Wind turbine generator gear ring fault diagnosis method based on energy concentration characterization model

Abstract

The invention provides a wind turbine generator gear ring fault diagnosis method based on an energy concentration characterization model, and the method comprises the steps: carrying out the angular domain resampling through order tracking based on the energy concentration characteristic of fault impact, reducing the influence of rotating speed fluctuation, carrying out the grouping according to a mode that all planetary gears traverse one gear ring as a period, and carrying out the fault diagnosis. According to the method, fault impact components in signals are enhanced through square demodulation, signals in a group are approximated as weighted superposition of fault impact and passing effect distribution, then the energy ratio of fault impact to passing effect is identified by using Gaussian mixture distribution, the fault impact energy ratio of the whole period is counted based on threshold setting in a healthy state, and the fault impact energy ratio of the whole period is calculated. When the threshold value is exceeded, a fault is diagnosed, and otherwise, a healthy state is realized. Fault diagnosis of the gear ring of the wind turbine generator is achieved through the angle of energy distribution, economic benefits and operation safety of the wind turbine generator are guaranteed, and the method has important significance in engineering practice.

Description

Technical Field

[0001] The invention relates to the technical field of wind turbine gearbox gear ring fault diagnosis, and in particular to a wind turbine gear ring fault diagnosis method based on an energy concentration characterization model. Background Art

[0002] Gearbox is a key component for transmitting power in wind turbines, so timely and effective fault diagnosis is very important. Compared with parallel gears, the rotation speed of planetary gears is lower than that of parallel gears, and the fault energy is difficult to be significantly excited. In addition, because the planetary structure is more complex and there are more interference factors, the fault diagnosis of planetary gears is difficult. The ring gear fault is more special. In a healthy state, each time the planetary gear passes through the ring gear, a passing effect will be formed. In the frequency domain, there will be a modulation of the rotation frequency with the meshing frequency as the carrier and the number of planetary gears as multiples. In the fault state, the fault frequency is also modulated on the meshing frequency with the rotation frequency as a multiple of the number of planetary gears. The fault characteristics of the two are difficult to directly identify in the frequency domain, making it difficult to quantify the fault characteristics and explain the fault frequency. Therefore, it is of great significance in engineering practice to carry out fault diagnosis research on the ring gear of wind turbines to ensure the economic benefits and operation safety of wind turbines. Summary of the invention

[0003] The present invention designs a wind turbine ring gear fault diagnosis method based on an energy concentration characterization model.

[0004] The technical solution adopted by the present invention to solve the technical problem is: to provide a wind turbine ring gear fault diagnosis method based on an energy concentration characterization model, comprising the following steps:

[0005] Step 1: Take the acceleration signal collected by the wind turbine as input, resample the original signal in the angular domain using order tracking based on the speed information and sampling frequency, and establish an energy concentration characterization model based on the fault cycle;

[0006] Step 2: Use square demodulation method to enhance fault impulse energy;

[0007] Step 3: Calculate the energy proportion of the fault impact based on the Gaussian mixture model;

[0008] Step 4: Use the energy ratio of the fault impact as an indicator to diagnose the ring gear fault.

[0009] 2. The wind turbine ring gear fault diagnosis method based on the energy concentration characterization model according to claim 1 is characterized in that the step 1 is specifically:

[0010] Step 1.1: The acceleration signal x collected by the wind turbine is {x1, x2, x3, …, x n}, n is the signal length, the sampling frequency is f s, the rotation frequency of the planet carrier is f c , the number of planetary gears is N p , taking the planet carrier rotation frequency as the reference order, O c =1, order tracking is used to resample the original time domain signal in the angular domain, and the angular domain sampling frequency is f sn , the angular domain resampled signal can be expressed as A = {a1, a2, …, a m}, m is the number of resampling points.

