Real-time monitoring method for rotating and static collision of high-speed turbine pump based on RSAF model

Abstract

The application provides a high-speed turbine pump rotating and static collision real-time monitoring method based on an RSAF model. First, features participating in the collision monitoring are determined, and two features of rotating speed and rotating speed change amount are added to reflect the working condition state. The displacement peak-peak value and the fundamental frequency value are introduced to reflect the influence of the overall vibration level under the current assembly condition, so that the data of different working conditions and states can be unified for analysis. An RSAF model is established based on historical normal data of the turbine pump, and is used as a representation of the normal state. The displacement signal of the turbine pump is monitored in real time, the collision fault features of the displacement signal are extracted, the similarity of the features at each moment and the RSAF model is calculated, the estimated value at the moment is obtained, and whether the estimated value deviates from the normal state is judged. The application can unify the data of the turbine pump under different rotating speeds for processing, and effectively avoids false alarms and missed alarms.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of turbine pump health monitoring, in particular to a high-speed turbine pump rotating and static collision real-time monitoring method based on an RSAF model. BACKGROUND

[0002] The turbine pump is one of the important and complex components in the rocket engine, which undertakes the key task of delivering fuel and oxidizer to the combustion chamber. During the operation of the turbine pump, the rotor may deviate from its ideal trajectory due to the influence of unbalanced force, thermal deformation, fluid dynamic effect and other factors. In severe cases, this deviation may lead to contact and collision between the rotor and the stationary parts of the turbine pump, such as the pump casing and the seal ring. The occurrence of collision phenomenon not only reduces the working efficiency of the turbine pump, but also causes a series of serious consequences, leading to mission failure, even personnel casualties. Therefore, monitoring the collision failure of the turbine pump and diagnosing its severity are crucial for subsequent health management.

[0003] Currently, the health monitoring of rocket engines mainly uses a red-line shutdown health monitoring method based on key parameters, which mainly monitors the collision failure by monitoring the vibration signal. However, due to the complexity of the working conditions of the turbine pump in actual operation, the actual fault features involved are numerous, so the threshold selection of the existing method is difficult, and the judgment logic is also complex, resulting in limited fault coverage. In addition, due to improper threshold setting, the existing method often appears false alarm or missed alarm in actual use. For example, the vibration level of a certain rocket engine is large, which leads to the fault feature obtained by monitoring the vibration signal easily exceeding the threshold, but at this time there is no collision failure, resulting in false alarm. For a rocket engine with small vibration level, although the fault feature does not exceed the set threshold, the actual collision phenomenon may have occurred. SUMMARY

[0004] To solve the problems existing in the prior art, the present application proposes a high-speed turbine pump rotating and static collision real-time monitoring method based on an RSAF model. Based on historical data, the method establishes a baseline that can unify the normal state of multiple working conditions for the rich harmonic component features of turbine pump vibration and displacement data, judges the collision based on the normal state, and quantifies the degree of collision to reduce the probability of false alarm or missed alarm.

[0005] The technical scheme of the present application is as follows:

[0006] The high-speed turbine pump rotating and static collision real-time monitoring method based on the RSAF model comprises the following steps:

[0007] Step 1: Determine the feature vector participating in the collision monitoring as

[0008] [nt Δnt DF ch,1 …DF ch,k …DF ch,K P] 1×n

[0009] Among them, nt is the speed of the turbo pump, Δnt is the speed change, P is the vector composed of 1st, 2nd and 3rd order positive and negative precessions, DF ch,k is the feature vector of the kth detection channel, k = 1, 2, ..., K:

[0010] DF ch,k =[ppv k fb k fl 1,k … fl i,k … fl I,k f c,k f 2,k f 3,k f 4,k ]

[0011] PPV in the above characteristics k is the peak-to-peak amplitude of the kth detection channel, fb k is the amplitude corresponding to the fundamental frequency of the kth detection channel, fl 1,k …fl i,k …fl I,k is the amplitude corresponding to the I frequency point in the low-frequency spectrum of the kth detection channel; f c,k is the subharmonic amplitude corresponding to the kth detection channel, f 2,k 、f 3,k 、f 4,k is the amplitude corresponding to the 2nd, 3rd and 4th frequency in the kth detection channel;

