A quantitative diagnosis method for rub-impact faults based on harmonic phase

Abstract

The present invention discloses a quantitative diagnosis method for rubbing faults based on harmonic phase, which relates to the technical field of vibration-based rubbing fault diagnosis. This method realizes the quantitative diagnosis of rubbing faults based on the processing of harmonic phase, has strong directivity in diagnosing rubbing faults, and enhances the feasibility of practical application. The technical solution of the present invention includes the following steps: extracting the harmonic phase in the vibration data under normal conditions, constructing an optimal harmonic number selection criterion, and selecting the optimal harmonic number. Calculating the fault feature index by fusing multi-harmonic phases, and then performing multi-component fusion to form the normal state threshold of the system. Combining the normal state phase processing method and the optimal harmonic number selection result, extracting and preprocessing the harmonic phase in the vibration data under the pending state. Based on the index result of the normal state, calculating the fault diagnosis index, and comparing the numerical value of the fault diagnosis index with the normal state threshold to determine the state of the system.

Description

Technical Field

[0001] The present invention relates to the technical field of rubbing fault diagnosis based on vibration, and particularly to a quantitative diagnosis method for rubbing faults based on harmonic phase. Background Art

[0002] The rotor system is the core of high-end rotating machinery, and the ability to accurately evaluate its state and diagnose faults is the key to ensuring the normal operation of the equipment. To improve the performance and efficiency of the equipment, a rotor system with a large number of blades is placed in a casing with a very small clearance, which easily causes contact between the blades and the casing, resulting in rubbing faults and affecting the normal operation of the equipment. To ensure the safety of high-end rotating machinery and the stability of production, it is extremely important to study the rubbing fault mechanism and then construct an index recognition system state.

[0003] At present, the latest research result for rubbing fault diagnosis is to construct an index based on the system's nonlinear output frequency response functions (NOFRFs) to judge the system state. However, this method has obvious limitations in practical applications.

[0004] Traditional methods for diagnosing rubbing faults based on the root mean square (RMS) of system vibration and the single change in the amplitude of some harmonics (the decrease in the amplitude of the rotor fundamental frequency) have the problem of weak directivity because, in addition to rubbing faults, other typical rotor faults such as rotor imbalance will also cause changes in the above characteristic values.

[0005] Although the current research-formed rubbing fault diagnosis method based on the nonlinear output frequency response function NOFRFs avoids the influence of other typical rotor faults under stable working conditions, its practicality in the actual fault diagnosis environment is doubtful. On the one hand, on the premise of known input, two test runs with the same waveform but different amplitudes are required to obtain the values of NOFRFs, which is almost impossible in the actual application process. On the other hand, the prerequisite of known system input is too harsh, and in actual situations, the input excitation of the system is almost unobtainable. Therefore, this results in the difficulty of effectively applying NOFRFs in the diagnosis of actual operating faults (such as rubbing).

[0006] Therefore, the existing solutions cannot solve the problems of weak directivity in diagnosing rubbing faults and poor practicality. Summary of the Invention

[0007] In view of this, the present invention provides a quantitative diagnosis method for rubbing faults based on harmonic phase. This method realizes the quantitative diagnosis of rubbing faults based on the processing of harmonic phase, has strong directivity when diagnosing rubbing faults, and enhances the feasibility of practical applications.

[0008] To achieve the above object, the technical solution of the present invention includes the following steps:

[0009] S1. Extract the harmonic phase in the vibration data in the normal state.

[0010] S2. Based on the harmonic phase extracted in the normal state, construct an optimal harmonic number selection criterion, and select the optimal harmonic number that can take into account both index sensitivity and robustness.

[0011] S3. Based on the optimal harmonic number, calculate the fault feature index by fusing multiple harmonic phases. Based on the fault feature index, perform multi-component fusion to form the normal state threshold of the system.

[0012] S4. Combine the normal state phase processing method and the optimal harmonic number selection result, extract the harmonic phase in the vibration data in the pending state and perform preprocessing.