[0011] Step 1.2: Calculate the number of rotations O based on the resampled angular signal A n =m / f sn , each circle contains the following points: O s =round(m / O n ), round(.) is the rounding function. Group them according to the period of all planetary gears traversing the gear ring once, and the number of points in each group is: g =N p *O s , the number of groups is J, J = round (m / O g ), then the regrouped set is R = {r1, r2, ..., r J}, r1={a1,a2,…,a Og}.

[0012] Step 1.3: Based on the different distributions of fault impact and through effect, the energy of fault impact in a single cycle is more concentrated than that of through effect. Therefore, the energy concentration characterization model can be expressed as a weighted superposition of the two distributions, expressed as:

[0013]

[0014] In the formula, Θ i , i=1,2 represents the energy concentration coefficient of fault impact and through effect, Γ i represents the energy distribution of fault impact and through effect, μ i ,σ i represents the mean and variance of the corresponding distribution, and the energy P(·) of a signal can be expressed as the weighted superposition of the two distributions.

[0015] 3. The wind turbine ring gear fault diagnosis method based on the energy concentration characterization model according to claim 1 is characterized in that the step 2 is specifically:

[0016] Step 2.1: In order to enhance the periodic characteristics of the fault impulse, square demodulation is used to enhance the signal. The square demodulation operator can be expressed as:

[0017] q(·)=(·) 2

[0018] Step 2.2: The set of enhanced period information is: Q = {q(r1),q(r2),…,q(r J )}.

[0019] 4. The wind turbine ring gear fault diagnosis method based on the energy concentration characterization model according to claim 1 is characterized in that the step 3 is specifically:

[0020] Step 3.1: Based on the set Q of enhanced information, for each period group, a Gaussian mixture model is constructed to identify the distribution state of the energy concentration representation model and evaluate the energy proportion Θ of the fault impact 1,j .

[0021] Step 3.2: Return the energy concentration factor of each distributed fault impulse: C = {Θ 1,1 ,Θ 1,2 ,…,Θ 1,J}.

[0022] 5. The wind turbine ring gear fault diagnosis method based on the energy concentration characterization model according to claim 1 is characterized in that the step 4 is specifically:

[0023] Step 4.1: Energy proportion C of health data normal The threshold value Th normal The calculation method is:

[0024]

[0025] In the formula, max(.) represents the maximum value function, mean(.) represents the mean function, N is the number of all samples, and the threshold Th normal As a criterion for evaluating health status.

[0026] Step 4.2: Count the energy proportion C of the global data and compare it with the threshold Th normal If the value exceeds the normal threshold, it is diagnosed as a fault, otherwise it is a healthy state. The beneficial effects of the present invention are:

[0027] The present invention proposes a wind turbine ring gear fault diagnosis method based on an energy concentration characterization model. Based on the energy concentration characteristics of the fault impact, order tracking is used for angular domain resampling to reduce the influence of speed fluctuations, and then grouping is performed according to the period of traversing one ring gear. The fault impact component in the signal is enhanced by square demodulation, and the signal in the group is approximated as a weighted superposition of the fault impact and the through effect distribution. Then, the Gaussian mixture distribution is used to identify the energy proportion of the fault impact and the through effect. Based on the threshold setting under the healthy state, the fault impact energy proportion of the whole cycle is counted. When it exceeds the threshold, it is diagnosed as a fault, otherwise it is a healthy state. The fault diagnosis of the wind turbine ring gear is realized from the perspective of energy distribution, which ensures the economic benefits and operation safety of the wind turbine, and has important significance in engineering practice. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 The present invention is a flow chart of a method for diagnosing wind turbine ring gear faults based on an energy concentration characterization model.

[0029] Figure 2 It is a time domain waveform diagram of the healthy state and fault state of the wind turbine group measured signal in Example 1 of a wind turbine group gear ring fault diagnosis method based on an energy concentration characterization model of the present invention.

[0030] Figure 3 It is a spectrum diagram of the healthy state and fault state of the wind turbine group measured signal in Example 1 of a wind turbine group gear ring fault diagnosis method based on an energy concentration characterization model of the present invention.

