Vibration diagnosis method and system for abrasion of main shaft bearing of turboshaft engine

Abstract

The invention relates to the technical field of bearing wear detection, and discloses a vibration diagnosis method and system for wear of a main shaft bearing of a turboshaft engine, and the method comprises the steps: collecting a vibration signal of the main shaft bearing of the turboshaft engine; constructing a cumulant projection matrix of the vibration signals; performing dimension reduction on the cumulant projection matrix to obtain a feature matrix of the vibration signal; judging whether the bearing is worn or not based on the characteristic matrix of the vibration signal; if the bearing is abraded, determining the abrasion position, specifically, calculating the delay inequality of an impact event in the vibration signal based on the characteristic matrix of the vibration signal; constructing a delay inequality feature vector of the impact event; and matching the time delay inequality feature vector with a preset time delay inequality map, and determining a wear position according to a matching result. According to the method, the wear judgment accuracy and the positioning precision can be improved, and the method has good adaptability under different working conditions.

Description

Technical Field

[0001] This application relates to the technical field of bearing wear detection, and particularly to a vibration diagnosis method and system for the wear of the main shaft bearing of a turboshaft engine. Background Art

[0002] In current industry practices, for the wear monitoring of the main shaft bearings of turboshaft engines, a variety of technical means have been developed. Each technology provides information for bearing condition assessment to a certain extent, but each also has obvious limitations. Traditional vibration monitoring technologies collect vibration signals generated during the operation of the main shaft bearings through vibration sensors installed at key parts of the engine, and judge the operating state of the bearings based on characteristics such as the amplitude, frequency, and phase of the signals. When the bearings are worn, their vibration characteristics will change. However, the actual operating environment of a turboshaft engine is complex, and the vibration signals are easily affected by various factors such as the vibration of other components, airflow pulsation, and electromagnetic interference, resulting in obvious background noise, making the weak characteristic signals related to bearing wear be submerged and difficult to accurately extract and identify. Existing vibration monitoring technologies have low sensitivity to early minor wear and often can only be detected when the wear has developed to a certain extent and the vibration characteristics have changed significantly, which is not conducive to early fault warning and maintenance. In addition, the rotational speed of a turboshaft engine often changes during operation. When existing technologies identify the wear location, it is very difficult to eliminate the influence of rotational speed fluctuations on the monitoring results. The signal changes caused by the rotational speed change will interfere with the judgment of the wear location characteristics, making the wear identification and the judgment results of the wear location inaccurate and unstable under different rotational speed conditions.

[0003] For example, Chinese Patent with the authorized announcement number CN110646202B discloses a method and device for detecting the wear of a bearing. The method includes: obtaining a time-domain signal representing the change in the vibration magnitude of the bearing in its radial direction over time; converting the time-domain signal into a frequency-domain signal representing the distribution of the vibration of the bearing in its radial direction in the frequency domain; and using the frequency-domain signal and a plurality of stored frequency-domain signal segments for detecting the wear of the bearing to judge whether the bearing is worn, wherein each of the plurality of frequency-domain signal segments has the characteristics of one of the various wears that the bearing may have. This method and device can improve the detection accuracy of bearing wear.

[0004] As disclosed in the Chinese patent application with the publication number CN113670615A, a vibration test method and system for bearing units belong to the technical field of bearing detection. The vibration of the bearing unit to be tested is detected, its vibration signal is obtained, and the vibration signal is converted into a digital signal. The digital signal is decomposed by VMD to obtain multiple band-limited intrinsic mode functions, and the power spectral energy of each band-limited intrinsic mode function is calculated. When the order of the band-limited intrinsic mode function with the largest power spectral energy among the band-limited intrinsic mode functions is greater than the set order, the bearing unit to be tested has wear-type defects; or, if the energy of one of the band-limited intrinsic mode functions is greater than the set threshold among the band-limited intrinsic mode functions, the bearing unit to be tested has wear-type defects. By adopting this technical solution, the bearing units can be detected quickly and in batches, reducing the detection time.

[0005] The above existing technologies all have the problems raised in this background technology: under different rotational speed conditions, the results of wear identification and wear position judgment are inaccurate and unstable.

[0006] The information disclosed in this background technology section is only intended to increase the understanding of the overall background of this application, and should not be regarded as an admission or any form of suggestion that this information constitutes the prior art already known to those of ordinary skill in the art. Summary of the Invention

[0007] The technical problem to be solved by this application is to overcome the defects of the prior art, and provide a vibration diagnosis method and system for the wear of the main shaft bearing of a turboshaft engine, improving the accuracy of wear judgment, the positioning accuracy, and the adaptability under different working conditions.