[0012] Step 2: Obtain historical test data from the turbopump under normal working conditions, extract feature vectors from the test data under normal working conditions, divide the obtained feature vectors into a learning set and a test set, and train the RSAF model with the learning set as a representation of the normal state; input the test set into the trained RSAF model to obtain feature vector estimates, and calculate the deviation between the feature vector estimates and the true value of the feature vector in the test set. If all components of the deviation are less than a set threshold, the final RSAF model is obtained; otherwise, the RSAF model is trained again using the learning set;

[0013] Step 3: Real-time monitoring of the detection signal of the turbine pump, calculating the feature vector of the measured detection signal according to the definition of step 1, inputting the obtained feature vector into the RSAF model obtained in step 2 to obtain the estimated value of the feature vector of the measured detection signal, calculating the deviation of the estimated value of the feature vector of the measured detection signal and the calculated value of the feature vector of the measured detection signal, if the deviation has components exceeding the set threshold for 3 consecutive calculation periods, it is considered that the working state of the turbine pump has been abnormal.

[0014] Further, in step 3, the threshold is 10% of the corresponding component of the estimated value of the feature vector of the measured detection signal.

[0015] Further, in step 3, the mean + 3σ of the corresponding component of the deviation of the estimated value of the feature vector of the test set in step 2 and the true value of the feature vector is taken as the threshold, and σ is the variance of the deviation of the estimated value of the feature vector of the test set and the true value of the feature vector.

[0016] Further, it further includes step 4: classifying the fault according to the type of the component in the deviation that exceeds the threshold.

[0017] Further, in step 4, if only the components corresponding to the fundamental frequency and the 2nd harmonic exceed the threshold in the deviation, it is considered that an imbalance and misalignment fault occurs; if other features corresponding components exceed the threshold, it is considered that a rub-impact fault occurs.

[0018] Further, it further includes step 5: if it is judged that a rub-impact fault occurs, the sum of the squares of each component in the deviation is taken as an index to measure the severity of the rub-impact.

[0019] Further, in step 3, the situation of sensor abnormality in the process of real-time monitoring of the detection signal of the turbine pump is considered:

[0020] (1) If the speed sensor fails, only the amplitude features of I frequency points in the low frequency band are used for judgment; when calculating the estimated value, the contributions of the speed, the speed change value, the frequency doubling features, the sub-harmonic features, and the precession features of each channel are set to 0 to shield the influence of invalid features; when calculating the deviation, the deviation of the speed, the speed change value, the frequency doubling features, the sub-harmonic features, and the precession features of each channel are set to 0.

[0021] (2) If a detection channel fails, when calculating the estimated value, the contribution of the feature corresponding to the detection channel is set to 0; when calculating the deviation, the deviation of the feature corresponding to the detection channel is set to 0.

[0022] Further, in step 2, the modeling process includes the following steps:

[0023] Step 2.1: According to the detection data of the turbo pump in the normal working state, the steady state is judged; in each steady state, the feature vectors of the maximum value time and the minimum value time of each feature component are obtained as a coarse typical state set {SE};

[0024] Step 2.2: In the coarse typical state set {SE}, if the similarity of the feature vectors of two time points is higher than the set threshold, one of the feature vectors of the time points is removed, and finally the subset {SE1} is reserved.

[0025] Step 2.3: According to the detection data of the turbo pump in the normal working state, the variable working condition section is judged; the feature vectors of all variable working condition sections are combined to form a typical state set {TE}; in the typical state set {TE}, if the similarity of the feature vectors of two time points is higher than the set threshold, one of the feature vectors of the time points is removed, and finally the subset {TE1} is reserved.

[0026] Step 2.4: The {SE1} and {TE1} are combined together to form the RSAF model.