[0031] Figure 4 It is the angular domain resampled signal and its spectrum diagram of the health state of the wind turbine group's measured signal in Example 1 of a wind turbine group gear ring fault diagnosis method based on an energy concentration characterization model of the present invention.

[0032] Figure 5 It is the angular domain resampled signal and its spectrum diagram of the fault state of the wind turbine ring gear fault in Example 1 of the wind turbine ring gear fault diagnosis method based on the energy concentration characterization model of the present invention.

[0033] Figure 6 This is a diagram of square demodulation results in Example 1 of a wind turbine ring gear fault diagnosis method based on an energy concentration characterization model of the present invention.

[0034] Figure 7 This is an illustration of an example of solving parameters of the energy concentration characterization model in a healthy state and a fault state in Embodiment 1 of a wind turbine ring gear fault diagnosis method based on an energy concentration characterization model of the present invention.

[0035] Figure 8This is a diagram of the gear ring fault diagnosis results in Example 1 of a wind turbine gear ring fault diagnosis method based on an energy concentration characterization model of the present invention. DETAILED DESCRIPTION

[0036] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and Example 1.

[0037] like Figure 1 FIG. 1 is a flow chart of a method for diagnosing a ring gear fault of a wind turbine generator system based on an energy concentration characterization model, which specifically includes the following steps:

[0038] Step 1: Take the acceleration signal collected by the wind turbine as input, resample the original signal in the angular domain using order tracking based on the speed information and sampling frequency, and establish an energy concentration characterization model based on the fault cycle;

[0039] Step 1.1: The acceleration signal x collected by the wind turbine is {x1, x2, x3, …, x n}, n is the signal length, the sampling frequency is f s , the rotation frequency of the planet carrier is f c , the number of planetary gears is N p , taking the planet carrier rotation frequency as the reference order, O c =1, order tracking is used to resample the original time domain signal in the angular domain, and the angular domain sampling frequency is f sn , the angular domain resampled signal can be expressed as A = {a1, a2, …, a m}, m is the number of resampling points.

[0040] Step 1.2: Calculate the number of rotations O based on the resampled angular signal A n =m / f sn , each circle contains the following points: O s =round(m / O n ), round(.) is the rounding function. Group them according to the period of all planetary gears traversing the gear ring once, and the number of points in each group is: g =N p *O s , the number of groups is J, J = round (m / O g ), then the regrouped set is R = {r1, r2, ..., r J}, r1={a1,a2,…,a Og}.

[0041] Step 1.3: Based on the different distributions of fault impact and through effect, the energy of fault impact in a single cycle is more concentrated than that of through effect. Therefore, the energy concentration characterization model can be expressed as a weighted superposition of the two distributions, expressed as:

[0042]

[0043] In the formula, Θ i , i=1,2 represents the energy concentration coefficient of fault impact and through effect, Γ i represents the energy distribution of fault impact and through effect, μ i ,σ i represents the mean and variance of the corresponding distribution, and the energy P(·) of a signal can be expressed as the weighted superposition of the two distributions.

[0044] Step 2: Use square demodulation method to enhance fault impulse energy;

[0045] The step 2 specifically includes the following sub-steps:

[0046] Step 2.1: In order to enhance the periodic characteristics of the fault impulse, square demodulation is used to enhance the signal. The square demodulation operator can be expressed as:

[0047] q(·)=(·) 2

[0048] Step 2.2: The set of enhanced period information is: Q = {q(r1),q(r2),…,q(r J )}.

[0049] Step 3: Calculate the fault impact energy distribution ratio based on the Gaussian mixture model;

[0050] The step 3 specifically includes the following sub-steps:

[0051] Step 3.1: Based on the set Q of enhanced information, for each period group, a Gaussian mixture model is constructed to identify the distribution state of the energy concentration representation model and evaluate the energy proportion Θ of the fault impact 1,j .