[0013] S5. Based on the index result in the normal state, calculate the fault diagnosis index, and compare the numerical value of the fault diagnosis index with the normal state threshold to determine the state of the system.

[0014] Further, in S1, to extract the harmonic phase in the vibration data in the normal state, the specific steps are as follows:

[0015] First, perform certain preprocessing on the harmonic phase in the normal state:

[0016] When , Perform the following preprocessing:

[0017]

[0018] wherein, represents the harmonic phase of the nth order and the kth segment of data; represents the harmonic phase of the nth order and the (k + 1)th segment of data, and the order represents the multiple of the harmonic compared to the rotational frequency;

[0019] When , perform the following preprocessing:

[0020]

[0021] where is the maximum change amount of the harmonic phase of the nth order, is the harmonic phase of the nth order.

[0022] Finally, subtract the mean value of each harmonic from the harmonic phase and then add 180° to obtain the extraction result of the harmonic phase.

[0023] Further, S2 is specifically as follows:

[0024] The system is under stable excitation. In the normal state, the phase of the harmonic is stable. If fluctuations occur in the system under normal conditions, it is due to the influence of noise.

[0025] Based on the vibration data under normal conditions, fully considering the role of noise, extract the harmonic phase therein. Based on the variance of the harmonic phase , where N is the upper limit of the selection of the number of harmonics, construct the optimal criterion S for selecting the number of harmonics, and its expression is:

[0026]

[0027] where respectively represent variance and mean; when S is the largest, the value of N is the optimal number of harmonics, that is, this number of harmonics is considered the optimal number of harmonics.

[0028] Further, in S3, based on the optimal number of harmonics, calculate the fault feature index that fuses multiple harmonic phases. Specifically, the fault feature index that fuses multiple harmonic phases is P:

[0029]

[0030] where respectively represent the harmonic order number, exponential constant, and the selected optimal number of harmonics.

[0031] Further, in S3, based on the fault feature index, perform multi-component fusion to form the normal state threshold of the system. Specifically:

[0032] In view of the expression of KL divergence, through the normal state data, amplify the fault difference, and construct the index PKL:

[0033]

[0034] where respectively represent the mean value of the index P value under the normal state of the system, the index P value under the pending state; under the normal state, after calculating the index P value for each harmonic phase, its value is , and its mean value is , and thus calculate the PKL under the normal state.

[0035] If under the normal state, the PKL calculated from each harmonic phase of the system is normal, then the threshold PKLH is set as the maximum value of the PKL calculated for each harmonic under different constants c:

[0036]

[0037] At this point, the processing of the normal state data is completed, and the system normal state threshold is PKLH.

[0038] Further, S4 is specifically:

[0039] The phase of the vibration signal under the condition to be determined is extracted. On the basis of the normal state phase processing, combined with the phase mean under the normal state, the harmonic phase is processed in stages, including the following three stages of processing:

[0040] Stage (1): If the absolute difference between the harmonic phase and its mean exceeds , it is determined to be an oscillation in a normal phase state.

[0041] Stage (2): If the absolute difference between the harmonic phase and its mean exceeds 180°, the phase is adjusted to minimize the absolute difference between it and the mean:

[0042] Stage (3): Further setting of upper limits on change , which specifies the maximum range of phase variation to prevent processing steps from affecting each other.

[0043] Furthermore, in S4, at stage (1), the absolute difference between the harmonic phase and its mean exceeds ,Right now , it is determined to be an oscillation in a normal phase state, and the following preprocessing is performed:

[0044]

[0045] in, for The mean of Indicates the harmonic phase of the nth order and kth segment of data. The order indicates the multiple of the harmonic compared to the rotation frequency.

[0046] Furthermore, in S4, in stage (2), if the absolute difference between the harmonic phase and its mean exceeds 180°, that is, , then adjust the phase to minimize its absolute difference with the mean:

[0047]

[0048] in They represent the order and the time sequence number of the data segment respectively.

