Method and system for health assessment of rolling bearing of gas turbine and method and system for fault diagnosis of rolling bearing of gas turbine

Abstract

Health assessment of a rolling bearing of a gas turbine is achieved by obtaining A1 an original vibration signal of the rolling bearing of the gas turbine and decomposing A2 the original vibration sig

Description

[0001] The present disclosure relates to the technical field of bearing performance evaluation, and in particular, to a method and system for health assessment of a rolling bearing of a gas turbine and a method and system for fault diagnosis of a rolling bearing of a gas turbine. BACKGROUND

[0002] Complexity and precision are important development directions of mechanical devices like a gas turbine and a power turbine. As one of important components of the mechanical device, a rolling bearing is very important to health of this kind of device, and is the key to ensure efficient and safe operation of this device. An unexpected fault of the rolling bearing may lead to a major and unpredictable fault of the mechanical device, resulting in device shutdown, major economic losses, casualties, and the like.

[0003] Currently, a large number of studies have proved that local damage or defects of the rolling bearing often cause the device to make a lot of noise and undergo abnormal vibration, and the like, but the information is often submerged by strong background noise and interference, so that it is difficult to directly obtain key information from a vibration signal. Therefore, there is an urgent need for a method that can achieve health assessment of a rolling bearing of a gas turbine and a method that can achieve fault diagnosis of a rolling bearing of a gas turbine, so as to improve reliability and safety of the whole gas turbine. SUMMARY

[0004] An objective of the present disclosure is to provide a method and system for health assessment of a rolling bearing of a gas turbine and a method and system for fault diagnosis of a rolling bearing of a gas turbine, which achieve decomposition and reconstruction of an original vibration signal of the rolling bearing of the gas turbine, facilitate subsequent quantitative evaluation of a health degree of the rolling bearing and fault diagnosis, and improve reliability and safety of the whole gas turbine.

[0005] To achieve the above objective, the present disclosure provides the following scheme:

[0006] In an aspect, the present disclosure provides a method for health assessment of a rolling bearing of a gas turbine, including:

[0007] obtaining an original vibration signal of the rolling bearing of the gas turbine;

[0008] decomposing the original vibration signal based on an improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) algorithm to obtain multiple decomposed signals;

[0009] determining a reconstructed signal of the rolling bearing of the gas turbine according to the multiple decomposed signals;

[0010] generating a distance curve according to the reconstructed signal based on a trained support vector data description (SVDD) model, where each data point on the distance curve represents a distance value between each sampling point of the reconstructed signal and a health signal;

[0011] smoothing the distance curve based on a curve smoothing mechanism to obtain a smoothed distance curve;

[0012] determining a health degree of the rolling bearing at each time according to the smoothed distance curve; and

[0013] determining a health state of the rolling bearing according to the health degree of the rolling bearing at a current time, where the health state of the rolling bearing is any one of healthy, early warning, and fault.

[0014] Optionally, the decomposing the original vibration signal based on an ICEEMDAN algorithm to obtain multiple decomposed signals specifically includes:

[0015] decomposing the original vibration signal by using the ICEEMDAN algorithm, and removing a low-frequency interference part from a fault signal to obtain multiple intrinsic mode function (IMF) subsignals and residuals;

[0016] adaptively estimating and removing noise from the residuals;

[0017] determining whether the IMF subsignals meet a preset condition, where the preset condition is that high-frequency signals in the IMF subsignals are obviously easy to identify; and

[0018] if yes, using the multiple IMF subsignals as the multiple decomposed signals; or

[0019] if no, proceeding to the step of adaptively estimating and removing noise from the residuals.

[0020] Optionally, the determining a reconstructed signal according to the multiple decomposed signals specifically includes:

[0021] calculating a kurtosis value of each decomposed signal; and

[0022] adding up a plurality of decomposed signals with respective kurtosis values greater than an average kurtosis value in the multiple decomposed signals to obtain the reconstructed signal.

[0023] Optionally, the generating a distance curve according to the reconstructed signal based on a trained SVDD model specifically includes:

[0024] training an SVDD model by using the health signal to obtain the trained SVDD model; and

[0025] inputting the reconstructed signal into the trained SVDD model to obtain the distance curve of the rolling bearing of the gas turbine.

