A Rolling Bearing Life Prediction Method Based on the Similarity Matching Optimization Theory

Abstract

The present invention discloses a rolling bearing life prediction method based on the similarity matching optimization theory. First, this method proposes to simulate the bearing degradation signal through an exponential decay function and a double-exponential function model, realizing the expansion of the sample dictionary set. Secondly, based on the Jensen-Renyi divergence health index of the Gaussian mixture model, the extraction of the degradation evolution trend of the rolling bearing vibration signal is realized. Then, by fitting the bearing degradation data with a Gaussian function, noise can be effectively reduced. The similarity is measured by the L2 norm of the function space, and the similarity is obtained based on traversing and searching the dictionary set and corresponding weights are assigned. Finally, the prediction result of the remaining service life of the rolling bearing is obtained. Through the simulation analysis of the bearing full-life cycle signal, the effectiveness of the proposed method is verified. The experimental data analysis results also show that this method can effectively predict the remaining service life of the rolling bearing.

Description

Technical Field

[0001] The present invention relates to a method for predicting the life of a rolling bearing, and particularly to a method for predicting the remaining life of a rolling bearing based on the theory of parameter similarity matching, belonging to the technical field of fault diagnosis. Background Art

[0002] Rolling bearings are one of the most widely used and crucial basic components in major high-end equipment. Malfunctions of rolling bearings can lead to abnormal operation or shutdown of equipment, causing huge economic losses and safety hazards. Therefore, real-time monitoring of the working state of bearings is a key link in fault and health management. Currently, there are two aspects in fault and health management: fault diagnosis and prediction of remaining useful life (RUL). However, most fault diagnosis methods only make diagnostic conclusions when a fault has occurred, so related research on prediction methods has received more attention.

[0003] Life prediction often requires a large amount of performance degradation data, but it is not easy to obtain in practice. Therefore, expansion through simulation technology is a solution. Idriss et al. proposed a dynamic wear evolution model based on the wear change and friction force change trends in the entire life cycle of the bearing. This model can effectively simulate the dynamic impact during the operation of rotating machinery and use multiple mechanical models to comprehensively estimate the transition points between wear evolution stages. By comparing the evolution model with the experimental model, its effectiveness is verified. However, the calculation amount of this model is extremely large, and it is impossible to construct a large number of sample data. Cui et al. divided the bearing degradation process into a stable stage, a defect initiation stage, a defect propagation and damage expansion stage. Considering the coupled excitation of the time-varying surface topography and stiffness, a comprehensive dynamic model was established to simulate the defect expansion. However, the calculation amount of this model is extremely large, and it cannot be quickly extended to the prediction of target bearings. Bin et al. established a bearing degradation simulation model based on the vibration response mechanism. The frequency components of the vibration response signal depend on the fault type and operating conditions, and the amplitude is related to the degradation degree. In this simulation, all parts other than the vibration response are regarded as noise. However, the degradation part amplitude depends on an exponential function, and the degradation form is single. Wang et al. constructed a set of random degradation points based on normal random variables. After determining the change points between the healthy part and the degradation part, a function is constructed by setting the mean and variance of the double normal distribution, and the parameter values when the function expectation is maximized are solved based on the recursive idea. However, this method requires relying on expert experience. The purpose of this paper is to construct a simulation signal by combining fault pulses and amplitude modulation, which not only conforms to the actual situation during the operation of the bearing but also ensures the diversity of the degradation trend of the fault signal.

[0004] Mining degradation information from bearings has received extensive attention from scholars at home and abroad, mainly including research in time domain, frequency domain, time-frequency domain, etc. Root mean square (RMS) is the most widely used health indicator (HI) in mechanical RUL prediction. There are also methods to extract RMS and peak values from wavelet coefficients to predict the RUL of bearings. Gaperin et al. extracted the power density of gear meshing frequency from the envelope spectrum to predict the RUL of gears. Medjaher et al. calculated the correlation coefficient between two series of vibration signals captured at different time periods as the HI. Ocak et al. used the Hidden Markov Model (HMM) to fuse multiple features and calculated the probability of the HMM in the healthy stage as the HI of the bearing. However, these methods have certain drawbacks, such as being insensitive to early defects and highly fluctuating with the increase in the severity of degradation. To overcome these drawbacks, this paper proposes a feature extraction algorithm based on Gaussian mixture model and Jensen-renyi divergence. This algorithm is more sensitive in the early stage of bearing state, while the sensitivity will decrease in the later stage, and it will not produce large random perturbations to interfere with the calculation of similarity.