[0049] Furthermore, in S4, stage (3) , further set the upper limit of change , which specifies the maximum range of phase variation to prevent processing steps from affecting each other.

[0050]

[0051] So far, the preprocessing of the harmonic phase in the vibration data in the to-be-determined state has been completed.

[0052] Further, S5 is specifically as follows:

[0053] First, calculate the value of index P in the to-be-determined state ; Based on the mean value of the index P values in the normal state , calculate the characteristic index PKLA:

[0054]

[0055] Based on PKLA, further adopt the maximum value PKLM of it under different constants c as the final fault determination basis:

[0056]

[0057] PKLM and its difference from the normal state threshold PKLH are used to determine the system state.

[0058] Beneficial effects:

[0059] 1: A quantitative diagnosis method for rubbing faults based on harmonic phase provided by the present invention. The harmonic phase becomes an effective index input parameter through preprocessing: Since the phase has a range of [0, 360°) and a base value, it is difficult to distinguish between normal and fault states (the advantage of the amplitude is that it changes significantly, and in the spectrum, the part without excitation is 0, and the change can be significantly observed). The present invention effectively extracts the phase. Through staged preprocessing, based on the unified mean value and change trend of the phase, it lays a foundation for constructing an index to determine the system state based on phase changes. The feasibility of the practical application of the present invention is enhanced: The present invention can be used in real time in the actual equipment status monitoring, without the need for multiple start-ups and shut-downs. Although normal state data is required as a basis, it conforms to the current data reserve situation (few fault cases and a large number of normal data).

[0060] 2: The present invention sets the determination thresholds for the system state and the fault state. Most of the previous technologies only set sensitive indicators, which change significantly when a fault occurs, but do not specifically indicate how much the change amount is when the system enters the fault state. This method sets the threshold for the normal state and can quantitatively diagnose the fault degree. Description of the Drawings

[0061] Figure 1 It is a flowchart of a quantitative diagnosis method for rubbing faults based on harmonic phase provided by the present invention;

[0062] Figure 2 It is the effect diagram of the normal state phase processing in the embodiment of the present invention;

[0063] Figure 3 This is the application effect diagram of the rub-impact fault diagnosis in the embodiment of the present invention. Specific embodiments

[0064] The present invention will be described in detail below with reference to the accompanying drawings and by way of examples.

[0065] The present invention provides a quantitative diagnosis method for rub-impact faults based on harmonic phase. The principle of this method is that the influence of rub-impact faults on the system can be transformed into the influence on the inherent characteristics of the system (such as stiffness, damping, etc.). Combining with the expression of the vibration lag angle :

[0066] (1)

[0067] Among them, respectively represent the system mass, the system stiffness, the additional stiffness generated by the rub-impact fault, the system damping, the additional damping generated by the rub-impact fault, and the harmonic frequency. The vibration lag angle is the harmonic phase.

[0068] It can be seen from equation (1) that the rub-impact fault will cause a change in the vibration lag angle of the system, while typical rotor faults will not affect the lag angle of the system. Therefore, a fault characteristic index can be constructed based on the lag angle of the harmonic to diagnose the rub-impact fault directionally.

[0069] The present invention provides a rub-impact fault diagnosis method based on harmonic phase, which can form a normal state threshold based on a large amount of existing normal state data, and can quantitatively diagnose the rub-impact fault without fault state data. The obtaining process is as Figure 1 shown. This method provides technical guidance for the healthy operation of mechanical equipment. Compared with the existing rub-impact fault diagnosis methods, this method can be directly used in the actual rub-impact fault diagnosis of rotating equipment, without repeated test runs, with strong practicability and a wider application range.