[0026] Optionally, the smoothing the distance curve based on a curve smoothing mechanism to obtain a smoothed distance curve specifically includes:

[0027] using a third data point on the distance curve as a current data point;

[0028] adding up the current data point and two adjacent data points in front of the current data point and averaging to obtain an average value of the three points;

[0029] repeating the step of adding up the current data point and two adjacent data points in front of the current data point and averaging to obtain an average value of the three points, sequentially with a next data point on the distance curve as the current data point, so as to obtain multiple average values of three points;

[0030] monotonically adjusting the multiple average values of three points to obtain the smoothed distance curve;

[0031] obtaining the average values of three points according to the following formula:

[0032] X*' = + + * / ) A Vz = 3,4,..., 77 X-1

[0033] where x, represents a numerical value of an 7th data point, ' represents an average value of three points obtained from a data point xh2, a data point Xt-i, and a data point Xt, i is a subscript of a data point, and n is a number of data points on the distance curve; and

[0034] monotonically adjusting the multiple average values of three points according to the following formula, yt

[0035] x\x-' > x- y 11     / -1 , Vz = 2,3,...,n -2, u =x If.,   I        777    “ J 1   1 ■ 1 «                      .A- ■ I

[0036] where yt represents a numerical value of the 7th data point on the smoothed distance curve.

[0037] Optionally, the health degree of the rolling bearing is determined according to the following formula:

[0038]    1

[0039] where hi is a health degree of the rolling bearing at an 7111 data point time, and y’ is an average value of all data points on the smoothed distance curve.

[0040] Optionally, the method for health assessment of a rolling bearing of a gas turbine further includes: inputting health degrees of the rolling bearing at first n times into a health prediction model, and predicting a health degree of the rolling bearing at a next time, where the health prediction model is a mixed model of at least two of an autoregressive model, a linear regression prediction model, and a radial basis function neural network model.

[0041] Optionally, the method for health assessment of a rolling bearing of a gas turbine further includes: determining a weight of each sub-model in the health prediction model by using a variance-covariance weight distribution method.

[0042] Corresponding to the method for health assessment of a rolling bearing of a gas turbine described above, the present disclosure further provides a system for health assessment of a rolling bearing of a gas turbine. When the system for health assessment of a rolling bearing of a gas turbine is run by a computer, the method for health assessment of a rolling bearing of a gas turbine described above is performed.

[0043] In another aspect, the present disclosure provides a method for fault diagnosis of a rolling bearing of a gas turbine, including:

[0044] performing the method for health assessment of a rolling bearing of a gas turbine described above;

[0045] determining, when a health state of the rolling bearing is a fault state, an optimal band-pass filter according to a reconstructed signal;

[0046] performing band-pass filtering on the reconstructed signal by using the optimal band-pass filter to obtain a filtered reconstructed signal;

[0047] determining an envelope spectrum according to the filtered reconstructed signal;

[0048] determining a theoretical characteristic frequency matching a characteristic frequency of the envelope spectrum from a theoretical characteristic frequency set, where the theoretical characteristic frequency set includes multiple fault types of rolling bearings and a theoretical characteristic frequency corresponding to each fault type; and

[0049] determining a fault type of the rolling bearing according to the theoretical characteristic frequency matching the characteristic frequency of the envelope spectrum.

[0050] Corresponding to the method for fault diagnosis of a rolling bearing of a gas turbine described above, the present disclosure further provides a system for fault diagnosis of a rolling bearing of a gas turbine, where when the system for fault diagnosis of a rolling bearing of a gas turbine is run by a computer, the method for fault diagnosis of a rolling bearing of a gas turbine described above is performed.

[0051] According to specific embodiments provided in the present disclosure, the present disclosure has the following technical effects:

