detection vector P (t) in the following way: p (t) PS(t)+PR(t)+PN(t) of (d). The fault catastrophe point detection vector P (t) mainly comprises elements of 0 or 1 or 2 or 3, wherein element values represent the sensitive characteristic number of specific step type catastrophe occurring at corresponding element positions, and when the static blade passing frequency amplitude, the movable blade passing frequency amplitude and the rotor power frequency amplitude all represent specific step type catastrophe at the same moment (namely, the element is output to be 3), the fact that the gas turbine movable blade has fracture fault at the moment is judged.
Specifically, the local extreme point peak vectors are superimposed, that is, the element values corresponding to the same sample point are directly added, so as to draw a sample point-model output value (superimposed) curve, as shown in fig. 5. Through the superposition operation among the peak value vectors, the element values corresponding to 13 th, 18 th, 19 th, 30 th, 31 th, 36 th, 40 th, 42 th, 50 th, 51 th, 61 th, 65 th, 78 th, 117 th, 124 th, 125 th, 139 th, 141 th and 142 th sample points in the constructed fault abrupt change point detection vector are 1, the element values corresponding to 69 th, 85 th and 111 th sample points are 2, the element value corresponding to 97 th sample point is 3, and the element values corresponding to the rest sample points are all 0. The element value represents the number of sensitive features of the current sample point position with a specific step type mutation.
Step (10): constructing a fault identification vector, judging whether each element in a fault catastrophe detection vector P (T) is equal to 3 or not, and if the condition is met, determining that P is 3 element position TPOutputting '1' and judging that a gas turbine compressor moving blade is at TPA fracture fault occurs at any moment; otherwise, the element is input at the position of P ≠ 3And (5) outputting '0', and considering that the blade fracture fault does not occur in the detection time period, thereby finally establishing a blade fracture fault identification vector.
According to the detection process (P ═ 3 element position output "1", P ≠ 3 element position output "0") in step (10), each element in the fault mutation point detection vector is screened and assigned, and a fault identification vector as shown in fig. 6 is constructed, wherein the model output value of the 97 th sample point is "1". And (3) judging that the gas turbine has a moving blade fracture fault within 0.64s corresponding to the 97 th sample point by combining the fault identification rule defined by the invention, namely realizing accurate capture of the blade fracture instant. The effectiveness of the method is verified through a certain real fault case of the blade fracture of the low-pressure compressor of the gas turbine.
The fault identification rules defined by the invention are as follows:
when the three fault-sensitive characteristic parameters all show specific step type sudden change at the same time, the fault of the gas turbine moving blade at the time is judged. As can be known from the construction principle of the fault mutation point detection vector, each element value in p (t) represents the number of sensitive features of a specific step type mutation in the current element position, so that whether a fault occurs or not and the fault occurrence time can be effectively judged by detecting the size and the position of each element in p (t).
The object of the present invention is fully effectively achieved by the above embodiments. Those skilled in the art will appreciate that the present invention includes, but is not limited to, what is described in the accompanying drawings and the foregoing detailed description. While the invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not to be limited to the disclosed embodiment, but on the contrary, is intended to cover various modifications within the spirit and scope of the appended claims.
The invention has not been described in detail and is part of the common general knowledge of a person skilled in the art.

Claims (7)