[0027] Further, in step 2, the test set is input into the RSAF model trained to obtain the estimated value of the feature vector:

[0028] For a feature vector in the test set, the similarity of the feature vector and all feature vectors in the RSAF model is calculated, the weight of each feature vector in the RSAF model is determined according to the similarity, and then the weighted sum of each feature vector in the RSAF model is obtained to obtain the estimated value of the feature vector.

[0029] Advantages

[0030] The high-speed turbo pump rotating and static rubbing real-time monitoring method based on the RSAF model can consider multiple rubbing fault features at the same time, generate a comprehensive evaluation index, effectively identify rubbing faults, and avoid missed alarms.

[0031] When the model is established, different normal reference values can be given to different working conditions instead of a single and absolute threshold value, so that different working conditions can be effectively included and the missed alarm in low working conditions or the false alarm in high working conditions due to an excessively large threshold value can be reduced.

[0032] The variable working condition state is considered when the model is established, and the threshold value of the variable working condition state can be given to monitor the steady state and the variable working condition section at the same time.

[0033] Additional aspects and advantages of the application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0034] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description, taken in conjunction with the following drawings of which:

[0035] Figure 1 : Time-domain waveform of horizontal direction vibration displacement amplitude of the turbine pump under normal condition;

[0036] Figure 2 : Turbine pump rotor axis trajectory diagram under normal condition;

[0037] Figure 3 : Time-domain waveform of horizontal direction vibration displacement amplitude of the turbine pump under rotating and static collision fault;

[0038] Figure 4 : Turbine pump rotor axis trajectory diagram under rotating and static collision fault;

[0039] Figure 5 : Comparison diagram of feature vector estimated value result and true value result of the normal signal;

[0040] Figure 6 : Total deviation amount diagram of feature vector estimated value and true value of the normal signal;

[0041] Figure 7 : Comparison diagram of feature vector estimated value result and true value result of the fault signal;

[0042] Figure 8 : Total deviation amount diagram of estimated value and true value of the fault signal. DETAILED DESCRIPTION

[0043] The present application is based on historical detection data under normal working condition of the turbine pump as sample data set, and constructs a specific feature vector to obtain the RSAF model representing the normal state of the turbine pump, aiming at the rich harmonic component features of the turbine pump detection data (rotation speed, vibration, displacement data). In the real-time monitoring process, the real-time features are estimated, and if the estimated result is inconsistent with the true result, it is considered that the state has changed or an abnormality has occurred.

[0044] Specifically, the following steps are included:

[0045] Step 1: Feature selection:

[0046] The amplitude signal is a direct representation of the turbine pump vibration, and the features in the amplitude signal are extracted as the basis for rub-impact fault analysis. Based on the dynamic mechanism, the amplitude representation of rub-impact fault mainly includes the fundamental frequency, the sub-harmonic (such as 0.5 times the frequency), and the harmonic (2, 3, and 4) corresponding amplitude. However, vibration is highly related to the working condition and assembly conditions, and there are great differences between different stages and different rotating speed conditions. Therefore, in this method, the rotating speed and the rotating speed change are added as two features to reflect the working condition, and the peak-to-peak value of the amplitude and the fundamental frequency corresponding amplitude are introduced to reflect the influence of the overall level of vibration under the current assembly condition, so that the data of different working conditions and states can be unified for analysis.

[0047] In addition, we found in the experiment that when severe rub-impact occurs, a larger radial force is easily generated, causing random energy peaks in the low-frequency band of the amplitude. Therefore, the amplitude features corresponding to the low-frequency band are added to the features for rub-impact monitoring. In addition, the appearance of counter-precession is also a significant feature when rub-impact occurs, so the 1st, 2nd, and 3rd order positive and negative precessions are added as detection features.