[0052] The present disclosure provides a method and system for health assessment of a rolling bearing of a gas turbine and a method and system for fault diagnosis of a rolling bearing of a gas turbine. In the method for health assessment, decomposition and reconstruction are performed on an original vibration signal by using an improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) algorithm to obtain a reconstructed signal of the rolling bearing of the gas turbine; a distance between each sampling point of the reconstructed signal and a health signal is obtained based on a trained support vector data description (SVDD) model to generate a distance curve, and then the distance curve is smoothed to obtain a smoothed distance curve, so as to more visually express a degree of the rolling bearing gradually deviating from the health signal and degrading in performance with time; and further, a health degree of the rolling bearing at each time can be accordingly quantified, so as to determine a health state of the rolling bearing. According to the present disclosure, high-frequency interference is eliminated from the original vibration signal based on the ICEEMDAN algorithm to obtain low-frequency key information, the decomposition and reconstruction of the original vibration signal of the rolling bearing of the gas turbine are achieved, and a change of the vibration signal relative to the health signal is obtained by means of the SVDD model, so that the health degree of the rolling bearing can be better quantified to determine the health state. In the method for fault diagnosis, based on the method for health assessment, when it is determined that the rolling bearing is in a fault state, the extracted reconstructed signal is filtered and then a characteristic frequency of an envelope spectrum is extracted, and accordingly a corresponding theoretical characteristic frequency is matched from a theoretical characteristic frequency set to determine a fault type of the rolling bearing. According to the present disclosure, the reconstructed signal extracted by using the ICEEMDAN algorithm is used to generate the envelope spectrum and the characteristic frequency of the envelope spectrum is compared with the theoretical characteristic frequencies, so that the most suitable theoretical characteristic frequency and the corresponding fault type thereof can be accurately and quickly found out, thereby improving reliability and safety of the whole gas turbine. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] To describe the technical solutions in embodiments of the present disclosure or in the prior art more clearly, the accompanying drawings required in the embodiments are briefly described below. Apparently, the accompanying drawings in the following description show merely some embodiments of the present disclosure, and other drawings can be derived from these accompanying drawings by those of ordinary skill in the art without creative efforts.

[0054] FIG. 1 is a flowchart of a method for health assessment of a rolling bearing of a gas turbine according to Embodiment 1 of the present disclosure;

[0055] FIGS. 2A-B are schematic diagram showing an effect of preprocessing an original vibration signal in the method according to Embodiment 1 of the present disclosure;

[0056] FIG. 3 is a schematic diagram showing construction of a hypersphere by using an SVDD algorithm in the method according to Embodiment 1 of the present disclosure;

[0057] FIG. 4 is a flowchart of step A5 of the method according to Embodiment 1 of the present disclosure;

[0058] FIG. 5 is a schematic diagram showing an effect after curve smoothing in the method according to Embodiment 1 of the present disclosure;

[0059] FIG. 6 is a schematic structural diagram of a system for health assessment of a rolling bearing of a gas turbine according to Embodiment 2 of the present disclosure;

[0060] FIG. 7 is a flowchart of a method for fault diagnosis of a rolling bearing of a gas turbine according to Embodiment 3 of the present disclosure; and

[0061] FIG. 8 is a schematic structural diagram of a system for fault diagnosis of a rolling bearing of a gas turbine according to Embodiment 4 of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] The technical solutions of the embodiments of the present disclosure are clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Apparently, the described embodiments are merely some rather than all of the embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.

[0063] An objective of the present disclosure is to provide a method and system for health assessment of a rolling bearing of a gas turbine and a method and system for fault diagnosis of a rolling bearing of a gas turbine, which achieve decomposition and reconstruction of an original vibration signal of the rolling bearing of the gas turbine, facilitate subsequent quantitative evaluation of a health degree of the rolling bearing and fault diagnosis, and improve reliability and safety of the whole gas turbine.

[0064] In order to make the above objective, features and advantages of the present disclosure clearer and more comprehensible, the present disclosure will be further described in detail below in combination with accompanying drawings and specific implementations.

[0065] Embodiment 1

[0066] This embodiment provides a method for health assessment of a rolling bearing of a gas turbine. As shown in a flowchart of FIG. 1, the method for health assessment includes the following steps.

[0067] Al: Obtain an original vibration signal of the rolling bearing of the gas turbine. In this embodiment, original vibration signal data of the rolling bearing of the gas turbine is imported in an off-line mode. The off-line mode means that the data obtained off-line is read in a specific format and then is transferred to other processing modules. The off-line import has the advantage that a user can choose data to be analyzed by himself / herself, which not only improves efficiency, but also is stable and reliable.

[0068] A2: Decompose the original vibration signal based on an ICEEMDAN algorithm to obtain multiple decomposed signals. Specifically, a commonly used decomposition method at present is empirical mode decomposition (EMD), which is a data-based adaptive signal decomposition method that can decompose any nonlinear or non-stationary signal into a series of intrinsic mode functions (IMFs), and each IMF represents different frequency components and vibration modes of the signal. The EMD has the advantages that the EMD does not need any prior knowledge or any assumptions about a signal, and thus any signal can be decomposed.