[0052] Step 3.2: Return the energy concentration factor of each distributed fault impulse: C = {Θ 1,1 ,Θ 1,2 ,…,Θ 1,J}.

[0053] Step 4: Use the fault impact energy ratio as an indicator to diagnose the ring gear fault.

[0054] The step 4 specifically includes the following sub-steps:

[0055] Step 4.1: Energy proportion C of health data normal The threshold value Th normal The calculation method is:

[0056]

[0057] In the formula, max(.) represents the maximum value function, mean(.) represents the mean function, N is the number of all samples, and the threshold Th normal As a criterion for evaluating health status.

[0058] Step 4.2: Count the energy proportion C of the global data and compare it with the threshold Th normal If it exceeds the normal threshold range, it is diagnosed as a fault, otherwise it is in a healthy state.

[0059] The specific steps of the present invention are as follows:

[0060] Step 1: Signal acquisition and angular domain resampling.

[0061] First, the acceleration signals collected by the wind turbine are used as input. These acceleration signals are usually obtained by sensors installed on the gearbox, reflecting the vibration of the gearbox during operation.

[0062] Next, based on the speed information and sampling frequency, order tracking is used to resample the original signal in the angular domain. Order tracking is an effective signal processing technology that can eliminate the impact of speed fluctuations on the signal, making the signal present a more stable characteristic in the angular domain. Through order tracking, the original time domain signal can be converted into an angular domain signal, which facilitates subsequent processing.

[0063] In the angular domain resampling process, it is necessary to determine the angular domain sampling frequency f sn . Usually the planet carrier rotation frequency f c For the reference order Oc=1, according to the sampling theorem and the speed information, calculate the appropriate angular domain sampling frequency f sn Then, the original time domain signal is resampled in the angular domain to obtain the angular domain resampled signal A = {a1, a2, ..., a m}, where m is the number of resampling points.

[0064] Finally, an energy concentration characterization model is established based on the fault cycle. Since the ring gear fault will generate a periodic impact signal, the angular domain resampled signals can be grouped according to the fault cycle, and each group contains a complete fault cycle signal. By analyzing these fault cycle signals, an energy concentration characterization model can be established for subsequent fault diagnosis.

[0065] Step 2: Square demodulation to enhance the fault impulse energy.

[0066] In order to enhance the periodic characteristics of fault impulse, the square demodulation method is used to enhance the signal. Square demodulation is an effective signal processing technology that can convert high-frequency components in the signal into low-frequency components, making it easier to detect and identify.

[0067] In the square demodulation process, the square value of the signal is first calculated, and then low-pass filtered to remove high-frequency noise components. The signal obtained after this processing is the square demodulated signal.

[0068] The square demodulated signals are grouped according to the fault period to obtain the set of enhanced period information Q = {q(r1),q(r2),…,q(r J )}, where J is the number of groups, r i The signal representing the i-th fault cycle.

[0069] Step 3: The Gaussian mixture model is used to calculate the fault impact energy ratio.

[0070] Based on the set Q of enhanced information, for each period group, a Gaussian mixture model is constructed to identify the distribution state of the energy concentration representation model and evaluate the energy proportion of the fault impact.

[0071] The Gaussian mixture model is a commonly used probability model that can describe a data set with multiple Gaussian distributions. In the present invention, the Gaussian mixture model is used to analyze the enhanced period information to obtain the energy distribution parameters (mean, variance, etc.) of the fault impact and the pass effect. Then, the energy proportion θ of the fault impact is calculated based on these parameters. 1,j , where j represents the jth periodic group.

[0072] Finally, the energy concentration coefficient C of each distributed fault impulse is returned = {Θ 1,1 ,Θ 1,2 ,…,Θ 1,J}. These coefficients will be used for subsequent fault diagnosis.

[0073] Step 4: Use the fault impact energy ratio to diagnose the ring gear fault.