[0008] To solve the above technical problems, this application provides the following technical solutions:

[0009] On the one hand, this application provides a vibration diagnosis method for the wear of the main shaft bearing of a turboshaft engine, including the following steps:

[0010] Collect the vibration signal of the main shaft bearing of the turboshaft engine; construct the cumulant projection matrix of the vibration signal;

[0011] Use truncated singular value decomposition to reduce the dimension of the cumulant projection matrix to obtain the feature matrix of the vibration signal;

[0012] Based on the feature matrix of the vibration signal, judge whether the bearing has wear;

[0013] If the bearing has wear, determine the wear position, specifically including:

[0014] Based on the feature matrix of the vibration signal, calculate the time delay difference of the impact events in the vibration signal;

[0015] Construct the time delay difference feature vector of the impact events;

[0016] Match the time delay difference feature vector with a preset time delay difference spectrum, and determine the wear position according to the matching result.

[0017] As a preferred solution of the vibration diagnosis method for the wear of the main shaft bearing of the turboshaft engine described in this application, wherein: the vibration signal is collected based on a vibration sensor; the cumulant projection matrix is the diagonal slice projection matrix of the third-order cumulant of the vibration signal; the method for constructing the cumulant projection matrix of the vibration signal is as follows:

[0018] Calculate the third-order cumulant of the vibration signal at different time delays;

[0019] Extract the diagonal slice data from the third-order cumulant;

[0020] Arrange the diagonal slice data into the cumulant projection matrix of the vibration signal.

[0021] As a preferred solution of the vibration diagnosis method for the wear of the main shaft bearing of the turboshaft engine described in this application, wherein: the truncated singular value decomposition is used to reduce the dimension of the cumulant projection matrix to obtain the feature matrix of the vibration signal, which specifically includes:

[0022] Perform singular value decomposition on the cumulant projection matrix to obtain the singular value diagonal matrix and the left singular vector of the cumulant projection matrix;

[0023] Determine the number of truncated main singular values, denoted as k; extract the largest k singular values from the singular value diagonal matrix;

[0024] Arrange the left singular vectors corresponding to the largest k singular values into the feature matrix of the vibration signal; any column in the feature matrix corresponds to a left singular vector.

[0025] As a preferred solution of the vibration diagnosis method for the wear of the main shaft bearing of the turboshaft engine described in this application, wherein: based on the feature matrix of the vibration signal, determine whether the bearing is worn, which specifically includes:

[0026] Based on the feature matrix of the vibration signal, extract the feature parameters of the vibration signal; the feature parameters include at least one of amplitude, frequency, kurtosis, and impulse factor;

[0027] Obtain the real-time working parameters of the main shaft of the turboshaft engine; the real-time working parameters include at least one of rotational speed, load, and cumulative operating time;

[0028] After normalizing and encoding the feature parameters of the vibration signal and the real-time working parameters of the main shaft of the turboshaft engine, input them into the trained wear prediction model to obtain the judgment result of whether there is wear;

[0029] The wear prediction model is any one of a support vector machine, a random forest model, and a gradient boosting decision tree, and the output is a binary judgment result of whether there is wear in the bearing or there is no wear in the bearing.

[0030] As a preferred solution of the vibration diagnosis method for the wear of the main shaft bearing of the turboshaft engine described in the present application, wherein: the time delay difference is the time difference between different impact events detected by the sensor in the feature matrix of the vibration signal; calculating the time delay difference of the impact events in the vibration signal specifically includes:

[0031] Perform Hilbert envelope demodulation on each column of left singular vectors in the feature matrix of the vibration signal to obtain an impact pulse sequence corresponding to each column of left singular vectors;

[0032] Calculate the cross-correlation coefficients between any two impact pulse sequences at different lag times and form a cross-correlation coefficient sequence corresponding to the two impact pulse sequences;

[0033] Extract the lag time corresponding to the maximum value in the cross-correlation coefficient sequence as the time delay difference between the impact events corresponding to the two impact pulse sequences.

[0034] As a preferred solution of the vibration diagnosis method for the wear of the main shaft bearing of the turboshaft engine described in the present application, wherein: constructing the time delay difference feature vector of the impact event specifically includes:

[0035] Construct a time delay difference matrix of the impact events in the vibration signal; the time delay difference matrix is a strictly upper triangular matrix, and the element in the i-th row and j-th column of the time delay difference matrix is the time delay difference between the i-th impact event and the j-th impact event in the vibration signal; wherein, i is greater than j, the value range of i is 2, 3,..., m, m is the number of columns of the feature matrix of the vibration signal; the value range of j is 1, 2,..., m-1;

[0036] Extract the non-zero elements in the time delay difference matrix row by row and arrange them into a time delay difference feature vector of the impact event.

[0037] As a preferred solution of the vibration diagnosis method for the wear of the main shaft bearing of the turboshaft engine described in the present application, wherein: the time delay difference spectrum includes a reference time delay difference feature vector corresponding to each known wear position; the method for constructing the time delay difference spectrum specifically includes: collecting historical vibration signals of bearings at known wear positions, calculating the time delay difference of the impact events in the historical vibration signals and constructing a time delay difference feature vector as the reference time delay difference feature vector corresponding to the wear position;

[0038] Matching the time delay difference feature vector with a preset time delay difference atlas, and determining the wear position according to the matching result specifically includes: calculating the similarity between the time delay difference feature vector and each reference time delay difference feature vector in the time delay difference atlas respectively; the wear position corresponding to the reference time delay difference feature vector with the highest similarity to the time delay difference feature vector is the current wear position of the main shaft bearing of the turboshaft engine.