[0069] However, the advantages of the EMD algorithm are destroyed because of aliasing of high-frequency and low-frequency signals during decomposition. Therefore, a method of reducing mode aliasing by adding auxiliary white noise is provided in follow-up research. ICEEMDAN, as an improved EMD method, adds adaptive noise and stability control on the basis of the EMD, so that the ICEEMDAN has better performance in noise interference and data short-term. The decomposition process of the ICEEMDAN includes the following steps.

[0070] (1): Select an initial parameter of the ICEEMDAN is selected, where when N = 100, a standard deviation of a noise amplitude is 0.2 times of a standard deviation of the signal.

[0071] (2): Use the ICEEMDAN to demodulate a low-frequency vibration component from a high-frequency resonance region, eliminate a low-frequency interference part in a fault signal, and obtain IMF subsignals by decomposition.

[0072] (3): Add adaptive noise, and adaptively estimate and remove noise from residuals after IMF decomposition to improve stability and accuracy of the decomposition.

[0073] (4): Set a threshold for an original signal, and control stability of the IMF subsignals obtained by the decomposition to avoid over-decomposition and under-decomposition.

[0074] (5): Repeat steps (3) and (4) until high-frequency signals in the obtained IMF subsignals are obviously easy to identify. In actual working conditions, signal components of a bearing fault signal, including a collected bearing acceleration result and speed result, are often submerged by noise signals, and cannot be observed directly easily. Therefore, in this embodiment, the ICEEMDAN method is proposed to change a noise-containing signal to IMFs, and the noise signals are removed, so that the fault signal can be observed more easily.

[0075] In this embodiment, step A2 specifically includes the following steps.

[0076] A21: Decompose the original vibration signal by using the ICEEMDAN algorithm, and remove a low-frequency interference part from a fault signal to obtain multiple IMF subsignals and residuals.

[0077] A22: Adaptively estimate and remove noise from the residuals.

[0078] A23: Determine whether the IMF subsignals meet a preset condition, where the preset condition is that high-frequency signals in the IMF subsignals are obviously easy to identify. If yes, perform step A24; or if not, proceed to step A22.

[0079] A24: Use the multiple IMF subsignals as the multiple decomposed signals.

[0080] A3: Determine a reconstructed signal of the rolling bearing of the gas turbine according to the multiple decomposed signals. In this embodiment, step A3 specifically includes the following steps.

[0081] A31: Calculate a kurtosis value of each decomposed signal. The kurtosis value is an index to describe peak characteristics of the signal in a frequency domain, and can reflect non-Gaussian and nonlinear characteristics of the signal in the frequency domain. When there are non-Gaussian and nonlinear components in the signal, the kurtosis value is large; or when the signal is Gaussian white noise, the kurtosis value is 0. In this embodiment, the kurtosis value is calculated according to the following formula: SK=^-2

[0082] ^2 (i),

[0083] where / / 2 is a second-order central moment of all decomposed signals (that is, a variance of the signals), and «4 is a fourth-order central moment of all decomposed signals, and is defined as:

[0084] Vn=E\(x-E(^^ (2),

[0085] where E( ) represents a mathematical expectation operator, x represents sample points in the decomposed signals, and n represents calculating an / / -order central moment.

[0086] A32: Add up a plurality of decomposed signals with respective kurtosis values greater than an average kurtosis value in the multiple decomposed signals to obtain the reconstructed signal.

[0087] Steps A2 and A3 are actually to preprocess an input signal, mainly to denoise the signal, and hybrid programming of Lab VIEW and MATLAB can be performed by means of an MATLAB Script node to achieve a human-computer interaction function. In this embodiment, the denoising method is ICEEMDAN decomposition, and the reconstructed signal with more obvious fault characteristics is obtained by recombining the decomposed signals and removing the redundant mode functions. As shown in FIGS. 2A-B, a curve on an upper left side is an imported original vibration signal, four diagrams on a right side are four mode functions obtained by ICEEMDAN decomposition, and a reconstructed signal is shown on a lower left side.

[0088] A4: Generate a distance curve according to the reconstructed signal based on a trained SVDD model, where each data point on the distance curve represents a distance value between each sampling point of the reconstructed signal and a health signal. As shown in FIG. 3, an SVDD algorithm can achieve division of target samples (health signals) and non-target samples (fault signals), and is often applied to anomaly detection, fault detection, and other fields. The principle of the SVDD algorithm is to construct, in space, a hypersphere with the smallest radius but containing almost all normal data by mapping normal vibration sample data at a power turbine case of a gas turbine to a high-dimensional inner product space. By calculation of a distance from a test sample to a spherical center of the hypersphere, a performance degradation degree of the bearing can be reflected. A greater distance indicates a farther deviation of performance of the rolling bearing represented by the test sample from a normal operating state, and a more obvious degradation degree.