[0074] First, according to the energy proportion C of health data normal Calculate the threshold Th normal . The energy proportion of health data C normal It can be obtained by analyzing the historical health status data. Then, according to the statistical characteristics of these health data (such as maximum value, mean value, etc.), the threshold Th is calculated. normal As a criterion for evaluating health status.

[0075] Next, the energy proportion C of the global data is counted and compared with the threshold Th normal The global data refers to the energy proportion data obtained by processing the acceleration signal collected during the actual operation of the wind turbine gearbox to be diagnosed. These data are compared with the threshold Th normal After comparison, if it exceeds the normal threshold range, it is diagnosed as a fault, otherwise it is in a healthy state.

[0076] like Figure 2 As shown, it is a time domain waveform diagram of the healthy state and faulty state of the measured signal of the wind turbine in Example 1 of a wind turbine ring gear fault diagnosis method based on an energy concentration characterization model of the present invention; wherein the black line represents the healthy state time domain signal, and the blue line represents the fault state time domain signal. It can be seen that the periodic impact of the blue line is more obvious.

[0077] Figure 3 As shown, it is a spectrum diagram of the healthy state and the faulty state of the measured signal of the wind turbine in Example 1 of a wind turbine ring gear fault diagnosis method based on an energy concentration characterization model of the present invention; it can be seen that the spectrum in the healthy state and the faulty state is relatively fuzzy, which is because the measured signal will have slight speed fluctuations, resulting in spectrum fuzziness.

[0078] Figure 4 As shown, it is the angular domain resampled signal and its spectrum of the healthy state of the wind turbine measured signal in Example 1 of a wind turbine ring gear fault diagnosis method based on an energy concentration characterization model of the present invention. Due to the random fluctuation of the rotation speed, the order tracking method is used to resample the time domain signal in the angular domain. It can be seen that the waveform of the angular domain signal in the healthy state is not distorted, and the spectrum is clearer, and the modulation components of the meshing frequency and the pass-through effect appear.

[0079] Figure 5 As shown, it is the angular domain resampled signal and its spectrum of the fault state of the wind turbine measured signal in Example 1 of the wind turbine ring gear fault diagnosis method based on the energy concentration characterization model of the present invention. It can be seen that the angular domain signal is not distorted in the fault state, and the frequency domain is also clearer. It is worth mentioning that the modulation frequency band energy of the meshing frequency is enhanced after the fault occurs, but no new modulation components appear. Therefore, it is difficult to explain the fault situation from the perspective of fault frequency.

[0080] Figure 6 FIG. 1 is a diagram showing the square demodulation result of a wind turbine ring gear fault diagnosis method based on an energy concentration characterization model in Example 1 of the present invention. It can be seen that the periodic impact is more obvious in the fault state and the result is more sparse, which means that the impact energy of the fault signal is more concentrated, while the energy of the healthy state signal is still relatively dispersed after square demodulation.

[0081] Figure 7The figure shows an example of solving the parameters of the energy concentration characterization model in the healthy state and the fault state in Example 1 of a wind turbine gear ring fault diagnosis method based on the energy concentration characterization model of the present invention. The black scatter plot represents the square demodulation result of traversing one circle of the gear ring in the healthy state, the blue scatter plot represents the square demodulation result of traversing one circle of the gear ring in the fault state, the red dotted circle represents the energy proportion of the passing effect, and the area outside the red dotted circle represents the energy proportion of the fault impact. It can be seen that the energy proportion of the fault impact in the fault state is higher than the energy proportion of the fault impact in the healthy state. Specifically, there are more blue scatter points outside the red dotted circle in the fault state.

[0082] Figure 8 The figure is a quantitative result diagram of Example 1 of a wind turbine ring gear fault diagnosis method based on an energy concentration characterization model of the present invention. The red dotted line represents the threshold set in the healthy state, the black line represents the quantitative index diagram of the fault energy ratio in the healthy state and the fault state, and the horizontal axis is the number of groups for each regrouping, r1 to r 10 For normal state grouping, r 11 To r 21 is the fault status group. It can be seen that the indicator is 11 When the threshold is exceeded, it is diagnosed as a fault state, which is consistent with the prior information and shows the effectiveness of the method.