1. A method for identifying a fracture fault of a moving blade of a gas turbine based on signal singularity detection is characterized by comprising the following steps of:
SS 1: acquiring real-time vibration signal data x (t) of the gas turbine under a certain stable working condition, and inputting the real-time vibration signal data x (t) as a fracture fault identification model of the moving blade of the gas turbine;
SS 2: carrying out fault feature extraction on the real-time vibration signal data x (t) acquired in the step SS1, and respectively constructing fault sensitive feature vectors BPF taking the static blade passing frequency amplitude, the moving blade passing frequency amplitude and the rotor power frequency amplitude as target elementsS(t)、BPFR(t) and N (t), wherein the calculation formulas of the passing frequency of the static blade and the moving blade are respectively as follows:
stator blade passing frequency (n-1) th stage stator blade number multiplied by rotor power frequency)
The passing frequency of the movable blades is equal to the n-th-stage movable blade number multiplied by the rotor power frequency;
SS 3: selecting a real function theta (t) with low-pass property as a smooth function, and constructing a wavelet function by taking a first derivative of the theta (t) as a wavelet base
Figure FDA0003600765100000011
Wavelet function described by a smoothing function theta (t) at a scale s
Figure FDA0003600765100000012
Can be expressed as:
Figure FDA0003600765100000013
in the formula, thetas(t) is a smoothing function at the scale s and satisfies:
Figure FDA0003600765100000014
SS 4: wavelet function constructed using step SS3
Figure FDA0003600765100000015
For step SS2 extractionTo fault-sensitive feature vector BPFS(t)、BPFR(t) and N (t) performing convolution wavelet transform to construct wavelet transform coefficient vectors Wf corresponding to the wavelet transform coefficientsS(t)、WfR(t) and WfN(t), wherein the convolution type wavelet transform satisfies the relation:
Figure FDA0003600765100000016
in the formula, Wf (s, t) is a wavelet transform coefficient; f (t) is an original signal to be detected; f (t) thetas(t) is a smoothing operator, representing the signal of the original signal after being "polished" by a smoothing function at a scale s;
SS 5: judging wavelet transform coefficient vector WfS(t) whether or not each element in (t) is a local extreme point. When the detection result is the local extreme point, the step SS6 is carried out, otherwise, 0 is output at the position of the non-local extreme point element;
SS 6: detecting WfSEach local extreme point in (t) corresponds to a peak value PWf；
SS 7: judgment of PWfWhether the peak value is a positive peak value or not is judged, if the condition is met, 1 is output at the positive peak value element position, otherwise 0 is output at the non-positive peak value element position, and a local extreme point peak value vector P is constructedS(t)；
SS 8: repeating the steps SS5 to SS7, and sequentially constructing local extreme point peak value vector PR(t) and PN(t): constructing a peak vector PRWhen (t), P should be judgedWfWhether it is a negative peak; to construct a peak vector PN(t), still taking the positive peak value as a judgment condition;
SS 9: constructing a fault catastrophe detection vector P (t) in the following way:
P(t)＝PS(t)+PR(t)+PN(t)；
SS 10: judging whether each element in the fault catastrophe detection vector P (T) is equal to 3 or not, and if the condition is met, judging the position T of the element when P is equal to 3POutputting '1' and judging that the moving blade of the gas compressor of the gas turbine is at TPA fracture fault occurs at any moment; otherwise, in P ≠And outputting '0' by the 3-element position, and considering that the blade fracture fault does not occur in the detection time period, thereby finally establishing a blade fracture fault identification vector.
2. The method for identifying the fracture fault of the moving blade of the gas turbine based on the signal singularity detection according to claim 1, wherein in the step SS2, the characteristic frequencies to be extracted mainly include a stator blade passing frequency, a moving blade passing frequency and a rotor power frequency, and the fault characteristic extraction is performed on an original vibration signal by using frequency amplitudes of the three frequency components as target elements.
3. The method for identifying the fracture fault of the moving blade of the gas turbine based on the signal singularity detection as claimed in claim 2, wherein in the step SS2, the original vibration signal is transformed into a frequency domain after Fourier transformation, and then the maximum amplitude value appearing in the range of the central frequency of +/-10 Hz calculated by theory is taken as the target amplitude value to be extracted.
4. The method for identifying a fracture fault of a moving blade of a gas turbine based on signal singularity detection according to claim 1, wherein in step SS3, the smoothing function θ (t) is a Gaussian function with low-pass property, and a wavelet function represented by the following formula is further constructed based on the Gaussian function
Figure FDA0003600765100000031
Figure FDA0003600765100000032
In the formula, the scale parameter s is 3.
5. The method for identifying the fracture fault of the moving blade of the gas turbine based on the signal singularity detection of claim 1, wherein in the step SS7, the local extreme point peak value vector PS(t) contains only elements"1" or "0", wherein the element "1" is the peak P corresponding to the local extremum pointWfWhen the frequency amplitude value is greater than 0, outputting a result which shows that the static blade has step-up sudden change at the current element position through the frequency amplitude value; the element "0" is detected as a non-local extreme point or a peak P corresponding to a local extreme pointWfAnd outputting a result when the frequency amplitude is less than or equal to 0, wherein the result indicates that the static blade passing frequency amplitude does not generate specific step type mutation which meets the screening condition at the corresponding element position.
6. The method for identifying the fracture fault of the moving blade of the gas turbine based on the signal singularity detection of claim 1, wherein in the step SS8, the local extreme point peak value vector PRThe (t) only contains the element "1" or "0", wherein the element "1" is the peak value P corresponding to the local extremum pointWfThe output result when the frequency amplitude value is less than 0 indicates that the movable blade has step-down sudden change at the current element position through the frequency amplitude value, and the element 0 is detected as a non-local extreme point or a peak value P corresponding to the local extreme pointWfWhen the frequency amplitude is larger than or equal to 0, the output result indicates that the moving blade does not have specific step type mutation meeting the screening condition at the corresponding element position through the frequency amplitude; local extreme point peak vector PN(t) is also composed of element "1" or "0", where element "1" is the peak P corresponding to the local extremum pointWfWhen the output result is more than 0, the output result indicates that the rotor power frequency amplitude has step-up sudden change at the current element position, and the element '0' is detected as a non-local extreme point or a peak value P corresponding to a local extreme pointWfAnd outputting a result when the power frequency amplitude of the rotor is less than or equal to 0, wherein the result indicates that the specific step type mutation meeting the screening condition does not occur at the position of the corresponding element of the power frequency amplitude of the rotor.
7. The method for identifying a fracture fault of a moving blade of a gas turbine according to claim 1, wherein in step SS9, the fault discontinuity detection vector p (t) is mainly composed of elements "0" or "1" or "2" or "3", the element value represents the number of sensitive features corresponding to the element position where a specific step change occurs, and when the stator blade passing frequency amplitude, the moving blade passing frequency amplitude, and the rotor power frequency amplitude all show a specific step change at the same time, it is determined that a fracture fault of the moving blade of the gas turbine has occurred at that time.
CN202210403016.1A 2022-04-18 2022-04-18 Gas turbine moving blade fracture fault identification method Pending CN114611562A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210403016.1A CN114611562A (en) 2022-04-18 2022-04-18 Gas turbine moving blade fracture fault identification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210403016.1A CN114611562A (en) 2022-04-18 2022-04-18 Gas turbine moving blade fracture fault identification method