[0005] You et al. introduced an extended weight function and established a robustness evaluation framework to evaluate the weight function. They improved the life estimation method for uncertainty. Liu et al. compared the characteristics of any two degradation processes through the similarity measure of the matching matrix (MM), thus being able to obtain a higher long-term prediction accuracy. However, the process of extracting the matching matrix is very complex and it is difficult to complete multiple repeated calculations of the rolling window. Liu Zhen et al. utilized the internal relationship between historical samples and on-site samples and calculated the internal difference in this relationship to obtain a more accurate performance degradation value. This method can make full use of the limited historical data set. However, by repeatedly using the same sample multiple times, it will lead to linear errors in prediction. You et al. studied the similarity life prediction weight function and studied the influence on prediction accuracy by adjusting parameters. However, the weight function used still cannot solve the prediction error caused by local fluctuations of degradation features. Gu et al. proposed a new method for calculating the weight function, adding state value variables to make the weight function have better adaptability to different working conditions. However, the similarity measure effect on the bearing vibration data under individual working conditions is poor. Summary of the Invention

[0006] The purpose of the present invention is to provide a rolling bearing life prediction method based on the similarity matching optimization theory to solve the above technical problems in the prediction of the remaining life of bearings by traditional similarity methods.

[0007] The innovation points of this technology are mainly reflected in: mining the degradation evolution trend of the bearing full life cycle signal based on the Jensen-renyi divergence (JRD) of the Gaussian mixture model, constructing a double-exponential function model to simulate the degradation signal, retaining the signal impact characteristics, and suppressing the non-impact components; using Gaussian function fitting to fit the degradation characteristics of the signal, calculating the similarity using the L2 norm of the function space, and finally realizing the prediction of the remaining service life of the rolling bearing. This method constructs a dictionary set according to the fault impact characteristics of the rolling bearing, making the simulation signal conform to the actual degradation process and solving the prediction error caused by the small sample data set. This method suppresses the non-degradation information of the signal by fitting the signal after feature extraction with a Gaussian function. Traverse and search the sample dictionary based on the degraded part of the test signal, and measure the similarity based on the L2 norm of the space function. Assign corresponding weights based on the similarity measurement results and normalize the weights. Finally, the weighted sum of the true life of the sample data and the corresponding weights is obtained to get the prediction result of the remaining life of the rolling bearing. Compared with the existing technology, the advantages of this method are reflected in: this method is a data-driven life prediction method, the calculation process of this method is simple, and better remaining life prediction accuracy can be obtained on the premise of sufficient data volume; existing mature life prediction methods and signal processing technologies, such as Wiener model, proportional hazard model, exponential regression model, etc., are difficult to predict the remaining life of the bearing without prior technology, while this method is an unsupervised prediction method. When using the current relatively advanced methods to predict the remaining life of the bearing, such as the physical model-based diagnosis method, neural network model, etc., a series of parameters in the model often need to be determined based on a large amount of calculation and the results of the bearing accelerated degradation experiment. The prediction accuracy is low without the parameters being iterated to the optimal solution. On the one hand, this method does not require prior experience guidance and relies on the degraded data as the theoretical support. On the other hand, in the process of model construction, without the need for complex calculations to determine the parameters, it can effectively eliminate the prediction impact caused by the degradation fluctuation of the bearing signal. The above are the innovation points and advantages of this method;

[0008] To achieve the above object, the technical solution adopted by the present invention is a rolling bearing life prediction method based on the similarity matching optimization theory. The method includes acquiring the vibration signal of the bearing full life cycle, constructing a sample dictionary set with the simulated vibration data, extracting the features of the bearing vibration signal, selecting the bearing degradation segment, fitting the degraded data with a Gaussian function to obtain the degradation trend and function parameters, calculating the similarity using the L2 norm of the function space, then calculating the weights of different samples for the future trend of the test data, and finally obtaining the remaining service life of the test data based on the weights and the remaining life of the samples.