[0070] The main steps for obtaining and using the diagnosis method are as follows:

[0071] S1. Extract the harmonic phase from the vibration data in the normal state;

[0072] Furthermore, in S1, certain preprocessing is required for the harmonic phase in the normal state, mainly excluding two aspects of factors: (1) Set the angle change range of the phase to [0, 360°). Due to the fluctuation of the actual phase measurement and the influence of the initial phase angle, there will be a problem that the harmonic phase fluctuating near the 0° range will have significant oscillations. To prevent such factors from affecting, set an angle threshold , if the change in the harmonic phase of the latter segment of data is greater than or equal to that of the previous segment of data , it is considered that this is a normal phase fluctuation. The data segment has the rotation of the rotating shaft for a set number of revolutions as one end of the data, and the set number of revolutions is at least fifty; (2) Not all harmonic phases are generated by a stable exciting force. Some harmonic phases are generated by noise excitation, and these harmonic phases need to be discarded. Set an angular threshold , if the maximum change in the harmonic phase of the nth order exceeds the threshold, then discard this phase.

[0073] Regarding (1), , and represent the harmonic phases of the nth order of the kth and k + 1th segment data. The order represents the multiple of the harmonic compared to the rotation frequency, perform the following preprocessing:

[0074] (2)

[0075] Regarding (2), , the maximum change in the harmonic phase of the nth order , perform the following preprocessing:

[0076] (3)

[0077] is the harmonic phase of the nth order;

[0078] Finally, in order to facilitate the subsequent setting of the state threshold, adjust the phase mean of all harmonics to 180°. Subtract the mean of each harmonic from the harmonic phase and then add 180° to obtain the extraction result.

[0079] The effect of this preprocessing method is as Figure 2 shown.

[0080] S2. Based on the harmonic phases extracted in the normal state, construct an optimal harmonic number selection criterion to select the harmonic number that can simultaneously balance the index sensitivity and robustness;

[0081] Furthermore, in S2, since it is impossible to fuse all harmonic phases indefinitely when constructing an index by fusing multiple harmonic phases. On the one hand, not all phases can well reflect faults; on the other hand, extracting a large number of phases and fusing them to calculate the index will consume a lot of time, thereby affecting the real-time performance of the method.

[0082] Assume that the system is under stable excitation. Then, in the normal state, the harmonic phases are stable, and their fluctuations are due to the influence of noise.

[0083] By using the vibration data in the normal state, fully considering the role of noise, extracting the harmonic phase therein, and based on the variance of the harmonic phase , where N is the upper limit of the selection of the number of harmonics, an optimal harmonic number selection criterion S is constructed, and its expression is:

[0084] (4)

[0085] where respectively represent the variance and the mean value. When S is the largest, the value of N is the optimal harmonic number, that is, this harmonic number is considered the optimal harmonic number.

[0086] According to the expression, the criterion S is constructed by the variance and the mean value of the harmonic phase variance. The construction of this expression mainly considers the following two aspects: (1) The index cannot be interfered by too much noise. Therefore, the harmonic phase variance reflects the influence level of noise, and its mean value is small, which means that the index is less interfered by noise; (2) The index needs to be sensitive to faults. Taking noise as the judgment factor, if harmonics can sensitively reflect the influence of noise, then they can also sensitively reflect faults. Therefore, the criterion S needs to include a part of sensitive harmonic phases, which will lead to being large.

[0087] In summary, the present invention constructs the optimal harmonic number selection criterion S and, based on this, selects the harmonic phase used for the index.

[0088] S3. Based on the optimal harmonic number, calculate the fault feature index by fusing multiple harmonic phases, and based on the index, perform multi-component fusion to form the normal state threshold of the system;

[0089] Furthermore, in S3, it is necessary to construct a fault feature index and set a threshold. First, construct the index P:

[0090] (5)

[0091] where respectively represent the harmonic order number, the exponential constant (commonly used), and the selected optimal harmonic number.

[0092] At the same time, in view of the expression of the KL divergence, by using the normal state data, the fault difference is amplified, and the index PKL is constructed:

[0093] (6)

[0094] where respectively represent the mean value of the index P value in the normal state of the system and the index P value in the pending state. In the normal state, after calculating the index P value for each harmonic phase, its value is , and its mean value is , thus the PKL under normal conditions can be calculated.