[0089] In this embodiment, step A4 specifically includes the following steps.

[0090] A41: Train an SVDD model by using the health signal to obtain the trained SVDD model. Specifically, a genetic algorithm is used to optimize a key parameter gamma value and a penalty coefficient of the SVDD model, and SVDD models are established for randomly selected two parameters and optimized parameters respectively. About first 20% sampling points of whole life data are used as health signals and set as training samples, and the training samples are sent to the SVDD models for training.

[0091] A42: Input the reconstructed signal into the trained SVDD model to obtain the distance curve of the rolling bearing of the gas turbine. The reconstructed signal is input into the trained SVDD model to obtain the distance curve of the rolling bearing of the gas turbine.

[0092] The optimization problem of the SVDD model for health assessment of the bearing is shown in the following formula: min F(a, R, F ) = R2+cy F

[0093] a’RA -i (3); and

[0094]                                  =

[0095] where A is a hypersphere radius of vibration data at the case when the bearing is operating £ normally, a is a spherical center, is a relaxation factor of an zth training sample, n herein represents a number of the training samples, i is a subscript and represents the / th training sample, C is a penalty parameter to weigh a volume and a misclassification rate of the hypersphere, and ) v is a mapping function in the SVDD algorithm. The SVDD algorithm uses the mapping function to map all the training samples to a high-dimensional space, and constructs a boundary region. Normal training samples are within the boundary of the hypersphere, while training samples outside the boundary of the hypersphere are referred to as fault samples. The algorithm can be transformed into the following formula with reference to a Lagrange dual problem: n                              n n max a, ^xt), G(x;- £ £ ap^ ), ¢( v))

[0096] * 1=1 1=1 (5), 0 <af <C, at = 1

[0097] where i=1 , and a, is a Lagrangian coefficient corresponding to a training sample x;. After this dual problem is solved, Lagrangian coefficients corresponding to all training samples can be obtained. According to the Lagrangian coefficient, the spherical center a and the radius R of the hypersphere can be calculated. A distance from a reconstructed signal x, to the spherical center can be obtained by using the following formula: d= (^(^),^))-2^^ (^)^(^))+0,^ (^M^j))

[0098] ’ !=1 !=1 7=1 (6).

[0099] If d <R, it is considered that the reconstructed signal is inside the hypersphere and belongs to a normal operating state; or if d >R, it is considered that the reconstructed signal is outside the hypersphere and belongs to a fault state, and a greater distance indicates a greater degradation degree.

[0100] A5: Smooth the distance curve based on a curve smoothing mechanism to obtain a smoothed distance curve. After the distance curve of the rolling bearing of the gas turbine is generated by means of the SVDD model, the distance curve has a too large fluctuation and a limited practical application effect when no further processing is performed, and thus the distance curve needs to be smoothed. In this embodiment, as shown in the flowchart of FIG. 4, step A5 specifically includes the following steps.

[0101] A51: Use a third data point on the distance curve as a current data point.

[0102] A52: Add up the current data point and two adjacent data points in front of the current data point and average to obtain an average value of the three points. In this embodiment, the average values of three points are obtained according to the following formula:

[0103] V = (^-2+^i+* / )AVz = 3,4,.„^ x. *

[0104] where x, represents a numerical value of an zth data point, ' represents an average value of three points obtained from a data point x,.2, a data point xm, and a data point x,, i is a subscript of a data point, n is a number of data points on the distance curve, and + i represents any value after the value of i is an equal sign.

[0105] A53: Repeat steps A52 and A53, sequentially with a next data point on the distance curve as the current data point, to obtain multiple average values of three points.

[0106] A54: Monotonically adjust the multiple average values of three points to obtain the smoothed distance curve. An effect of smoothing the distance curve by using the smoothing mechanism of step A5 is shown in FIG. 5.

[0107] In this embodiment, the multiple average values of three points are monotonically adjusted according to the following formula: X = 1    ,,      ,’ Vz = 2,3,..., / 7-2, yl = x1 X , , X <X ,

[0108] I / -1 ’ r / -1 (8),

[0109] where j, represents a numerical value of the zth data point on the smoothed distance curve.