[0039] As a preferred scheme of the vibration diagnosis method for the wear of the main shaft bearing of the turboshaft engine described in this application, wherein: the method for calculating the similarity between the time delay difference feature vector and any reference time delay difference feature vector in the time delay difference atlas is as follows:

[0040] Normalize the time delay difference feature vector and the reference time delay difference feature vector respectively;

[0041] Construct a distance matrix; the element in the p-th row and q-th column of the distance matrix is denoted as ; is the Euclidean distance between the p-th element in the time delay difference feature vector and the q-th element in the reference time delay difference feature vector; the value range of p is 1, 2,..., ; is the length of the time delay difference feature vector; the value range of q is 1, 2,..., ; is the length of the reference time delay difference feature vector;

[0042] Initialize the cumulative distance matrix; the dimension of the cumulative distance matrix is the same as that of the distance matrix;

[0043] Iteratively calculate the cumulative distance value based on the distance matrix, and fill the elements of the cumulative distance matrix row by row based on the cumulative distance value;

[0044] Extract the cumulative distance value in the lower right corner of the cumulative distance matrix as the similarity between the time delay difference feature vector and the reference time delay difference feature vector.

[0045] As a preferred scheme of the vibration diagnosis method for the wear of the main shaft bearing of the turboshaft engine described in this application, wherein: iteratively calculating the cumulative distance value based on the distance matrix, and filling the elements of the cumulative distance matrix row by row based on the cumulative distance value specifically includes:

[0046] Let the element in the p-th row and q-th column of the cumulative distance matrix be ; fill the elements of the cumulative distance matrix row by row, and fill the elements in each row in the order from left to right; for the element with any values of p and q , if there exists , , in the cumulative distance matrix, then The value is and , , the sum of the minimum values among; otherwise, The value is .

[0047] In a second aspect, the present application provides a vibration diagnosis system for wear of the main shaft bearing of a turboshaft engine, including a data acquisition module, a data processing module, a wear diagnosis module, and a wear location module; wherein:

[0048] The data acquisition module is used to collect the vibration signals of the main shaft bearing of the turboshaft engine and collect the real-time working parameters of the main shaft of the turboshaft engine, including rotational speed, load, and cumulative operating time;

[0049] The data processing module is used to construct a feature matrix of the vibration signal and extract feature parameters of the vibration signal based on the feature matrix, including amplitude, frequency, kurtosis, and impulse factor;

[0050] The wear diagnosis module determines whether the bearing is worn based on the real-time working parameters of the main shaft of the turboshaft engine and the feature parameters of the vibration signal. If the bearing is worn, a wear location instruction is sent to the wear location module;

[0051] The wear location module is used to respond to the wear location instruction to determine the wear location; the wear location module is configured with a time delay difference map; the wear location module constructs a time delay difference feature vector of the impact event based on the feature matrix of the vibration signal, matches the time delay difference feature vector with the time delay difference map, and determines the wear location according to the matching result.

[0052] Compared with the prior art, the beneficial effects achieved by the present application are as follows:

[0053] The present application realizes the structured coding of the high-order statistical features of the original vibration signal by constructing a third-order cumulant diagonal slice projection matrix of the vibration signal, effectively suppresses Gaussian noise, amplifies the impact components related to wear in the vibration signal, and makes the vibration characteristics of wear more obvious.

[0054] Based on the feature matrix of the vibration signal, various feature parameters such as amplitude, frequency, kurtosis, and impulse factor are extracted, and combined with the real-time working parameters such as rotational speed, load, and cumulative operating time of the main shaft of the turboshaft engine, and joint analysis is carried out through a machine learning model to comprehensively reflect the operating state of the bearing, thereby improving the accuracy of judging whether the bearing is worn.

[0055] By calculating the cross-correlation coefficients between different impact pulse sequences, the time delay difference of impact events is accurately extracted. A time delay difference feature vector is constructed and matched with a preset time delay difference spectrum map to achieve the accurate determination of the bearing wear position; the direct correlation between the impact event propagation path and time domain characteristics and the wear position is fully utilized, effectively overcoming the influence of rotational speed fluctuations on the judgment of the wear position. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. Among them:

[0057] Figure 1 is a flowchart of a vibration diagnosis method for wear of the main shaft bearing of a turboshaft engine provided by the present application;

[0058] Figure 2 is a schematic diagram of a vibration diagnosis system for wear of the main shaft bearing of a turboshaft engine provided by the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] The technical solutions of the present application will be described in detail below through the drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical solutions of the present application, rather than limitations on the technical solutions of the present application. Without conflict, the technical features in the embodiments of the present application and the embodiments can be combined with each other.

[0060] Embodiment 1

[0061] This embodiment introduces a vibration diagnosis method for wear of the main shaft bearing of a turboshaft engine. Referring to Figure 1 , the method includes the following steps:

[0062] Collect the vibration signal of the main shaft bearing of the turboshaft engine; construct the cumulant projection matrix of the vibration signal;

[0063] The vibration signal is collected based on a vibration sensor. In this embodiment, it is preferably to install a vibration sensor on the bearing housing of the turboshaft engine to accurately obtain the vibration signal related to the main shaft bearing.