[0009] S1 Vibration signal acquisition;

[0010] To verify the effectiveness of the proposed method, the accelerated degradation experimental data of rolling bearings are used for verification and analysis. The experimental system consists of a bearing test bench, a data acquisition instrument, and a laptop computer. Four bearings are installed on one shaft. An AC motor connected to the shaft through a friction belt keeps the rotational speed at 2000 revolutions per minute. A radial load of 6000 pounds is applied to the shaft and bearings through a spring mechanism. Rexnord ZA-2115 double-row bearings are installed on the shaft. PCB 353B33 high-sensitivity quartz ICP accelerometers are installed on the bearing housing (two accelerometers for each bearing [x-axis and y-axis]). All failures occur after the bearing exceeds its design life (exceeding 100 million revolutions).

[0011] S2 Dictionary set construction;

[0012] Let the time-domain representation of the collected vibration signal be x(t), and a dictionary set is constructed according to its fault impact characteristics;

[0013] The full-life trajectory signal of the bearing is divided into two parts: the healthy part and the degraded part. Among them, the bearing signal in the healthy state is mainly environmental noise without impact signals. Bearings tend to be accompanied by fault impacts during the fault stage, and as the service life increases, the fault impacts become larger. The form of the impact is a periodic pulse based on the fault frequency, and the amplitude of the pulse changes randomly.

[0014] The bearing impulse response signal can be expressed as:

[0015]

[0016] In the formula, I represents the number of pulses; J represents the number of system modes; A ij is the amplitude of the jth system frequency at the ith pulse, and the specific value is a random value in [0, 0.5]; T is the theoretical period of the pulse; τ i is the difference between the theoretical period and the actual pulse time; ε j is the damping ratio corresponding to different modes; f dj is the system frequency corresponding to different modes.

[0017] Among them, the rotational speed is 2000 rpm, the bearing fault frequency is 236.4 Hz, and the frequency components before and after the fault are 2000 Hz and 4000 Hz; the damping ratios in different modes are 0.1 and 0.05; the sampling rate is 3 kHz; and the length of each sampling is 1 s.

[0018] The degradation simulation signal consists of three components: vibration response signal, environmental noise, and system noise. Environmental noise is a constant noise level at the mechanical equipment's installation location and does not vary significantly. System noise is a type of vibration interference that may originate from within the equipment and increases with the severity of bearing damage. Because bearing degradation exhibits nonlinear behavior, a double exponential function is used in the simulation signal construction to simulate degradation trends.

[0019] The full life simulation signal of the bearing can be expressed as:

[0020]

[0021] In the formula, a1 and a2 are parameters reflecting the change in vibration amplitude; b1 and b2 are parameters reflecting the bearing degradation rate; η1 and η2 are environmental noise and system noise, respectively. The environmental noise and system noise are assumed to be the signal-to-noise ratio of -150dB in a healthy bearing state. λ reflects the rate at which system noise increases with increasing damage.

[0022] S3 degradation feature extraction;

[0023] S3.1: Extract various traditional time domain and frequency domain health indicators of bearings as components of Gaussian mixture model. The feature set is X=(x1,x2,…x N ).

[0024] S3.2: Train the GMM model using the feature vectors extracted from the signal samples in the healthy state of the bearing. Calculate the minimum BIC value and select the optimal GMM.

[0025]

[0026] Where m is the number of estimated parameters and N is the number of observations.

[0027] S3.3: Provide the test feature vector to the optimal GMM model and calculate the posterior probability of the corresponding GMM component. i ,σ i ), i=1,2,……,M represents the Gaussian density component.