[0110] A6: Determine a health degree of the rolling bearing at each time according to the smoothed distance curve. In this embodiment, the health degree of the rolling bearing is determined according to the following formula:

[0111] h.=e (y / / j)2,z = 1,2,..., / 7 - 2

[0112] where h, is a health degree of the rolling bearing at an zth data point time, and y' is an average value of all data points on the smoothed distance curve. This formula can map the distance between 0 and 1, so that an evaluation result is more visual and accurate.

[0113] A7: Determine a health state of the rolling bearing according to the health degree of the rolling bearing at a current time, where the health state of the rolling bearing is any one of healthy, early warning, and fault. In this embodiment, three different ranges are set to distinguish the health state of the rolling bearing, namely, healthy (health degree of the rolling bearing G [1-0.6]), early warning (health degree of the rolling bearing c [0.4-0.6]), and fault (health degree of the rolling bearing e [0-0.4]).

[0114] In addition to determining the health state of the rolling bearing at the current time, in this embodiment, at least two of an autoregressive model, a linear regression prediction model, and a radial basis function neural network model are further combined into a health prediction model in advance, and health degrees of the rolling bearing at first n times are input into the health prediction model to predict the health degree of the rolling bearing at the next time. In the following embodiments, before this, the weight of each sub-model in the health prediction model is determined by using a variance-covariance weight distribution method. After device and object replacement, the following method may be used to determine new weight distribution: Wi

[0115] W2

[0116] ________^22^33________ ^11^22+^11^33+^22^33 (10)' _________^11^33_________ ^11^22+^11^33+^22^33 (11)'and 3 i U™

[0117] ^11^22+^11^33+^22^33 (12),

[0118] where W2 and w? are weights of the autoregressive model, the radial basis function (y cy o neural network model and the linear regression prediction model respectively, 11, 22 and 33 are prediction error variances of the autoregressive model, the radial basis function neural network model and the linear regression prediction model respectively, and finally calculated weights are changed again in a program. In this embodiment, the weight of the autoregressive model was set to 0.3608, the weight of the radial basis function neural network model was set to 0.1524, and the weight of the linear regression prediction model was set to 0.4868.

[0119] In the method for health assessment of a rolling bearing of a gas turbine according to this embodiment, decomposition and reconstruction are performed on an original vibration signal by using an ICEEMDAN algorithm to obtain a reconstructed signal of the rolling bearing of the gas turbine; a distance between each sampling point of the reconstructed signal and a health signal is obtained based on a trained SVDD model to generate a distance curve, and then the distance curve is smoothed to obtain a smoothed distance curve, so as to more visually express a degree of the rolling bearing gradually deviating from the health signal and degrading in performance with time; and further, a health degree of the rolling bearing at each time can be accordingly quantified, so as to determine a health state of the rolling bearing. According to the present disclosure, high-frequency interference is eliminated from the original vibration signal based on the ICEEMDAN algorithm to obtain low-frequency key information, the decomposition and reconstruction of the original vibration signal of the rolling bearing of the gas turbine are achieved, and a change of the vibration signal relative to the health signal is obtained by means of the SVDD model, so that the health degree of the rolling bearing can be better quantified to determine the health state.

[0120] Embodiment 2

[0121] In addition, the method for health assessment according to Embodiment 1 of the present disclosure may also be implemented by means of an architecture of a system for health assessment of a rolling bearing of a gas turbine shown in FIG. 6. As shown in FIG. 6, the system for health assessment may include a signal import module, a data preprocessing module, a distance curve generation module, a health degree determining module, and a health state assessment module. Some modules may further have subunits configured to implement functions thereof. Of course, the architecture shown in FIG. 6 is only exemplary, and one or at least two assemblies in the system shown in FIG. 6 may be omitted according to actual needs when different functions are achieved.

[0122] Embodiment 3

[0123] This embodiment provides a method for fault diagnosis of a rolling bearing of a gas turbine. As shown in a flowchart of FIG. 7, the method for fault diagnosis includes the following steps.

[0124] Bl: Perform the method for health assessment of a rolling bearing of a gas turbine described above.

[0125] B2: Determine, when a health state of the rolling bearing is a fault state, an optimal band-pass filter according to a reconstructed signal. Specifically, in this embodiment, a fast kurtosis diagram is calculated for the reconstructed signal, so as to determine main frequency components of the reconstructed signal, and the optimal band-pass filter can be determined according to the main frequency components.