[0064] The cumulant projection matrix is the diagonal slice projection matrix of the third-order cumulant of the vibration signal; the method for constructing the cumulant projection matrix of the vibration signal is as follows:

[0065] Calculate the third-order cumulant of the vibration signal at different time delays; extract the diagonal slice data from the third-order cumulant; arrange the diagonal slice data into a cumulant projection matrix of the vibration signal.

[0066] This application preferably uses a Hankel matrix as the specific form of the cumulant projection matrix. The third-order cumulant is a high-order statistic. Arranging the diagonal slice data in the third-order cumulant into a matrix in Hankel structure realizes the structured encoding of the high-order statistical features of the original vibration signal. The non-linear impact features of the original vibration signal are projected into the space composed of high-order statistics, realizing the suppression of Gaussian noise in the vibration signal (the high-order cumulant of Gaussian noise is zero), and amplifying the impact components of non-Gaussian noise related to wear in the vibration signal.

[0067] Perform dimensionality reduction on the cumulant projection matrix by using truncated singular value decomposition to obtain a feature matrix of the vibration signal; specifically including:

[0068] Perform singular value decomposition on the cumulant projection matrix to obtain a singular value diagonal matrix and left singular vectors of the cumulant projection matrix; perform singular value decomposition on the cumulant projection matrix through a mathematical algorithm software library to obtain a singular value diagonal matrix and a left singular matrix; among them, the singular value diagonal matrix contains singular values, and any column in the left singular matrix is a left singular vector.

[0069] Determine the number of truncated dominant singular values, denoted as k; extract the largest k singular values from the singular value diagonal matrix; the preferred way to determine the number of truncated dominant singular values in the embodiments of this application is as follows: calculate the total energy of all elements in the cumulant projection matrix; if the sum of the energies of the elements in the k column vectors in the cumulant projection matrix corresponding to the largest k singular values accounts for more than 95% of the total energy of all elements in the cumulant projection matrix, then k is the number of truncated dominant singular values.

[0070] Arrange the left singular vectors corresponding to the largest k singular values into a feature matrix of the vibration signal; any column in the feature matrix corresponds to a left singular vector. Through singular value truncation, only the left singular vectors corresponding to k dominant singular values are retained, realizing feature dimensionality reduction, removing secondary information such as noise corresponding to smaller singular values, and retaining impact events related to wear corresponding to larger singular values, achieving the purpose of highlighting signal features.

[0071] The impact signals generated by bearing wear propagate through different physical paths, such as directly propagating or reflecting along the inner ring, outer ring, rolling elements, etc. to the sensor; in this embodiment, by calculating the third-order cumulant diagonal slice matrix of the vibration signal, each column of which corresponds to the third-order cumulant slices with different time delays, this matrix implicitly contains the multi-path propagation time delay information of the impact events in the vibration signal. Through truncated singular value decomposition, the dominant impact events are extracted; the column space of the left singular vectors reflects the physical propagation paths of the impact signals. Each column of the left singular vectors represents an impact event component.

[0072] Based on the characteristic matrix of the vibration signal, determine whether the bearing is worn; specifically including:

[0073] Based on the characteristic matrix of the vibration signal, extract the characteristic parameters of the vibration signal; the characteristic parameters include at least one of amplitude, frequency, kurtosis, and impulse factor;

[0074] Obtain the real-time working parameters of the main shaft of the turboshaft engine; the real-time working parameters include at least one of rotational speed, load, and cumulative operating time;

[0075] After normalizing and encoding the characteristic parameters of the vibration signal and the real-time working parameters of the main shaft of the turboshaft engine, input them into the trained wear prediction model to obtain the judgment result of whether there is wear;

[0076] The wear prediction model is any one of a support vector machine, a random forest model, and a gradient boosting decision tree, and the output is a binary judgment result of whether the bearing has wear or the bearing does not have wear.

[0077] By constructing the cumulant projection matrix of the vibration signal and further reducing the dimension of the cumulant projection matrix, the obtained characteristic matrix highlights the impact events caused by wear in the vibration signal. By analyzing the characteristic matrix, it is possible to judge whether the bearing has wear and the degree of wear, etc. Through the analysis of the historical vibration data and experimental data calibration of a large number of bearings with known wear states, the laws and characteristics of the characteristic parameters corresponding to typical wear patterns can be obtained. For example, if the amplitude extracted from the characteristic matrix is significantly higher than the amplitude in the normal operating state and exceeds the preset threshold, it represents that the bearing has wear; bearing wear will cause the appearance or enhancement of specific frequency components. For example, inner ring wear appears as a peak at multiples of the rotational frequency, and outer ring wear has obvious frequency components at the characteristic frequencies related to the rolling elements. Through spectrum analysis, if the energy of the frequency components extracted from the characteristic matrix significantly increases at the above characteristic frequencies, the bearing has wear; when the kurtosis value extracted from the characteristic matrix exceeds the normal range or the impulse factor significantly increases and is consistent with the change trend of the corresponding indicators in the known wear cases, the bearing has wear.