[0048] The final features involved in rub-impact monitoring include

[0049] [nt Δnt DF ch,1 … DF ch,k … DF ch,K P] 1×n

[0050] Where nt is the rotating speed of the turbine pump, Δnt is the rotating speed change, P is a vector composed of the 1st, 2nd, and 3rd order positive and negative precessions, and DF ch,k is the feature vector of the kth detection channel, k = 1, 2, …, K:

[0051] DF ch,k = [ppv k fb k fl 1，k … fl i,k … fl I,k f c,k f 2,k f 3,k f 4,k ]

[0052] ppv k in the above features is the peak-to-peak value of the kth detection channel, fb k is the fundamental frequency corresponding amplitude of the kth detection channel, fl 1，k … fl i,k … fl I,k is the amplitude corresponding to I frequency points in the low-frequency band of the kth detection channel; f c,k is the sub-harmonic corresponding amplitude of the kth detection channel, and f2,k , f 3,k , f 4,k is the amplitude corresponding to 2 times frequency, 3 times frequency, 4 times frequency in the kth detection channel.

[0053] By setting the above features, when comparing the monitoring state with the typical normal state, the weights of the typical cases under different rotating speeds and different rotating speed change rates can be distinguished, so that the steady state and the variable working condition are considered at the same time.

[0054] Step 2: Establish a model:

[0055] Obtain the detection data under the historical normal working state of the turbine pump, extract the feature vector of the detection data under the historical normal working state, divide the obtained feature vector into a learning set and a test set, train the RSAF model with the learning set as the representation of the normal state; and input the test set into the trained RSAF model to obtain the feature vector estimate value, calculate the deviation of the feature vector estimate value and the true value of the test set feature vector, if all components in the deviation are less than a set threshold, then the final RSAF model is obtained, otherwise the RSAF model is retrained with the learning set.

[0056] The modeling process is:

[0057] Step 2.1: According to the detection data under the historical normal working state of the turbine pump, determine the steady state; in each steady state, obtain the feature vector of the maximum value time and the minimum value time of each feature component as the coarse typical state set {SE};

[0058] Step 2.2: In the coarse typical state set {SE}, if the similarity of the feature vectors of two time points is higher than a set threshold, then one of the feature vectors is removed, and finally the subset {SE1} is retained;

[0059] Step 2.3: According to the detection data under the historical normal working state of the turbine pump, determine the variable working condition section; the feature vectors of all variable working condition sections form a typical state set {TE}; in the typical state set {TE}, if the similarity of the feature vectors of two time points is higher than a set threshold, then one of the feature vectors is removed, and finally the subset {TE1} is retained;

[0060] Step 2.4: Combine {SE1} and {TE1} together to form an RSAF model.

[0061] For the obtained RSAF model, the process of inputting the test set into the trained RSAF model to obtain the feature vector estimate value is:

[0062] For a certain feature vector in the test set, the similarity of the feature vector with all feature vectors in the RSAF model is calculated, the weight of each feature vector in the RSAF model is determined according to the similarity, and then the estimated value of the feature vector is obtained by weighted summation of each feature vector in the RSAF model.

[0063] Step 3: Real-time monitoring of the detection signal of the turbine pump, according to the definition of step 1, calculating the feature vector of the measured detection signal, inputting the obtained feature vector into the RSAF model obtained in step 2 to obtain the estimated value of the feature vector of the measured detection signal, calculating the deviation of the estimated value of the feature vector of the measured detection signal from the calculated value of the feature vector of the measured detection signal, and if the deviation has components exceeding the set threshold for 3 consecutive calculation periods, it is considered that the working state of the turbine pump has been abnormal.

[0064] Here, the threshold value can be determined according to experience, taking 10% of the corresponding component of the estimated value of the feature vector of the measured detection signal as the threshold value of the corresponding component; or taking the mean + 3σ of the corresponding component of the deviation of the estimated value of the feature vector of the test set from the true value of the feature vector as the threshold value, and σ is the variance of the deviation of the estimated value of the feature vector of the test set from the true value of the feature vector.

[0065] Step 4: After judging that the working state of the turbine pump has been abnormal, further classifying the fault according to the type of the component in the deviation that exceeds the threshold value. If only the components corresponding to the fundamental frequency and the 2nd harmonic exceed the threshold value in the deviation, it is considered that an imbalance or misalignment fault has occurred; if other characteristic components exceed the threshold value, it is considered that a rub-impact fault has occurred, and if a rub-impact fault is judged to have occurred, the sum of the squares of each component in the deviation is taken as an index to measure the severity of the rub-impact.