[0028]

[0029] where μ i and σ i denote the mean vector and covariance matrix respectively. Therefore, a complete GMM is parameterized by the mean vector μ i Covariance matrix σ i and the mixing weight w i , composed of λ=(w i ,μ i ,σ i)。The parameter τ obtained by maximizing the likelihood function of GMM in the above formula is solved using the maximum likelihood estimation (EM) algorithm. is a Gaussian mixture model:

[0030]

[0031] S3.4: Calculate the JRD for the GMM-PDs related to the test feature vectors. The calculation formula of JRD is as follows:

[0032]

[0033] w1, w2, … w n is the weight vector corresponding to the probability distribution (PD). H α is the Renyi entropy of order α when the discrete probability distribution is f = (f1, f2, … f n ). Using the mathematical properties of JRD, the health index of the bearing can be extracted. When the PDs corresponding to the health status of the bearing throughout its life are estimated and then the JRD is calculated, the degree of degradation can be evaluated. When the bearing remains in a healthy state, the JRD measurement value will be close to zero and will increase once early degradation of the bearing occurs.

[0034] S4 Calculate the similarity between the test data and the sample dictionary set;

[0035] S4.1 Determine the threshold T. Find the starting point of degradation of the bearing's full life cycle signal, screen out the bearing's degradation information, and discard the healthy part of the bearing's signal.

[0036] S4.2 Obtain the degraded parts of the bearing test data and sample data according to the threshold T, perform Gaussian function fitting on the degraded parts, retain the degradation trend of the signal, and eliminate the interference error caused by local fluctuations. f(x) is the Gaussian fitting function, and a1, b1, c1 are the fitting parameters.

[0037]

[0038] S4.3 Place the fitting functions of the test data and the sample data in the same domain and calculate the L2 norm of the two function spaces.

[0039]

[0040] (a1, b1, c1), (a2, b2, c2) are the parameter vectors of the fitting results of the test data and the sample data respectively. The simplified result is as follows:

[0041]

[0042] Take π as 3.14.

[0043] S5 Predicting the remaining service life of rolling bearings based on similarity

[0044] S5.1 Based on the traversal search method, calculate the similarity between the test data and the sample data one by one. And record the result as a vector.

[0045] S5.2 Assign weights to the vector obtained in S5.1.

[0046] S = e -dis

[0047] Where dis is the element in the vector. Through this weight function, the similarity vector is transformed into a weight vector. And normalize the weight vector.

[0048] S5.3 Record the true remaining service life of each sample in the sample dictionary set at the current monitoring point of the test data. Weighted sum it with the corresponding weight to obtain the remaining service life of the test data.

[0049] S5.4 By progressively testing the current monitoring point of the test data, obtain the remaining life curve of the test data, and realize the prediction of the remaining service life of the rolling bearing.

[0050] Compared with the existing technology, the present invention has the following beneficial effects.

[0051] The present invention proposes a method for predicting the life of rolling bearings based on the similarity matching optimization theory. A feature extraction method combining GMM and JRD is proposed, which can effectively extract the degradation components from the bearing vibration signals. It is more sensitive to the early degradation of the bearing, and the random fluctuation is smaller in the later stage of bearing failure. A simulation signal of the entire life cycle of the bearing is constructed by the impulse function and the double-exponential function, which solves the problem of low prediction accuracy caused by small sample data. The combination of the linear function fitting and the sliding window algorithm is used to solve the starting degradation point of each group of bearing degradation data. A large amount of healthy stage data that does not contain degradation information is removed, which improves the credibility of the similarity prediction results and reduces the redundant calculation amount for the similarity prediction method. According to the analysis of the simulation and experimental results, the combination of Gaussian function fitting and parameter similarity proposed in this paper can effectively improve the accuracy of similarity measurement. It significantly improves the accuracy of life prediction. The present invention combines the above methods and applies them to the field of bearing life prediction for the first time, forming a complete method for predicting the remaining service life of rolling bearings. Description of the Drawings

[0052] Figure 1 It is the flowchart of the method for predicting the life of rolling bearings based on the similarity matching optimization theory in the present invention.

[0053] Figure 2 It is the flowchart of the similarity matching optimization method in the present invention.

[0054] Figure 3 is the vibration signal of the bearing full - life cycle experiment in the present invention.

[0055] Figure 4 is the simulation result based on the actual situation of the function fault pulse in the present invention.

[0056] Figure 5 is the simulation vibration signal of the bearing full - life cycle based on the combination of the fault pulse and the double - exponential function in the present invention.

[0057] Figure 6 is the degradation feature of the remaining service life of the bearing extracted by applying the combination of the Gaussian mixture model and JRD in the present invention.