[0078] In this embodiment, a machine learning model (such as a support vector machine) is preferably used to automatically learn the laws and characteristics of the above-mentioned characteristic parameters. At the same time, real-time operating parameters such as the rotational speed, load, and cumulative operating time of the main shaft of the turboprop engine are considered for joint analysis, thereby improving the accuracy of wear judgment.

[0079] If there is no wear in the bearing, continue to collect the vibration signal of the main shaft bearing of the turboprop engine and continuously judge whether there is wear in the bearing; if there is wear in the bearing, determine the wear position, and at the same time continue to collect the vibration signal of the main shaft bearing of the turboprop engine and perform subsequent processing and judgment.

[0080] The method for determining the wear position is as follows:

[0081] Based on the characteristic matrix of the vibration signal, calculate the time delay difference of the impact events in the vibration signal;

[0082] The time delay difference is the time difference when different impact events are detected by the sensor in the characteristic matrix of the vibration signal; the calculation of the time delay difference of the impact events in the vibration signal specifically includes:

[0083] Perform Hilbert envelope demodulation on each column of the left singular vectors in the characteristic matrix of the vibration signal respectively to obtain the impact pulse sequence corresponding to each column of the left singular vectors; perform Hilbert transform on each column of the left singular vectors and calculate its envelope signal as the impact pulse sequence to eliminate high-frequency carrier interference and highlight the time-domain characteristics of the impact events in the vibration signal.

[0084] Calculate the cross-correlation coefficients between any two impact pulse sequences at different lag times and form the cross-correlation coefficient sequence corresponding to the two impact pulse sequences;

[0085] Extract the lag time corresponding to the maximum value in the cross-correlation coefficient sequence as the time delay difference between the impact events corresponding to the two impact pulse sequences.

[0086] When the bearing is worn, the impact events caused by the wear reach the vibration sensor along different propagation paths; the waveforms of the impact pulse sequences corresponding to different propagation paths are similar, but due to the different arrival times at the sensor, there is a time delay difference in the waveforms. By calculating the cross-correlation coefficients between two impact pulse sequences at different lag times, the time difference between the arrival times of the impact events corresponding to these two impact pulse sequences at the sensor, that is, the time delay difference, can be determined; the time delay difference reflects the time difference caused by different propagation paths of the impact events generated by the wear in the bearing. When the lag time is equal to the time delay difference, the cross-correlation coefficient is the largest, that is, the waveforms of the impact pulse sequences are the most similar at this lag time.

[0087] Construct the time delay difference feature vector of the impact event; specifically, it includes: constructing the time delay difference matrix of the impact event in the vibration signal; the time delay difference matrix is a strictly upper triangular matrix, and the element in the i-th row and j-th column of the time delay difference matrix is the time delay difference between the i-th impact event and the j-th impact event in the vibration signal; where i > j, the value range of i is 2, 3, ……, m, m is the number of columns of the feature matrix of the vibration signal, that is, the number of impact events in the feature matrix; the value range of j is 1, 2, ……, m - 1;

[0088] Extract the non-zero elements in the time delay difference matrix row by row and arrange them into the time delay difference feature vector of the impact event.

[0089] The wear of the bearing will cause periodic impacts as the bearing rotates. The propagation path and time domain characteristics of the impact event are directly related to the wear position, and the time delay difference records the wear position information. When the rotational speed changes, the time difference of the periodic impacts will change accordingly, that is, the magnitude of the time delay difference of the impact event will change; in this embodiment, by constructing the time delay difference matrix of the impact event in the vibration signal, which is a matrix composed of the time delay differences between all pairs of impact modes, and then constructing the time delay difference feature vector of the impact event. Even if any one of the time delay differences fluctuates with the rotational speed, the relative magnitudes of the multiple time delay differences in the time delay difference feature vector can remain constant (all time delay differences are scaled proportionally). Therefore, while retaining the wear position information, the time delay difference feature vector eliminates the influence of rotational speed fluctuations on the wear position information, making the judgment of the wear position more accurate.

[0090] Match the time delay difference feature vector with the preset time delay difference spectrum, and determine the wear position according to the matching result.

[0091] The time delay difference spectrum includes the reference time delay difference feature vectors corresponding to each known wear position; the method of constructing the time delay difference spectrum specifically includes: collecting the historical vibration signals of the bearings with known wear positions, calculating the time delay differences of the impact events in the historical vibration signals and constructing the time delay difference feature vectors as the reference time delay difference feature vectors corresponding to the wear positions; for the wear positions lacking historical vibration signals, supplement the historical vibration signals through simulation experiments and construct the corresponding reference time delay difference feature vectors.

[0092] Matching the time delay difference feature vector with the preset time delay difference spectrum and determining the wear position according to the matching result specifically includes: calculating the similarity between the time delay difference feature vector and each reference time delay difference feature vector in the time delay difference spectrum respectively; the wear position corresponding to the reference time delay difference feature vector with the highest similarity to the time delay difference feature vector is the current wear position of the main shaft bearing of the turboprop engine.