[0095] Under normal conditions, the PKL calculated for each harmonic phase of the system is normal. Then the threshold PKLH is set as the maximum value of the PKL calculated for each harmonic under different constants c:

[0096] (7)

[0097] In summary, S1, S2, and S3 have completed the processing of normal state data. The normal state threshold of the system is PKLH.

[0098] S4. Combine the normal state phase processing method and the optimal harmonic number selection result to extract the harmonic phase from the vibration data in the pending state and perform preprocessing.

[0099] Furthermore, in S4, the phase of the vibration signal in the pending state is extracted. Based on the normal state phase processing and combined with the phase mean value in the normal state, the harmonic phase is processed in stages. The rejection of the noise excitation phase in the normal state is adopted in the present invention:

[0100] (1) If the absolute difference between the harmonic phase and its mean value exceeds , it is determined as an oscillation in the normal phase state, and its processing method is

[0101] (8)

[0102] Among them, is the mean value of; represents the harmonic phase of the nth order and the kth segment of data. The order represents the multiple of the harmonic compared to the rotational frequency.

[0103] (2) If the absolute difference between the harmonic phase and its mean value exceeds 180°, the present invention adjusts the phase to minimize its absolute difference from the mean value:

[0104] (9)

[0105] Among them respectively represent the order number and the time serial number of the data segment.

[0106] (3) . Step (2) makes the difference between all phases and the harmonic mean value not greater than 180°. The present invention further sets the change upper limit , which stipulates the maximum change range of the phase to prevent mutual influence between processing steps.

[0107] (10)

[0108] In summary, the technology of the present invention has completed the preprocessing of the harmonic phase in a specific state.

[0109] S5. Based on the index results in the normal state, calculate the fault diagnosis index, and compare the index value with the normal state threshold to determine the state of the system.

[0110] Further, in S5, first calculate the value of index P in the specific state based on the fault feature index formula (5) .

[0111] Based on the mean value of the value of index P in the normal state , the present invention calculates the characteristic index PKLA:

[0112] (11)

[0113] On the basis of PKLA, further adopt the maximum value PKLM of it under different constants c as the final fault determination basis:

[0114] (12)

[0115] It should be noted the difference between PKLM and its normal state threshold PKLH. PKLH is the sum of the maximum values of each PKL under different constants c, while PKLM is the maximum value of the sum results of PKL under the same constant c. Therefore, PKLH must be greater than or equal to the normal PKLM in the specific state. When PKLM exceeds PKLH, it indicates that the system has entered an abnormal state.

[0116] The effect diagram of this method in practical application is as Figure 3 shown: The horizontal line in the figure is the normal state threshold, and below the red line is the normal state. It can be seen that the technology of the present invention not only quantitatively identifies the development of the fault, but also determines the stage when the system returns to the normal state after the rubbing fault ends.