[0126] B3: Perform band-pass filtering on the reconstructed signal by using the optimal band-pass filter to obtain a filtered reconstructed signal.

[0127] B4: Determine an envelope spectrum according to the filtered reconstructed signal. Specifically, in this embodiment, Hilbert transform is performed on the filtered reconstructed signal to obtain an envelope signal, and the envelope spectrum is obtained by fast Fourier transform (FFT).

[0128] B5: Determine a theoretical characteristic frequency matching a characteristic frequency of the envelope spectrum from a theoretical characteristic frequency set, where the theoretical characteristic frequency set includes multiple fault types of rolling bearings and a theoretical characteristic frequency corresponding to each fault type. In this embodiment, fast Fourier transform of Lab View and a MATLAB program are used to analyze a frequency spectrum of the read reconstructed signal and extract a fault characteristic frequency.

[0129] B6: Determine a fault type of the rolling bearing according to the theoretical characteristic frequency matching the characteristic frequency of the envelope spectrum. In addition, a fault frequency of the reconstructed signal and an error between a calculated result and a theoretical value may be calculated.

[0130] In the method for fault diagnosis of a rolling bearing of a gas turbine according to this embodiment, based on the method for health assessment according to Embodiment 1, when it is determined that the rolling bearing is in a fault state, the extracted reconstructed signal is filtered and then a characteristic frequency of an envelope spectrum is extracted, and accordingly a corresponding theoretical characteristic frequency is matched from a theoretical characteristic frequency set to determine a fault type of the rolling bearing. According to the present disclosure, the reconstructed signal extracted by using the ICEEMDAN algorithm is used to generate the envelope spectrum and the characteristic frequency of the envelope spectrum is compared with the theoretical characteristic frequencies, so that the most suitable theoretical characteristic frequency and the corresponding fault type thereof can be accurately and quickly found out.

[0131] Embodiment 4

[0132] In addition, the method according to Embodiment 1 of the present disclosure may also be implemented by means of an architecture of a system for fault diagnosis of a rolling bearing of a gas turbine shown in FIG. 8. As shown in FIG. 8, the system for fault diagnosis may include a health assessment module, a filter determining module, a band-pass filtering module, a characteristic frequency extraction module, and a fault diagnosis module. Some modules may further have subunits configured to implement functions thereof. Of course, the architecture shown in FIG. 8 is only exemplary, and one or at least two assemblies in the system shown in FIG. 8 may be omitted according to actual needs when different functions are achieved.

[0133] Specific examples are used herein, but the above description is only a description of the principle and implementations of the present disclosure, and the above embodiment is only used to help understand the method of the present disclosure and its core ideas. Those skilled in the art should understand that the above modules or steps of the present disclosure can be implemented by means of a general-purpose computer apparatus, or optionally, by means of program code executable by a computing apparatus, such that they can be stored in a storage apparatus and executed by the computing apparatus, or they can be made separately into integrated circuit modules, or a number of modules or steps in them are made into a single integrated circuit module for implementation. The present disclosure is not limited to any specific hardware and software combination.

[0134] In addition, those of ordinary skill in the art can make various modifications in terms of specific implementations and the scope of application according to the ideas of the present disclosure. In conclusion, the content of this description shall not be construed as limitations to the present disclosure.

Claims

1. A method for health assessment of a rolling bearing of a gas turbine, comprising:obtaining an original vibration signal of the rolling bearing of the gas turbine;decomposing the original vibration signal based on an improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) algorithm to obtain multiple decomposed signals;determining a reconstructed signal of the rolling bearing of the gas turbine according to the multiple decomposed signals;generating a distance curve according to the reconstructed signal based on a trained support vector data description (SVDD) model, wherein each data point on the distance curve represents a distance value between each sampling point of the reconstructed signal and a health signal;smoothing the distance curve based on a curve smoothing mechanism to obtain a smoothed distance curve;determining a health degree of the rolling bearing at each time according to the smoothed distance curve; anddetermining a health state of the rolling bearing according to the health degree of the rolling bearing at a current time, wherein the health state of the rolling bearing is any one of healthy, early warning, and fault.

2. The method for health assessment of a rolling bearing of a gas turbine according to claim 1, wherein the decomposing the original vibration signal based on an ICEEMDAN algorithm to obtain multiple decomposed signals specifically comprises:decomposing the original vibration signal by using the ICEEMDAN algorithm, and removing a low-frequency interference part from a fault signal to obtain multiple intrinsic mode function (IMF) subsignals and residuals;adaptively estimating and removing noise from the residuals;determining whether the IMF subsignals meet a preset condition, wherein the preset condition is that high-frequency signals in the IMF subsignals are obviously easy to identify; andif yes, using the multiple IMF subsignals as the multiple decomposed signals; orif no, proceeding to the step of adaptively estimating and removing noise from the residuals.

3. The method for health assessment of a rolling bearing of a gas turbine according to claim 2, wherein the determining a reconstructed signal according to the multiple decomposed signals specifically comprises:calculating a kurtosis value of each decomposed signal; andadding up a plurality of decomposed signals with respective kurtosis values greater than an average kurtosis value in the multiple decomposed signals to obtain the reconstructed signal.

4. The method for health assessment of a rolling bearing of a gas turbine according to claim 1, wherein the generating a distance curve according to the reconstructed signal based on a trained SVDD model specifically comprises:training an SVDD model by using the health signal to obtain the trained SVDD model; andinputting the reconstructed signal into the trained SVDD model to obtain the distance curve of the rolling bearing of the gas turbine.

5. The method for health assessment of a rolling bearing of a gas turbine according to claim 1, wherein the smoothing the distance curve based on a curve smoothing mechanism to obtain a smoothed distance curve specifically comprises:using a third data point on the distance curve as a current data point;adding up the current data point and two adjacent data points in front of the current data point and averaging to obtain an average value of the three points;repeating the step of adding up the current data point and two adjacent data points in front of the current data point and averaging to obtain an average value of the three points, sequentially with a next data point on the distance curve as the current data point, so as to obtain multiple average values of three points;monotonically adjusting the multiple average values of three points to obtain the smoothed distance curve;obtaining the average values of three points according to the following formula:V = +xi-i + ^,) / 3,Vz = 3,4,. ..,nX-wherein xt represents a numerical value of an zth data point, ' represents an average value of three points obtained from a data point Xt-2, a data point x«, and a data point xt, i is a subscript of a data point, and n is a number of data points on the distance curve; andmonotonically adjusting the multiple average values of three points according to the followingformula,X.X / i = 2,X...,n-2, yx = xxwhereinj / , represents a numerical value of the zth data point on the smoothed distance curve.

6. The method for health assessment of a rolling bearing of a gas turbine according to claim 5, wherein the health degree of the rolling bearing is determined according to the following formula:ht = e~^yi'y^2 J = 1,2,...,« - 2wherein ht is a health degree of the rolling bearing at an z* data point time, and y’ is an average value of all data points on the smoothed distance curve.

7. The method for health assessment of a rolling bearing of a gas turbine according to claim 1, further comprising:inputting health degrees of the rolling bearing at first n times into a health prediction model, and predicting a health degree of the rolling bearing at a next time, wherein the health prediction model is a mixed model of at least two of an autoregressive model, a linear regression prediction model, and a radial basis function neural network model.

8. The method for health assessment of a rolling bearing of a gas turbine according to claim 7, further comprising:determining a weight of each sub-model in the health prediction model by using a variance-covariance weight distribution method.

9. A method for fault diagnosis of a rolling bearing of a gas turbine, comprising:performing the method for health assessment of a rolling bearing of a gas turbine according to any one of claims 1 to 8;determining, when a health state of the rolling bearing is a fault state, an optimal band-pass filter according to a reconstructed signal;performing band-pass filtering on the reconstructed signal by using the optimal band-pass filter to obtain a filtered reconstructed signal;determining an envelope spectrum according to the filtered reconstructed signal;determining a theoretical characteristic frequency matching a characteristic frequency of the envelope spectrum from a theoretical characteristic frequency set, wherein the theoretical characteristic frequency set comprises multiple fault types of rolling bearings and a theoretical characteristic frequency corresponding to each fault type; anddetermining a fault type of the rolling bearing according to the theoretical characteristic frequency matching the characteristic frequency of the envelope spectrum.

10. A system for fault diagnosis of a rolling bearing of a gas turbine, wherein when the system for fault diagnosis of a rolling bearing of a gas turbine is run by a computer, the method for fault diagnosis of a rolling bearing of a gas turbine according to claim 9 is performed.

Images