[0093] The method for calculating the similarity between the time delay difference feature vector and any one reference time delay difference feature vector in the time delay difference spectrum is as follows:

[0094] Normalize the time delay difference feature vector and the reference time delay difference feature vector respectively; through the normalization process, the scaling of the time delay difference caused by the rotational speed change is eliminated, and the relative ratio of the time delay difference is retained.

[0095] Construct a distance matrix; the element in the p-th row and q-th column of the distance matrix is denoted as ; is the Euclidean distance between the p-th element in the time delay difference feature vector and the q-th element in the reference time delay difference feature vector; the value range of p is 1, 2, ……, , is the length of the time delay difference feature vector; the value range of q is 1, 2, ……, , is the length of the reference time delay difference feature vector;

[0096] Initialize the cumulative distance matrix; the dimension of the cumulative distance matrix is the same as that of the distance matrix;

[0097] Iteratively calculate the cumulative distance value based on the distance matrix, and fill the elements of the cumulative distance matrix row by row based on the cumulative distance value;

[0098] Extract the cumulative distance value in the lower right corner of the cumulative distance matrix as the similarity between the time delay difference feature vector and the reference time delay difference feature vector.

[0099] Iteratively calculate the cumulative distance value based on the distance matrix, and fill the elements of the cumulative distance matrix row by row based on the cumulative distance value, specifically including:

[0100] Let the element in the p-th row and q-th column of the cumulative distance matrix be ; fill the elements of the cumulative distance matrix row by row, and fill the elements in each row in the order from left to right; for the element with any values of p and q , if there exists , , in the cumulative distance matrix, then is the sum of the minimum values of and , , ; otherwise, is .

[0101] By iteratively calculating the cumulative distance value, the minimum Euclidean distance of the time delay differences between the time delay difference feature vectors and the reference time delay difference feature vectors is accumulated each time. Thus, the time axes of the cumulative time delay difference feature vectors and the reference time delay difference feature vectors are dynamically aligned, ignoring the scaling of the time delay differences caused by rotational speed changes and focusing on comparing the waveform similarities of the time delay difference feature vectors. In this way, the best match between the time delay difference feature vectors obtained from real-time detection calculations and the time delay difference spectrogram is found through a non-linear alignment path, avoiding misjudgment of the wear position caused by changes in the magnitude of the time delay differences.

[0102] The rotational speed fluctuation will cause the overall scaling of the absolute value of the time delay difference, but the periodic characteristics of the time delay differences corresponding to different wear positions remain unchanged. For example, when the inner ring is worn, the impact events are transmitted through the rotating rolling elements and modulated by the loaded area. The periodic change of the time delay difference is affected by the rotational speed and the bearing geometric parameters. When the outer ring is worn, the impact events are mainly transmitted along the stationary outer ring with a fixed path, and the periodicity of the time delay difference is relatively stable. When the rolling element is worn, the impact events occur periodically with the revolution of the rolling element, and the time delay difference shows a quasi-periodic oscillation. Based on the method provided in this application, in the preset time delay difference spectrogram, multiple time delay differences in the reference time delay difference feature vector corresponding to the outer ring wear are close to the theoretical time delay difference of the outer ring wear, such as 0.01 seconds. During actual detection, multiple time delay differences in the time delay difference feature vector constructed from the vibration signal are close to integer multiples of 0.01 seconds. By calculating the highest similarity with the reference time delay difference feature vector of the outer ring wear, the wear position is determined to be the outer ring. This method successfully converts the spatial characteristics of the wear position into the relative ratio of the time delay differences and realizes high-precision wear positioning under variable rotational speed conditions through the elastic alignment ability of calculating the similarity by cumulative distance.

[0103] Embodiment 2

[0104] This embodiment is the second embodiment of this application; based on the same inventive concept as Embodiment 1, referring to Figure 2 , this embodiment introduces a vibration diagnosis system for the wear of the main shaft bearing of a turboshaft engine, including a data acquisition module, a data processing module, a wear diagnosis module, and a wear positioning module; where:

[0105] The data acquisition module is used to collect the vibration signals of the main shaft bearing of the turboshaft engine and collect the real-time working parameters of the main shaft of the turboshaft engine, including rotational speed, load, and cumulative operating time; this module includes a vibration sensor, and the vibration sensor performs continuous sampling according to the set sampling frequency to obtain the vibration signals related to the main shaft bearing, providing a data basis for subsequent analysis.

[0106] The data processing module is used to construct the feature matrix of the vibration signal and extract the feature parameters of the vibration signal based on the feature matrix, including amplitude, frequency, kurtosis, and impulse factor. First, the module constructs a cumulant projection matrix, calculates the third-order cumulants of the vibration signal at different time delays, extracts the diagonal slice data, and arranges them into a matrix in Hankel structure to realize the structured encoding of the high-order statistical features of the original vibration signal, suppress Gaussian noise, and amplify the impact components related to wear. Then, truncated singular value decomposition is used to reduce the dimension of the cumulant projection matrix, and the left singular vectors corresponding to the largest k singular values are extracted to form the feature matrix of the vibration signal, highlighting the impact events related to wear and providing effective data for wear judgment and positioning. The module also extracts feature parameters based on the feature matrix of the vibration signal for wear identification.