[0117] In summary, the above is only the preferred embodiment of the present invention, and is not used to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A quantitative diagnosis method for rubbing fault based on harmonic phase, characterized in that: The steps include: S1. Extracting the harmonic phase from the vibration data under normal conditions; S2. Based on the harmonic phase extracted from the normal state, the optimal harmonic number selection criteria are constructed to select the optimal harmonic number that can take into account both the sensitivity and robustness of the indicator; S2, specifically: The system is stably excited. Under normal conditions, the phase of the harmonics is stable. If the system fluctuates under normal conditions, it is due to the influence of noise. Through the vibration data under normal conditions, the effect of noise is fully considered, the harmonic phase is extracted, and the variance of the harmonic phase is calculated based on the , N is the upper limit of the harmonic number, and the optimal harmonic number selection criterion S is constructed, and its expression is: in represent the variance and mean respectively; when S is the largest, the value of N is the optimal harmonic number, that is, the harmonic number is considered to be the optimal harmonic number; S3. Based on the optimal harmonic number, calculate the fault characteristic index of the fused multi-harmonic phase, and perform multi-component fusion based on the fault characteristic index to form a system normal state threshold; S3 is specifically, the fault characteristic index of the fused multi-harmonic phase is P: in Respectively represent the harmonic order, exponential constant and the optimal harmonic number selected; is the harmonic phase; Given the expression of KL divergence, the indicator PKL is constructed by amplifying the fault difference through normal state data: in They represent the mean value of the index P under the normal state of the system and the index P value under the pending state respectively; under the normal state, after the index P value is calculated, the value of each harmonic phase is , whose mean is , from which the PKL under normal conditions is calculated; If the PKL calculated by the phase of each harmonic of the system is normal under normal conditions, then the threshold PKLH setting is the maximum value of the PKL calculated by using each harmonic under different constants c: At this point, the processing of the normal state data is completed, and the system normal state threshold is PKLH; S4, combining the normal state phase processing method and the optimal harmonic number selection result, extracting the harmonic phase in the vibration data under the pending state and performing preprocessing; S5. Based on the indicator results of the normal state, calculate the fault diagnosis indicator, and compare the fault diagnosis indicator value with the normal state threshold to determine the state of the system; S5, specifically: First, calculate the value of the indicator P in the pending state ; Based on the mean value of the indicator P under normal conditions , calculate the characteristic index PKLA: On the basis of PKLA, its maximum value PKLM under different constants c is further adopted as the final fault judgment basis: PKLM and its difference from the normal state threshold PKLH are used to determine the system state.

2. A method for quantitative diagnosis of rubbing faults based on harmonic phase according to claim 1, characterized in that: In S1, the harmonic phase in the vibration data under normal conditions is extracted, and the specific steps are as follows: First, preprocess the harmonic phase under normal conditions: when hour, Perform the following preprocessing: in, Indicates the harmonic phase of the nth order and kth segment of data; Indicates the harmonic phase of the nth order, k+1th segment of data. The order indicates the multiple of the harmonic compared to the rotation frequency; when When , the following preprocessing is performed: in is the maximum change in the phase of the nth order harmonic, is the harmonic phase of the nth order; Finally, the harmonic phase is subtracted from the mean value of each harmonic and then added with 180° to obtain the extracted result of the harmonic phase.

3. A method for quantitative diagnosis of rubbing fault based on harmonic phase as claimed in claim 1, characterized in that: The S4 is specifically: The phase of the vibration signal under the condition to be determined is extracted. On the basis of the normal state phase processing, combined with the phase mean under the normal state, the harmonic phase is processed in stages, including the following three stages of processing: Stage (1): If the absolute difference between the harmonic phase and its mean exceeds , it is determined to be an oscillation in a normal phase state; Stage (2): If the absolute difference between the harmonic phase and its mean exceeds 180°, the phase is adjusted to minimize the absolute difference between it and the mean: Stage (3): Further setting of upper limits on change , which specifies the maximum range of phase variation to prevent processing steps from affecting each other.

4. A method for quantitative diagnosis of rubbing fault based on harmonic phase as claimed in claim 3, characterized in that: In S4, in stage (1), the absolute difference between the harmonic phase and its mean exceeds ,Right now , it is determined to be an oscillation in a normal phase state, and the following preprocessing is performed: in, for The mean of Indicates the harmonic phase of the nth order and kth segment of data. The order indicates the multiple of the harmonic compared to the rotation frequency.

5. A method for quantitative diagnosis of rubbing fault based on harmonic phase as claimed in claim 3, characterized in that: In S4, in stage (2), if the absolute difference between the harmonic phase and its mean value exceeds 180°, that is, , then adjust the phase to minimize its absolute difference with the mean: in They represent the order and the time sequence number of the data segment respectively.

6. A method for quantitative diagnosis of rubbing fault based on harmonic phase as claimed in claim 3, characterized in that: In S4, the stage (3) , further set the upper limit of change , which specifies the maximum range of phase variation to prevent the processing steps from affecting each other; So far, the preprocessing of the harmonic phase in the vibration data under the predetermined state is completed.

Images