[0066] Of course, attention should be paid to the case of sensor abnormality in the monitoring of the detection signal. The present application adapts to the case of sensor abnormality in the following two ways:

[0067] (1) If the speed sensor fails, only the amplitude characteristics of I frequency points in the low frequency band are used for judgment; when calculating the estimated value, the contributions of the speed, speed change value, frequency doubling characteristics, sub-harmonic characteristics and precession characteristics of each channel are set to 0 to shield the influence of invalid characteristics and avoid false alarms; when calculating the deviation, the deviation of the speed, speed change value, frequency doubling characteristics, sub-harmonic characteristics and precession characteristics of each channel is set to 0.

[0068] (2) If a certain detection channel fails, the contribution of the feature corresponding to the detection channel is set to 0 when calculating the estimated value, and only the other channels are used for estimated value calculation; when calculating the deviation, the deviation of the feature corresponding to the detection channel is set to 0 to avoid false alarms caused by invalid features.

[0069] The embodiments of the present application are described in detail below, which are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application.

[0070] The present embodiment divides the rotor into disc, shaft segment, bearing and other units, establishes a four-pivot rotor model for fault simulation, applies fault force to the model by referring to real fault data, simulates the rubbing fault, and obtains the fault signal and the normal signal, as shown in Figures 1-4 .

[0071] Based on the normal signal, the RSAF model is established, and the comparison chart of the estimated value result of the model and the real value result of the normal signal is as shown in Figure 5 and Figure 6 It can be seen that different rotating speed conditions are well fitted, the normal condition can be correctly estimated, and the model is effective.

[0072] The comparison of the estimated value of the RSAF model and the real value result of the rubbing fault data is as shown in Figure 7 and Figure 8 It can be seen that from the beginning, the estimated value and the real value have a large deviation, but at this time, the amplitude corresponding to the fundamental frequency of the low speed condition is close to the amplitude corresponding to the fundamental frequency of the normal condition of the high speed condition, but at this time, the abnormality can be detected, if the same threshold is used for all conditions, it is difficult to achieve, thereby avoiding false alarm; and all conditions can be uniformly processed, and multiple features can be considered at the same time, which is convenient for the definition of threshold and the evaluation of severity.

[0073] It can be seen that the method can effectively identify the rubbing fault, and can uniformly process the data of different rotating speed conditions, effectively avoiding false alarm and missing alarm.

[0074] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as a limitation of the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments without departing from the principles and purposes of the present application within the scope of the present application.

Claims

1. A real-time monitoring method for high-speed turbine pump rotational-static friction based on the RSAF model, characterized by: The method comprises the following steps: Step 1: determining a feature vector participating in the rubbing monitoring as wherein, is the turbine pump rotation speed, is the rotation speed change amount, P is a vector composed of 1, 2, 3 order positive and negative precession, is the feature vector of the kth detection channel, k = 1, 2, …, K: Among the above features is the peak-to-peak value of the amplitude of the kth detection channel, is the fundamental frequency corresponding amplitude of the kth detection channel, is the I frequency point corresponding amplitude in the low frequency band spectrum of the kth detection channel; is the sub-harmonic corresponding amplitude of the kth detection channel, , , is the 2 times frequency, 3 times frequency, 4 times frequency corresponding amplitude in the kth detection channel; Step 2: obtaining detection data in a historical normal working state of the turbopump, extracting a feature vector of the detection data in the historical normal working state, dividing the obtained feature vector into a learning set and a test set, training an RSAF model by using the learning set, taking the RSAF model as a representation of a normal state, combining the test set and the trained RSAF model to obtain an estimated value of the feature vector, and calculating a deviation amount of the estimated value of the feature vector from a true value of the feature vector of the test set; if all components in the deviation amount are less than a set threshold, a final RSAF model is obtained, otherwise, the RSAF model is retrained by using the learning set. The process of establishing the RSAF model is as follows: Step 2.1: judging a steady state section according to the detection data in the historical normal working state of the turbopump; in each steady state section, a feature vector at a maximum value time and a minimum value time of each feature component is obtained as a coarse typical state set {SE}; Step 2.2: in the coarse typical state set {SE}, if the similarity of the feature vectors at two time points is higher than a set threshold, one of the feature vectors at the two time points is removed, and finally a subset {SE1} is reserved; Step 2.3: judging a variable working condition section according to the detection data in the historical normal working state of the turbopump; a typical state set {TE} is composed of the feature vectors of all variable working condition sections; in the typical state set {TE}, if the similarity of the feature vectors at two time points is higher than a set threshold, one of the feature vectors at the two time points is removed, and finally a subset {TE1} is reserved; Step 2.4: combining {SE1} and {TE1} to form the RSAF model; The process of obtaining the estimated value of the feature vector by combining the test set and the trained RSAF model is as follows: For a feature vector in the test set, the similarity of the feature vector with all feature vectors in the RSAF model is calculated, the weight of each feature vector in the RSAF model is determined according to the similarity, and the estimated value of the feature vector is obtained by weighted summation of each feature vector in the RSAF model Step 3: real-time monitoring of a detection signal of the turbopump, calculation of a feature vector of the measured detection signal according to the definition in step 1, obtaining of an estimated value of the feature vector of the measured detection signal by combining the feature vector of the measured detection signal and the final RSAF model obtained in step 2, calculation of a deviation amount of the estimated value of the feature vector of the measured detection signal from the calculated value of the feature vector of the measured detection signal, and determination that the working state of the turbopump is abnormal if a component of the deviation amount exceeds a set threshold for three continuous calculation periods.