[0107] The wear diagnosis module determines whether there is wear in the bearing based on the real-time working parameters of the main shaft of the turboshaft engine and the feature parameters of the vibration signal. If there is wear in the bearing, it sends a wear positioning instruction to the wear positioning module. After normalizing and encoding these parameters, they are input into a trained wear prediction model, such as a support vector machine, random forest model, gradient boosting decision tree, etc., and the judgment result of whether there is wear in the bearing is output. The module comprehensively analyzes the feature parameters of the vibration signal and the real-time working parameters of the engine, and uses the trained machine learning model to accurately judge whether there is wear in the bearing.

[0108] The wear positioning module is used to respond to the wear positioning instruction and determine the wear position. The wear positioning module is configured with a time delay difference map. Based on the feature matrix of the vibration signal, the wear positioning module constructs a time delay difference feature vector of the impact event, matches the time delay difference feature vector with the time delay difference map, and determines the wear position according to the matching result. After determining that there is wear in the bearing, the module performs Hilbert envelope demodulation on each column of the left singular vectors of the vibration signal feature matrix to obtain an impact pulse sequence, then calculates the cross-correlation coefficient of any two impact pulse sequences at different lag times, finds the lag time corresponding to the maximum value as the time delay difference, and constructs a time delay difference feature vector. The time delay difference feature vector is matched with the time delay difference map, and the wear position is determined by calculating the similarity.

[0109] For the specific function implementation of the above modules, refer to the relevant content in the vibration diagnosis method for the wear of the main shaft bearing of the turboshaft engine described in Embodiment 1, which will not be elaborated here.

[0110] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0111] The embodiments of the present application have been described above in conjunction with the accompanying drawings. However, the present application is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Those of ordinary skill in the art, under the inspiration of the present application and without departing from the purpose and scope protected by the present application, can also make many forms, and these all fall within the protection scope of the present application.

Claims

1. Vibration diagnosis method for wear of the main shaft bearing of a turboshaft engine, characterized in that: It includes the following steps: Collect the vibration signals of the main shaft bearing of the turboshaft engine; construct the cumulant projection matrix of the vibration signals; Reduce the dimension of the cumulant projection matrix by using truncated singular value decomposition to obtain the feature matrix of the vibration signals; Based on the feature matrix of the vibration signals, determine whether there is wear on the bearing; If there is wear on the bearing, determine the wear position, specifically including: Based on the feature matrix of the vibration signals, calculate the time delay difference of the impact events in the vibration signals; Construct the time delay difference feature vector of the impact events; Match the time delay difference feature vector with the preset time delay difference spectrum, and determine the wear position according to the matching result.

2. The vibration diagnosis method for the wear of the main shaft bearing of a turboshaft engine according to claim 1, characterized in that: The vibration signals are collected based on vibration sensors; the cumulant projection matrix is the diagonal slice projection matrix of the third-order cumulant of the vibration signals; the method for constructing the cumulant projection matrix of the vibration signals is as follows: Calculate the third-order cumulants of the vibration signals at different time delays; Extract the diagonal slice data from the third-order cumulants; Arrange the diagonal slice data into the cumulant projection matrix of the vibration signals.

3. The vibration diagnosis method for the wear of the main shaft bearing of a turboshaft engine according to claim 2, characterized in that: The reduction of the dimension of the cumulant projection matrix by using truncated singular value decomposition to obtain the feature matrix of the vibration signals specifically includes: Perform singular value decomposition on the cumulant projection matrix to obtain the singular value diagonal matrix and the left singular vectors of the cumulant projection matrix; Determine the number of truncated dominant singular values, denoted as k; extract the largest k singular values from the singular value diagonal matrix; Arrange the left singular vectors corresponding to the largest k singular values into the feature matrix of the vibration signals; any column in the feature matrix corresponds to a left singular vector.

4. The vibration diagnosis method for the wear of the main shaft bearing of a turboshaft engine according to claim 3, characterized in that: Based on the feature matrix of the vibration signals, determine whether there is wear on the bearing, specifically including: Based on the feature matrix of the vibration signals, extract the feature parameters of the vibration signals; the feature parameters include at least one of amplitude, frequency, kurtosis, and impulse factor; Obtain the real-time working parameters of the main shaft of the turboshaft engine; the real-time working parameters include at least one of rotational speed, load, and cumulative operation time; After normalizing and encoding the feature parameters of the vibration signals and the real-time working parameters of the main shaft of the turboshaft engine, input them into the trained wear prediction model to obtain the judgment result of whether there is wear; The wear prediction model is any one of a support vector machine, a random forest model, and a gradient boosting decision tree, and the output is a binary judgment result of whether there is wear on the bearing or there is no wear on the bearing.