[0058] Figure 7 is the remaining service life of the bearing obtained by applying the optimization matching principle in the present invention. Specific implementation manners

[0059] The present invention will be further described below in conjunction with the accompanying drawings and specific implementation manners.

[0060] Figure 1 is the flow chart of the rolling bearing life prediction method based on the similarity matching optimization theory of the present invention. The principle of the bearing remaining service life prediction method optimized by dictionary matching will be described in detail below in conjunction with the flow chart.

[0061] (1) Construct a simulation signal dictionary set. By constructing an exponentially - decaying pulse signal and performing amplitude modulation through a double - exponential function, and finally adding noise.

[0062] (2) Calculate the probability density distribution of the bearing signal through GMM, then calculate JRD based on the distribution result, and finally transform it into CV to perform data pre - processing on the bearing signal.

[0063] (3) Calculate the degradation starting point of the bearing signal, and extract the degraded part of the bearing to establish a dictionary set.

[0064] (4) Perform Gaussian fitting on the test signal and the degraded signals in the dictionary set, and query the dictionary to solve the similarity using function parameters.

[0065] (5) Assign corresponding weights based on the similarity, and calculate the remaining service life of the test data by weighted summation with the true remaining life of the sample signal.

[0066] Figure 2 is the method flow chart of the optimization matching principle of the present invention. The specific process is as follows:

[0067] (1) Expand the sample dictionary set through simulation signals.

[0068] (2) Extract features from the bearing signal to extract the degradation evolution trend.

[0069] (3) Divide the bearing health state. For a set of data, fitting its scatter values into a linear function curve is the first-order regression analysis. The first-order regression model is as follows:

[0070] HI(k) = mk + c

[0071] where m is the slope of the linear function and c is the constant term. Their respective estimations are as

[0072]

[0073] where p is the window size of the health index intercepted by the regression analysis.

[0074] According to the observation of the bearing health index in the whole life cycle, when the health state changes to the degradation state, HI will mutate and show a sharp upward trend. Therefore, for distinguishing the two states, that is, determining the TSP point, the present invention adopts the threshold alarm technology (Alarm bound technique, ABT) to achieve it. Perform a first-order regression fitting on the data within the selected window. Use the ratio of the health indices of the first and last points of the window estimated by the fitting curve as the warning value, and judge whether it exceeds the preset warning line. If it does not exceed, it is considered that no bearing degradation trace is detected. If it exceeds, it is considered that the bearing has started to degrade.

[0075] (4) Fit the degraded part with a Gaussian function to retain the degradation trend of the signal and eliminate the interference error caused by local fluctuations. f(x) is the Gaussian fitting function, and a1, b1, c1 are the fitting parameters.

[0076]

[0077] (5) Calculate the similarity between the two based on the obtained fitting function parameters through the following.

[0078]

[0079] Convert the similarity corresponding to the sample into a weight vector through the weight function. And perform normalization processing on the weight vector.

[0080] (6) Calculate the remaining service life of the unfailed bearing based on the remaining life of the sample and the corresponding weight.

[0081] Figure 3 is the time-domain diagram of the vibration signal x(t) of the bearing full-life cycle experiment. The Rexnord ZA-2115 double-row bearing is installed on the shaft. A radial load of 6000 pounds is applied to the shaft and the bearing through the spring mechanism. The AC motor connected to the shaft through the friction belt keeps the rotational speed at 2000 revolutions per minute. The sampling interval is 10 min and the sampling frequency is 20480. It can be seen from the time-domain diagram that there is a general degradation trend, but it is not intuitive enough.

[0082] Figure 4 It is a simulation based on the fault impact pulses generated during the actual operation of the bearing. It can be seen from the figure that the simulation can effectively generate periodic fault pulses, and the fault amplitude fluctuates randomly, which is in line with the actual operation state of the bearing.

[0083] Figure 5 It is a simulation vibration signal diagram of the bearing's full life cycle based on the combination of fault pulses and double-exponential functions. It can be seen from the time-domain diagram the healthy state and the degradation state of the bearing. In the degradation state, the degradation trend of the bearing can be observed, and the degradation evolution trend of the bearing experimental signal is effectively simulated. It is also not intuitive enough.