Claims

1. A wind turbine ring gear fault diagnosis method based on an energy concentration characterization model, characterized in that: The following steps are involved: Step 1: Take the acceleration signal collected by the wind turbine as input, resample the original signal in the angular domain using order tracking based on the speed information and sampling frequency, and establish an energy concentration characterization model based on the fault cycle; Step 2: Use square demodulation method to enhance fault impulse energy; Step 3: Calculate the energy proportion of the fault impact based on the Gaussian mixture model; Step 4: Use the energy ratio of the fault impact as an indicator to diagnose the ring gear fault.

2. A wind turbine ring gear fault diagnosis method based on energy concentration characterization model according to claim 1, characterized in that: The step 1 is specifically as follows: Step 1.1: The acceleration signal x collected by the wind turbine is {x1, x2, x3, …, x n }, n is the signal length, the sampling frequency is f s , the rotation frequency of the planet carrier is f c , the number of planetary gears is N p , taking the planet carrier rotation frequency as the reference order, O c =1, order tracking is used to resample the original time domain signal in the angular domain, and the angular domain sampling frequency is f sn , the angular domain resampled signal can be expressed as A = {a1, a2, …, a m }, m is the number of resampling points; Step 1.2: Calculate the number of rotations O based on the resampled angular signal A n =m / f sn , each circle contains the following points: O s =round(m / O n ), round(.) is the rounding function; group them according to the period of all planetary gears traversing the gear ring for one circle, and the number of points in each group is: g =N p *O s , the number of groups is J, J = round (m / O g ), then the regrouped set is R = {r1, r2, ..., r J }, r1={a1,a2,…,a Og }; Step 1.3: Based on the different distributions of fault impact and through effect, the energy of fault impact in a single cycle is more concentrated than that of through effect; therefore, the energy concentration characterization model can be expressed as a weighted superposition of the two distributions, expressed as: In the formula, Θ i , i=1,2 represents the energy concentration coefficient of fault impact and through effect, Γ i represents the energy distribution of fault impact and through effect, μ i ,σ i represents the mean and variance of the corresponding distribution, and the energy P(·) of a signal can be expressed as the weighted superposition of the two distributions.

3. The wind turbine ring gear fault diagnosis method based on the energy concentration characterization model according to claim 1 is characterized in that: The step 2 is specifically as follows: Step 2.1: In order to enhance the periodic characteristics of the fault impulse, square demodulation is used to enhance the signal. The square demodulation operator can be expressed as: q(·)＝(·) 2 Step 2.2: The set of enhanced period information is: Q = {q(r1),q(r2),…,q(r J )}.

4. The wind turbine ring gear fault diagnosis method based on the energy concentration characterization model according to claim 1 is characterized in that: The step 3 is specifically as follows: Step 3.1: Based on the set Q of enhanced information, for each period group, a Gaussian mixture model is constructed to identify the distribution state of the energy concentration representation model and evaluate the energy proportion Θ of the fault impact 1,j ; Step 3.2: Return the energy concentration factor of each distributed fault impulse: C = {Θ 1,1 ,Θ 1,2 ,…,Θ 1,J }.

5. The wind turbine ring gear fault diagnosis method based on the energy concentration characterization model according to claim 1 is characterized in that: The step 4 is specifically as follows: Step 4.1: Energy proportion C of health data normal The threshold value Th normal The calculation method is: In the formula, max(.) represents the maximum value function, mean(.) represents the mean function, N is the number of all samples, and the threshold Th normal As a criterion for evaluating health status; Step 4.2: Count the energy proportion C of the global data and compare it with the threshold Th normal If it exceeds the normal threshold range, it is diagnosed as a fault, otherwise it is in a healthy state.

Images