5. The vibration diagnosis method for the wear of the main shaft bearing of a turboshaft engine according to claim 4, characterized in that: The time delay difference is the time difference when different impact events are detected by the sensor in the feature matrix of the vibration signals; the calculation of the time delay difference of the impact events in the vibration signals specifically includes: Perform Hilbert envelope demodulation on each column of left singular vectors in the feature matrix of the vibration signals respectively to obtain the impact pulse sequence corresponding to each column of left singular vectors; Calculate the cross-correlation coefficients between any two impact pulse sequences at different lag times and form the cross-correlation coefficient sequence corresponding to the two impact pulse sequences; Extract the lag time corresponding to the maximum value in the cross-correlation coefficient sequence as the time delay difference between the impact events corresponding to the two impact pulse sequences.

6. The vibration diagnosis method for wear of the main shaft bearing of a turboshaft engine according to claim 5, characterized in that: Construct the time delay difference feature vector of the impact event, specifically including: Construct the time delay difference matrix of the impact event in the vibration signal; the time delay difference matrix is a strictly upper triangular matrix, and the element in the i-th row and j-th column of the time delay difference matrix is the time delay difference between the i-th impact event and the j-th impact event in the vibration signal; where i > j, the value range of i is 2, 3, ……, m, and m is the number of columns of the feature matrix of the vibration signal; the value range of j is 1, 2, ……, m - 1; Extract the non-zero elements in the time delay difference matrix row by row and arrange them into the time delay difference feature vector of the impact event.

7. The vibration diagnosis method for the wear of the main shaft bearing of a turboshaft engine according to claim 6, characterized in that: The time delay difference map includes the reference time delay difference feature vectors corresponding to each known wear position; The method for constructing the time delay difference map specifically includes: collecting the historical vibration signals of the bearings with known wear positions, calculating the time delay differences of the impact events in the historical vibration signals and constructing the time delay difference feature vectors as the reference time delay difference feature vectors corresponding to the wear positions; Match the time delay difference feature vector with the preset time delay difference map, and determine the wear position according to the matching result, specifically including: calculating the similarity between the time delay difference feature vector and each reference time delay difference feature vector in the time delay difference map respectively; the wear position corresponding to the reference time delay difference feature vector with the highest similarity to the time delay difference feature vector is the current wear position of the main shaft bearing of the turboshaft engine.

8. The vibration diagnosis method for the wear of the main shaft bearing of a turboshaft engine according to claim 7, wherein: The method for calculating the similarity between the time delay difference feature vector and any reference time delay difference feature vector in the time delay difference map is as follows: Normalize the time delay difference feature vector and the reference time delay difference feature vector respectively; Construct a distance matrix; the element in the p-th row and q-th column of the distance matrix is denoted as ; is the Euclidean distance between the p-th element of the time delay difference feature vector and the q-th element of the reference time delay difference feature vector; the value range of p is 1, 2, ……, , is the length of the time delay difference feature vector; the value range of q is 1, 2, ……, , is the length of the reference time delay difference feature vector; Initialize the cumulative distance matrix; the dimension of the cumulative distance matrix is the same as that of the distance matrix; Iteratively calculate the cumulative distance value based on the distance matrix, and fill the elements of the cumulative distance matrix row by row based on the cumulative distance value; Extract the cumulative distance value in the lower right corner of the cumulative distance matrix as the similarity between the time delay difference feature vector and the reference time delay difference feature vector.

9. The vibration diagnosis method for the wear of the main shaft bearing of a turboshaft engine according to claim 8, characterized in that: Iteratively calculate the cumulative distance value based on the distance matrix, and fill the elements of the cumulative distance matrix row by row based on the cumulative distance value, specifically including: Let the element in the p-th row and q-th column of the cumulative distance matrix be ; fill the elements of the cumulative distance matrix row by row, and fill the elements of each row in the order from left to right; for the element where p and q take any values , if there exists , , in at least one of them, then is plus the minimum value of , , ; otherwise, is .

10. A vibration diagnosis system for a main shaft bearing wear of a turboshaft engine, which is used to implement the vibration diagnosis method for the main shaft bearing wear of the turboshaft engine as described in any one of claims 1-9, characterized in that: It includes a data acquisition module, a data processing module, a wear diagnosis module, and a wear location module; where: The data acquisition module is used to acquire the vibration signals of the main shaft bearing of the turboshaft engine and collect the real-time working parameters of the main shaft of the turboshaft engine, including rotational speed, load, and cumulative operating time; The data processing module is used to construct the feature matrix of the vibration signal and extract the feature parameters of the vibration signal based on the feature matrix, including amplitude, frequency, kurtosis, and impulse factor; The wear diagnosis module determines whether the bearing is worn based on the real-time working parameters of the main shaft of the turboshaft engine and the feature parameters of the vibration signal. If the bearing is worn, it sends a wear location instruction to the wear location module; The wear location module is used to respond to the wear location instruction and determine the wear position; the wear location module is configured with a time delay difference map; the wear location module constructs the time delay difference feature vector of the impact event based on the feature matrix of the vibration signal, matches the time delay difference feature vector with the time delay difference map, and determines the wear position according to the matching result.

Images