Publications (1)

Publication Number Publication Date
CN114611562A true CN114611562A (en) 2022-06-10

Family

ID=81869412

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210403016.1A Pending CN114611562A (en) 2022-04-18 2022-04-18 Gas turbine moving blade fracture fault identification method

Country Status (1)

Country Link
CN (1) CN114611562A (en)

Similar Documents

Publication Publication Date Title
US9739167B2 (en) Method and system for monitoring rotating blade health
Megerle et al. Numerical and experimental investigation of the aerodynamic excitation of a model low-pressure steam turbine stage operating under low volume flow
Sun et al. HHT-based feature extraction of pump operation instability under cavitation conditions through motor current signal analysis
US7698942B2 (en) Turbine engine stall warning system
EP2728331B1 (en) Methods and Systems for Monitoring Health of Blades
CN112461934B (en) Aero-engine blade crack source positioning method based on acoustic emission
Hu et al. Blade crack detection of centrifugal fan using adaptive stochastic resonance
Logan et al. Singular spectrum analysis as a tool for early detection of centrifugal compressor flow instability
Champ et al. Incipient surge detection in large volume energy systems based on wigner–ville distribution evaluated on vibration signals
Bornassi et al. Blade synchronous vibration measurements of a new upgraded heavy duty gas turbine MGT-70 (3) by using tip-timing method
Wu et al. Blade tip timing: from raw data to parameters identification
Zhou et al. Periodic analysis on gas path fault diagnosis of gas turbines
JP2015125147A (en) Methods and systems to monitor health of rotor blades
Alabied et al. Empirical mode decomposition of motor current signatures for centrifugal pump diagnostics
Wang et al. The method for identifying rotating blade asynchronous vibration and experimental verification
Stajuda et al. Aerodynamic instabilities detection via empirical mode decomposition in centrifugal compressors
Zhang et al. Aerodynamic system instability identification with sample entropy algorithm based on feature extraction
CN114611562A (en) Gas turbine moving blade fracture fault identification method
Bhattacharya et al. System to monitor blade health in axial flow compressors
Richards et al. Unsteady acoustic forcing on an impeller due to coupled blade row interactions
Lim et al. Motor fault detection method for vibration signal using FFT residuals
Wu et al. Foreign object damage diagnosis of aero-engine compressor based on damping averaging built-in matrix method
Gill et al. Experimental investigation of flutter in a single stage unshrouded axial-flow fan
CN113533530B (en) Blade natural frequency detection method of single blade end timing sensor
Li et al. Fault identification method of micro turbine blade based on the order spectral entropy of bearing vibration

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination