Method for evaluating residual service life of bearing

Abstract

The invention discloses a method for evaluating the residual service life of a bearing, and particularly relates to the technical field of bearings, which comprises the following steps of: applying a common-mode voltage of a simulation inverter driving motor to an intact bearing until the bearing is invalid so as to observe a key event of severely damaging the bearing; acquiring data of vibration, current and temperature in the process from operation to failure of the bearing to obtain original vibration, current and temperature samples; determining a discharge event in the original current sample based on a wavelet decomposition method; key vibration characteristics in the original vibration sample are extracted based on a time domain statistical moment method; based on the temperature data, the key vibration characteristics and the discharge event, a prediction result of the residual life of the bearing is obtained by using an extended Kalman filtering method; using the surge of the bearing discharge event as a cue to start estimating the bearing RUL, this new RUL estimation algorithm provides a more accurate result than the RUL estimation starting from the bearing life, and requires less computational time.

Description

Technical Field

[0001] The present invention belongs to the technical field of bearings, and particularly relates to a method for evaluating the remaining service life of bearings. Background Art

[0002] Bearings are commonly used components in mechanical equipment, carrying the rotational motion of mechanical equipment. Therefore, the performance of bearings directly affects the performance and reliability of the entire mechanical equipment. In recent years, the field of bearing fault diagnosis and prediction has received much attention. There are many reasons for bearing failure, including overheating, excessive axial and radial loads, and electrical stresses such as bearing current. During the normal operation of the motor, all these factors may cause the degradation of the bearing to accelerate, shortening the service life of the motor bearing. Designing a method for evaluating the life of the motor bearing can make the performance of the motor bearing in the equipment better, ensuring its safe, efficient, and long-term operation.

[0003] Among the various factors affecting the life of the motor bearing, the most influential one is the common-mode voltage generated by the frequency converter, which then induces shaft current on the bearing. The most destructive way among the various generated shaft currents is the electro-discharge machining (EDM) current. The EDM current will cause a lot of damage to the bearing, generating pits in the rotating elements of the bearing, and ultimately leading to bearing failure. Therefore, it is necessary to design a method for evaluating the remaining service life of the bearing. It is very important to accurately diagnose and predict bearing faults, providing a reliable basis for aspects such as the design, use, and maintenance of the motor bearing field to ensure the normal operation and extend the service life of the motor bearing. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for evaluating the remaining service life of bearings, solving the technical problems of difficult condition assessment and inability to accurately predict the faults and life of motor bearings during the actual operation of current motor bearings.

[0005] To achieve the above purpose, a method for evaluating the remaining service life of bearings is provided. The evaluation method includes the following steps:

[0006] S1. Apply the simulated inverter-driven motor common-mode voltage to a sound bearing until the bearing fails to observe the key events of the severely damaged bearing;

[0007] S2. Obtain the data of vibration, current, and temperature during the bearing operation to failure to obtain the original vibration, current, and temperature samples;

[0008] S3. Based on the statistical moment method in the time domain, extract the key vibration characteristics from the original vibration samples;

[0009] S4. Based on the wavelet decomposition method, determine the discharge events in the original current samples;

[0010] S5. Based on the temperature data, key vibration characteristics, and discharge events, and using the extended Kalman filter method, obtain the prediction result of the remaining life of the bearing;

[0011] S6. According to the operating conditions of the motor bearing, make timely adjustment strategies for it.

[0012] Furthermore, the method for obtaining the data of vibration, current, and temperature during the process of the bearing running to failure in step S2 to obtain the original vibration, current, and temperature samples specifically includes:

[0013] The original temperature sample is measured by a T-type thermocouple and obtained using a timed sampling method;

[0014] The original vibration sample is measured by placing an acceleration sensor in the horizontal direction of the bearing base, and the measurement interval time and sampling time are fixed;

[0015] The original current sample is collected in real time through a current sensor and a voltage sensor.

[0016] Furthermore, the time-domain statistical moment method in step S3 includes:

[0017] Select the root mean square frequency feature, and the formula is as follows:

[0018]

[0019] where n is the number of frequency samples, and X fft,i is the i-th sample of the Fourier transform of the vibration signal.

[0020] Furthermore, the method for determining the discharge event in the original current sample based on the wavelet decomposition method in step S4 specifically includes:

[0021] S4-1: Wavelet decomposition decomposes the signal into different scales and frequency bands through multi-resolution analysis. The specific steps are as follows:

[0022] (1) Select Haar as the wavelet basis function;

[0023] (2) According to the sampling frequency and signal characteristics, select the decomposition level;

[0024] (3) Use the discrete wavelet transform to decompose the original current signal into low-frequency (approximate part) and high-frequency (detail part) components;

[0025] S4-2: The specific steps for extracting the discharge event characteristics are as follows:

[0026] (1) Locate the frequency range of the discharge event. The discharge event usually shows high-frequency components, and locate its level in the wavelet decomposition;

[0027] (2) Extract high-frequency features and extract feature signals from the decomposed high-frequency detail coefficients.

[0028] (3) Calculate energy features. The energy of the discharge event is concentrated in the high frequency. Calculate the energy of each layer of high-frequency coefficients, and highlight the layer with energy jump, which may be the discharge event.

[0029] (4) Find mutation points and use the threshold method to locate the mutation points in the high-frequency coefficients.

[0030] S4-3: The specific steps for determining the discharge event are as follows:

[0031] (1) Reconstruct the signal, perform wavelet reconstruction on the high-frequency component containing the discharge event, and restore the time-domain signal of this frequency band.

[0032] (2) Combine time features, analyze the time-domain waveform of the reconstructed signal, and determine whether there are short-time pulses that conform to the characteristics of the discharge event.

[0033] (3) Statistical characteristics, perform statistical analysis on the data collected multiple times, and verify the repeatability and distribution characteristics of the discharge event.

[0034] Furthermore, in step S5, based on temperature data, key vibration characteristics, and discharge events, and using the extended Kalman filter method, the method for obtaining the prediction result of the remaining life of the bearing specifically includes:

[0035] S5-1: According to the discharge event, establish a historical record of the discharge event within a specified time of m minutes, accumulate the number of discharges within a certain time, and fit a linear function a + bt to the current discharge event within the last m minutes. Calculate the normalized mean square error between this fitted line and the actual data points. If the normalized mean square error exceeds the preset threshold, trigger an event flag, find the inflow point of the current discharge event, and set an appropriate threshold to capture the occurrence of the high-discharge current inflow point.

[0036] S5-2: Taking the event of a sharp increase in the bearing discharge event within a short time as an indicator, start predicting the remaining service life. The training of the extended Kalman filter starts after m minutes, and then the discharge event is tracked over time to detect the currently incoming discharge event. If this incoming event is realized, the prediction of the remaining service life of the bearing based on vibration will start through the extended Kalman filter.

[0037] Furthermore, according to the calculation of the root mean square frequency feature, the best-fitting exponential function is determined to be ae bt form.

[0038] Furthermore, the method for predicting the remaining service life of the bearing based on vibration through the extended Kalman filter specifically includes:

[0039] (1) Perform curve fitting on the training data, extract the root mean square frequency feature, and establish an exponential function with the best fit form of ae bt as the observation model B;

[0040] (2) Based on the new measurement points, dynamically update the Kalman filter parameters and continue with the estimation of the remaining useful life based on ;

[0041] (3) Predict the remaining useful life of the bearing based on the Kalman filter parameters and the fault threshold.

[0042] Furthermore, the method for dynamically updating the Kalman filter parameters based on the new measurement points and continuing with the estimation of the remaining useful life based on specifically includes:

[0043] In the extended Kalman filter, the state transition model and the observation model are shown by the following formulas:

[0044] x k = f(x k-1 , u k-1 ) + w k-1

[0045] z k = h(x k ) + v k

[0046] where x k is the state to be estimated, f is the non-linear function of the state, h is the measurement function, u k is the input, z k is the measurement at time sample k, w k and v k are zero-mean Gaussian noises with covariance matrices Q k and R k , respectively;

[0047] Once f and h have been determined, they must be locally linearized around the estimated state by calculating their respective Jacobians, which respectively produce the matrices F and H;

[0048] where, describes how small changes in the state vector around time k affect the state at time k + 1; describes how small changes in the state vector affect the observables;

[0049] If the state transition model is defined by the continuous-time state equation dx / dt = Ax(t) + Lw(t), where A and L are matrices characterizing the model, x is the state vector, and w is Gaussian white noise with power spectral density Q c , then use to discretize it, where δt k is the time step to find the state transition matrix F; the matrix L is used to calculate Q k :

[0050]

[0051] where τ represents the integration variable of the matrix exponential function, and T represents the transpose of the matrix L;

[0052] The first step of the extended Kalman filter is prediction. Using the predicted state estimate value and the error covariance matrix

[0053]

[0054] Assuming no input, when a measurement value is received, the difference is used to update the extended Kalman filter, using the relationship between the predicted value and the measurement value:

[0055]

[0056] where K k is the Kalman gain; represents the transpose of the observation matrix at time k;

[0057] To predict the remaining useful life of the bearing, a failure threshold is defined:

[0058]

[0059] where γ i is the value of the training set at each failure, K is the number of training data sets, and is determined using a data set containing only vibration data.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] This life evaluation method uses the surge of bearing discharge events as a clue to start estimating the RUL of the bearing. Compared with the RUL estimation starting from the bearing life, this new RUL estimation algorithm provides more accurate results and requires less computing time. This method also provides some insights into the hidden health state of the bearing by relating the cumulative discharge to the bearing vibration.

[0062] Based on the implementation manners provided in the above aspects, the present application can be further combined to provide more implementation manners. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0064] Figure 1 This is a flowchart of the method for evaluating the remaining service life of the bearing according to the present invention. Detailed implementation manners

[0065] In order to enable those skilled in the art of the present technology to better understand the solutions of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, rather than all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0066] The embodiments of the present invention provide a method for evaluating the remaining service life of a bearing. Refer to Figure 1 , Figure 1 This is a flowchart of a method for evaluating the remaining service life of a bearing provided by the embodiments of the present invention. The evaluation method includes the following steps:

[0067] S1. Apply a simulated inverter-driven motor common-mode voltage to a sound bearing until the bearing fails to observe the key events of a severely damaged bearing;

[0068] S2. Obtain the data of vibration, current, and temperature during the operation of the bearing until failure to obtain the original vibration, current, and temperature samples;

[0069] S3. Extract the key vibration characteristics from the original vibration samples based on the statistical moment method in the time domain;

[0070] S4. Determine the discharge events in the original current samples based on the wavelet decomposition method;

[0071] S5. Based on the temperature data, key vibration characteristics, and discharge events, and using the extended Kalman filter (EKF) method, obtain the prediction result of the remaining useful life (RUL) of the bearing;

[0072] S6. Make timely adjustment strategies according to the operating conditions of the motor bearing.

[0073] Based on an assessment method for the remaining useful life of a bearing provided by an embodiment of the present invention, the surge of bearing discharge events is used as a clue to start estimating the RUL of the bearing. Compared with the RUL estimation starting from the beginning of the bearing life, this new RUL estimation algorithm provides more accurate results and requires less calculation time. This method also provides some insights into the hidden health state of the bearing by relating the cumulative discharge amount to the bearing vibration.

[0074] In one embodiment, in step S1, a sound bearing in actual use is subjected to electrical stress during the operation of its bearing shaft to ensure that the bearing starts to operate until it fails, so as to observe the key events of a severely damaged bearing. The applied voltage is designed to simulate the common-mode voltage from an inverter-driven motor, thereby allowing the EDM current to flow through the bearing, and this current is used as the main cause of the bearing's accelerated aging.

[0075] In one embodiment, in step S2, data on vibration, current, and temperature during the operation of the bearing until failure are obtained. The specific methods for obtaining the original vibration, current, and temperature samples include:

[0076] The original temperature sample is measured by a T-type thermocouple and obtained using a timed sampling method; the original vibration sample is measured by placing an acceleration sensor in the horizontal direction of the bearing base, and the measurement interval time and sampling time are fixed; the original current sample is collected in real time by a current sensor and a voltage sensor.

[0077] In one embodiment, in step S4, key vibration characteristics of the bearing are extracted based on the statistical moments in the time domain from the bearing vibration data; in the initial stage of operation, the vibration amplitude is at the lowest point, and in the final stage of operation, when the bearing starts to fail, the vibration amplitude grows exponentially; some common features for capturing trends extracted from the bearing vibration data are the statistical moments from the time domain, such as variance, and the root mean square frequency from the frequency domain. The root mean square frequency is used in this method, and the formula is as follows:

[0078]

[0079] where n is the number of frequency samples, and X fft,i is the i-th sample of the Fourier transform of the vibration signal.

[0080] In one embodiment, in step S3, the method for determining the discharge events in the original current sample based on wavelet decomposition specifically includes:

[0081] S3-1: Wavelet decomposition decomposes the signal into different scales and frequency bands through multi-resolution analysis. The specific steps are as follows:

[0082] (1) Wavelet decomposition decomposes the signal into different scales and frequency bands through multi-resolution analysis (MRA) to facilitate the extraction of the characteristics of discharge events;

[0083] (2) Select the decomposition level according to the sampling frequency and signal characteristics;

[0084] (3) Use discrete wavelet transform to decompose the original current signal into low-frequency (approximate part) and high-frequency (detail part) components;

[0085] S3-2: The specific steps for extracting the characteristics of discharge events are as follows:

[0086] (1) Locate the frequency range of the discharge event. The discharge event usually shows high-frequency components. Locate its level in the wavelet decomposition;

[0087] (2) Extract high-frequency features and extract the feature signal from the decomposed high-frequency detail coefficients;

[0088] (3) Calculate the energy characteristics. The energy of the discharge event is concentrated in the high frequency. Calculate the energy of each layer of high-frequency coefficients, and highlight the layer with energy jump, which may be the discharge event;

[0089] (4) Find the mutation points and use the threshold method to locate the mutation points in the high-frequency coefficients;

[0090] S3-3: The specific steps for determining the discharge event are as follows:

[0091] (1) Reconstruct the signal, perform wavelet reconstruction on the high-frequency component containing the discharge event, and restore the time-domain signal of this frequency band;

[0092] (2) Combine the time characteristics, analyze the time-domain waveform of the reconstructed signal, and determine whether there are short-time pulses that conform to the characteristics of the discharge event;

[0093] (3) Statistical characteristics, perform statistical analysis on the data collected multiple times to verify the repeatability and distribution characteristics of the discharge event.

[0094] Specifically, for the detection and tracking of EDM (electrical discharge machining current): During the actual operation of the bearing, there are three states of the bearing: ohmic, capacitive, and discharge. As the bearing degrades, it begins to switch between the three electrical states. When a spike appears in the current waveform recorded within a period of time, it indicates that a discharge event has occurred.

[0095] Within a short period of time, a large number of inflowing bearing discharge events cause significant and irreversible damage to the bearing, thus accelerating failure.

[0096] This method uses a wavelet decomposition-based method to detect discharge events in the original current samples: First, perform Haar wavelet decomposition on each current sample. Assume the decomposition level is 8 levels. The reason for choosing the Haar wavelet is that it is very similar to the square wave nature of the bearing current data and is an ideal choice for detecting discontinuous wavelets. In the wavelet domain, each discharge event can be observed in the signal reconstructed in the subspace spanned by the given 8-level wavelet function. The following are the 8-level wavelet functions:

[0097]

[0098] where n is the length of the signal, d 8,k is the 8th-level wavelet coefficient, ψ is the wavelet function, which is the Haar wavelet at this time; this reconstruction or projection of this subspace will generate a significant peak when the current discharge event occurs; on the contrary, the current sample without a discharge event has no obvious peak; to determine whether a fault has occurred, select the threshold T = μ(D8)+4σ(D8), where D8 = [D8(1) D8(2)... D8(N)] is the vector of the projected signal samples in this subspace, μ is the mean value, and σ is the standard deviation; whenever the reconstructed signal crosses this threshold, T is classified as a bearing discharge event;

[0099]

[0100] DE(K) is a switching function that outputs 1 or 0 according to whether D8(k) exceeds the threshold T;

[0101] Then, perform time tracking on these discharge events to obtain the cumulative sum; at a certain moment during operation, the number of discharge events per minute increases, and after the critical period of the discharge event profile, the amplitude of the vibration also begins to increase exponentially.

[0102] In one embodiment, the method for obtaining the prediction result of the bearing remaining life based on the temperature data, key vibration characteristics, and discharge events and using the extended Kalman filter method in step S5 specifically includes:

[0103] S5-1: According to the discharge events, establish a historical record of the discharge events in the specified time m minutes, accumulate the number of discharges within a certain time, and fit a linear function a + bt to the current discharge event within the last m minutes. Calculate the normalized mean square error (NMSE) between this fitted line and the actual data points. If the normalized mean square error exceeds the preset threshold, trigger an event flag, find the inflow point of the current discharge event, and set an appropriate threshold to capture the occurrence of the high-discharge current inflow point;

[0104] S5-2: Using the event of a sharp increase in bearing discharge events within a short period as an indicator, start predicting the remaining service life. The training of the extended Kalman filter starts after m minutes, and then track the discharge events over time to detect the currently incoming discharge events. If this incoming event is achieved, start predicting the remaining service life of the bearing based on vibration through the extended Kalman filter.

[0105] In one embodiment, according to the calculated root mean square frequency feature, it is determined that the best fit exponential function is ae bt form.

[0106] In one embodiment, the method for predicting the remaining service life of a bearing based on vibration through the extended Kalman filter specifically includes:

[0107] (1) Perform curve fitting on the training data, extract the root mean square frequency feature and establish an exponential function with the best fit form of ae bt as the observation model h;

[0108] (2) Based on the new measurement points, dynamically update the Kalman filter parameters and continue the estimation of the remaining service life based on ;

[0109] (3) Predict the remaining service life of the bearing based on the Kalman filter parameters and the fault threshold.

[0110] In one embodiment, the method for dynamically updating the Kalman filter parameters based on the new measurement points and continuing the estimation of the remaining service life based on specifically includes:

[0111] In the extended Kalman filter, the state transition model and the observation model are shown by the following formulas:

[0112] x k = f(x k-1 , u k-1 ) + w k-1

[0113] z k = h(x k ) + v k

[0114] Where, is the state to be estimated, f is the non-linear function of the state, h is the measurement function (non-linear function of the state), u k is the input, z k is the measurement at time sample k, w k and v k are zero-mean Gaussian noises with covariance matrices Q k and R k , respectively;

[0115] Once f and h have been determined, they must be linearized locally around the estimated state by computing their respective Jacobians, yielding matrices F and H, respectively;

[0116] where, it describes how small changes in the state vector around time k affect the state at time k+1, it describes how small changes in the state vector affect the observables.

[0117] If the state transition model is defined by the continuous-time state equation dx / dt = Ax(t) + Lw(t), where A and L are matrices characterizing the model, x is the state vector, and w is Gaussian white noise with power spectral density Q c then it is discretized using (δt k (where δt is the time step) to find the state transition matrix F; matrix L is used to compute Q k ;

[0118]

[0119] where τ represents the integration variable of the matrix exponential function and T represents the transpose of matrix L;

[0120] The first step of the EKF is prediction, using the predicted state estimate and the error covariance matrix

[0121]

[0122] Assuming no input, when a measurement is received, the difference is used to update the EKF, using the relationship between the predicted value and the measurement value:

[0123]

[0124] where K k is the Kalman gain; represents the transpose of the observation matrix at time k;

[0125] To predict the RUL of the bearing, a failure threshold is defined:

[0126]

[0127] where γi is the value of each failure in the training set, K is the number of training data sets, and is determined using a data set containing only vibration data.

[0128] Specifically, first, curve fitting is performed on the training data, appropriate observables are extracted, and the model h is established. For the root mean square frequency feature, the best fitting form is the exponential function of ae bt ; second, the state variables are selected as x = [a b ω], where dω / dt = b; it is also assumed that a and b are linearly updated: da / dt = w a and db / dt = w b , where w a and w b are Gaussian white processes; the continuous-time state equation adopted by this method is:

[0129]

[0130] where:

[0131]

[0132] To predict the RUL of the bearing, a fault threshold is defined:

[0133]

[0134] where γ i is the value of each independent training set at the time of failure, K is the number of training data sets, and is determined using the data set containing only vibration data; second, the EKF is initialized, and the prediction step is repeatedly run until the ae bt value reaches the failure threshold The time when the threshold is reached is taken as the RUL. Finally, the EKF parameters are updated using each new measurement point, and the RUL estimation based on is continued.

[0135] The method of step S5 is outlined as follows: The training of the EKF starts after m minutes, and then the discharge event is tracked over time to detect the current inflow event. Once this event is achieved, the RUL estimation based on vibration is started through the EKF.

[0136] Finally, according to the operating conditions of the motor bearing, an adjustment strategy is made for it in a timely manner.

[0137] The above has described a specific embodiment of the present invention in detail, but the content is only the preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of the present invention application should still fall within the scope covered by the patent of the present invention.

Claims

1. A method for evaluating the remaining service life of a bearing, characterized in that: The evaluation method comprises the following steps: S1, applying a common mode voltage of a simulated inverter-driven motor to a good bearing until the bearing fails, in order to observe the key events that severely damage the bearing; S2, obtaining the vibration, current and temperature data of the bearing during the process of running to failure, and obtaining the original vibration, current and temperature samples; S3. Extracting key vibration characteristics from the original vibration sample based on a statistical moment method in the time domain; S4. Determine the discharge events in the original current sample based on a wavelet decomposition method; S5. Based on the temperature data, key vibration characteristics and discharge events, and using an extended Kalman filter method, a prediction result of the remaining life of the bearing is obtained; S6. Make timely adjustment strategies based on the operating conditions of the motor bearings.

2. The method for evaluating the remaining service life of a bearing according to claim 1, characterized in that: The method of obtaining the vibration, current and temperature data of the bearing during the process of running to failure in step S2 to obtain the original vibration, current and temperature samples specifically includes: The original temperature sample is measured by a T-type thermocouple and obtained using a timed sampling method; The vibration original sample is measured by placing an acceleration sensor in the horizontal direction of the bearing base, and the measurement interval and sampling time are fixed; The original current samples are collected in real time by current sensors and voltage sensors.

3. The method for evaluating the remaining service life of a bearing according to claim 1, characterized in that: The time domain statistical moment method described in step S3 includes: Select the root mean square frequency feature, the formula is as follows: Where n is the number of frequency samples, X fft,i is the i-th sample of the Fourier transform of the vibration signal.

4. The method for evaluating the remaining service life of a bearing according to claim 1, characterized in that: The method for determining the discharge event in the original current sample based on the wavelet decomposition method in step S4 specifically includes: S4-1: The wavelet decomposition decomposes the signal into different scales and frequency bands through multi-resolution analysis, and the specific steps are as follows: (1) Select Haar as the small basis wave function; (2) Select the number of decomposition levels based on the sampling frequency and signal characteristics; (3) Decomposing the original current signal into low-frequency and high-frequency components using discrete wavelet transform; S4-2: The specific steps of extracting discharge event features are: (1) Locate the frequency range of discharge events. Discharge events usually appear as high-frequency components, and locate their levels in wavelet decomposition; (2) Extract high-frequency features and extract characteristic signals from the high-frequency detail coefficients obtained by decomposition; (3) Calculate the energy characteristics. The energy of the discharge event is concentrated in the high frequency. Calculate the energy of the high frequency coefficient of each layer, and highlight the layer with energy jump, which may be a discharge event; (4) Find the mutation point and use the threshold method to locate the mutation point in the high-frequency coefficient; S4-3: The specific steps for determining a discharge event are: (1) Reconstruct the signal by performing wavelet reconstruction on the high-frequency components containing the discharge event and restoring the time domain signal of the frequency band; (2) Analyze the time domain waveform of the reconstructed signal in combination with the time characteristics to determine whether there is a short-duration pulse that meets the characteristics of the discharge event; (3) Statistical characteristics: Statistical analysis is performed on multiple acquisition data to verify the repeatability and distribution characteristics of the discharge events.

5. The method for evaluating the remaining service life of a bearing according to claim 1, characterized in that: The method for obtaining the prediction result of the remaining life of the bearing based on the temperature data, key vibration characteristics and discharge events and using the extended Kalman filter method in step S5 specifically includes: S5-1: According to the discharge event, a historical record of the discharge event of the specified time m minutes is established, the number of discharges is accumulated within a certain period of time, and a linear function a+bt is fitted to the current discharge event within the last m minutes, and the normalized mean square error between the fitting line and the actual data point is calculated. If the normalized mean square error exceeds a preset threshold, an event flag is triggered, and the inflow point of the current discharge event is found, and a suitable threshold is set to capture the occurrence of a high discharge inflow point; S5-2: Using the sharp increase in bearing discharge events in a short period of time as an indicator, the remaining service life is predicted. The training of the extended Kalman filter begins after m minutes, and then the discharge events are tracked over time to detect the current inflow of discharge events. If this inflow event is realized, the remaining service life of the bearing will be predicted based on vibration through the extended Kalman filter.

6. The method for evaluating the remaining service life of a bearing according to claim 3, characterized in that: Based on the calculation of the RMS frequency characteristics, it is determined that the best fitting exponential function is ae bt form.

7. The method for evaluating the remaining service life of a bearing according to claim 5, characterized in that: The method for predicting the remaining service life of a bearing based on vibration by using an extended Kalman filter in step S5-2 specifically includes: (1) Perform curve fitting on the training data, extract the root mean square frequency characteristics and establish the best fitting form as ae bt The exponential function of is used as the observation model B; (2) Based on the new measurement point, dynamically update the Kalman filter parameters and continue to perform An estimate of the remaining useful life of the (3) Based on the Kalman filter parameters and fault threshold, the remaining service life of the bearing is predicted.

8. The method for evaluating the remaining service life of a bearing according to claim 7, characterized in that: Based on the new measurement point, the Kalman filter parameters are dynamically updated, and the The methods for estimating the remaining useful life include: In the extended Kalman filter, the state transition model and observation model are shown in the following formulas: x k =f(x k-1 ,u k-1 )+w k-1 z k =h(x k )+v k Among them, x k is the state to be estimated, f is the nonlinear function of the state, h is the measurement function, u k is the input, z k is the measurement of sample k at time instant, w k and v k The covariance matrix is ​​Q k and R k , zero-mean Gaussian noise; Once f and h have been determined, they must be locally linearized around the estimated state by computing their respective Jacobians, producing matrices F and H, respectively; in, It describes how small changes in the state vector around time k affect the state at time k+1; It describes how small changes in the state vector affect the observables; If the state transition model is defined by the continuous-time state equation dx / dt=Ax(t)+Lw(t), where A and L are matrices representing the model, x is the state vector, and w is a vector with power spectral density Q c Gaussian white noise, and then use To discretize it, where δt k is the time step to find the state transition matrix F; the matrix L is used to calculate Q k : Among them, τ represents the integral variable of the matrix exponential function, and T represents the transpose of the matrix L; The first step of the extended Kalman filter is prediction, using the predicted state estimate and the error covariance matrix Assuming there is no input, when the measurement value is received, the difference is updated to the extended Kalman filter, using the relationship between the predicted value and the measured value: Among them, K k is the Kalman gain; represents the transpose of the observation matrix at time k; In order to predict the remaining service life of a bearing, a failure threshold is defined: Among them, γ i is the value of the training set at each failure, K is the number of training data sets, and is determined using a data set containing only vibration data.

Images