2. The method of claim 1, wherein the method is a real-time monitoring method for the rotating and static rubbing of a high-speed turbine pump based on the RSAF model. In step 3, the threshold is 10% of the corresponding component of the estimated value of the feature vector of the measured detection signal.

3. The method of claim 1, wherein the method is a real-time monitoring method for rub-impact between rotating and stationary components of a high-speed turbopump based on the RSAF model. In step 3, the threshold is the mean value of the corresponding components of the deviation amount of the estimated value of the feature vector of the test set from the true value of the feature vector plus 3σ, and σ is the variance of the deviation amount of the estimated value of the feature vector of the test set from the true value of the feature vector.

4. The method of claim 1, wherein the method is a real-time monitoring method for rub-impact between rotating and stationary components of a high-speed turbopump based on the RSAF model. Step 4: classifying the fault according to the type of the component of the deviation amount exceeding the threshold.

5. The method of claim 4, wherein the method is based on the RSAF model. In step 4, if only the components corresponding to the fundamental frequency and the 2nd harmonic exceed the threshold in the deviation amount, it is considered that an unbalance or misalignment fault occurs; if other characteristic components exceed the threshold, it is considered that a rub fault occurs.

6. The method of claim 5, wherein the method is a real-time monitoring method for the rotating and stationary rubbing of a high-speed turbopump based on the RSAF model. Step 5 is further included: if it is judged that a rub fault occurs, the sum of squares of each component in the deviation amount is taken as an index for measuring the severity of the rub.

7. The method of claim 1, wherein the method is a real-time monitoring method for rub-impact between rotating and stationary components of a high-speed turbopump based on the RSAF model. In step 3, the abnormality of the sensor is considered in the process of real-time monitoring of the turbine pump detection signal: (1) If the speed sensor fails, only the amplitude characteristics of I frequency points in the low-frequency band spectrum are relied on for judgment; In the calculation of the estimated value, the contributions of the speed, the speed change value, the frequency multiplication characteristics, the sub-harmonic characteristics and the precession characteristics of each channel are set to 0 to shield the influence of invalid characteristics; in the calculation of the deviation amount, the deviation amounts of the speed, the speed change value, the frequency multiplication characteristics, the sub-harmonic characteristics and the precession characteristics of each channel are set to 0; (2) If a detection channel fails, in the calculation of the estimated value, the contribution of the characteristic corresponding to the detection channel is set to 0; in the calculation of the deviation amount, the deviation amount of the characteristic corresponding to the detection channel is set to 0.

Images