[0084] Figure 6 It is a degradation feature diagram of the remaining service life of the bearing extracted by applying the combination of Gaussian mixture model and JRD. It can be directly seen from the figure the degradation trend of the bearing, effectively eliminating the local fluctuations of the time-domain signal, and as time goes by, the degradation rate of the bearing is getting faster and faster.

[0085] Figure 7 It is a prediction diagram of the remaining service life of the bearing obtained by applying the principle of optimized matching. It can be seen from the figure that the method proposed by the present invention effectively predicts the remaining service life of the rolling bearing, and the accuracy is significantly improved compared with the traditional method.

Claims

1. A rolling bearing life prediction method based on the similarity matching optimization theory, characterized in that: Obtain the vibration signals of the bearing throughout its life cycle, construct a sample dictionary set with simulation vibration data, extract features from the bearing vibration signals, select the bearing degradation segments, perform Gaussian function fitting on the degradation data to obtain the degradation trend and function parameters, calculate the similarity by computing the L2 norm of the function space, then calculate the weights of different samples for the future trend of the test data, and finally obtain the remaining service life of the test data based on the weights and the remaining life of the samples; The implementation steps of vibration signal feature extraction are as follows; S3.1: Extract various traditional time-domain and frequency-domain health indicators of the bearing as components of the Gaussian mixture model; the feature set is X = (x1, x2, … x N ); S3.2: Train the GMM model using the feature vectors extracted from the signal samples in the healthy state of the bearing; calculate the minimum BIC value and select the optimal GMM; where m is the number of estimated parameters and N is the number of observations; S3.3: Provide the test feature vector to the optimal GMM model and calculate the posterior probability of the corresponding GMM component; z(x, μ i , σ i ), where i = 1, 2, ……, M represents the Gaussian density components; where μ i and σ i represent the mean vector and covariance matrix respectively; thus, a complete GMM is parameterized by the mean vector μ i the covariance matrix σ i and the mixing weights w i , and is composed of λ = (w i , μ i , σ i ); the parameter τ obtained by maximizing the likelihood function of GMM in the above formula is solved using the Expectation-Maximization (EM) algorithm; is the Gaussian mixture model: S3.4: Calculate the JRD for the GMM-PDs related to the test feature vectors; the calculation formula of JRD is as follows: w1, w2, … w n are weight vectors corresponding to the probability distribution (PD); H α is the Renyi entropy of order α when the discrete probability distribution is f = (f1, f2, … f n ); The mathematical properties of JRD are used to extract the health indicators of the bearing; When the PDs corresponding to the health conditions during the entire life of the bearing are estimated, and then JRD is calculated to evaluate the degree of degradation; When the bearing remains in a healthy state, the measured value of JRD will be close to zero and will increase once early degradation of the bearing occurs; The steps to predict the remaining service life of a rolling bearing by similarity are as follows; S5.1 Based on the traversal search method, calculate the similarity between the test data and the sample data one by one; and record the results as a vector; S5.2 Assign weights to the vector obtained in S5.1; S = e -dis where dis is the element in the vector; convert the similarity vector into a weight vector through the above function; and perform normalization processing on the weight vector; S5.3 Record the true remaining service life of each sample in the sample dictionary set under the current monitoring point of the test data; perform weighted summation with the corresponding weights to obtain the remaining service life of the test data; S5.4 By advancing the current monitoring point of the test data, obtain the remaining life curve of the test data, and realize the prediction of the remaining service life of the rolling bearing.

2. The rolling bearing life prediction method based on the similarity matching optimization theory according to claim 1, wherein: The implementation steps of vibration signal acquisition are as follows; Use the rolling bearing accelerated degradation experiment data for verification and analysis; the experimental system consists of a bearing test bench, a data acquisition instrument, and a laptop computer; four bearings are installed on one shaft; an AC motor connected to the shaft through a friction belt keeps the rotational speed at 2000 revolutions per minute; a radial load of 6000 pounds is applied to the shaft and bearings through a spring mechanism; Rexnord ZA-2115 double-row bearings are installed on the shaft; PCB 353B33 high-sensitivity quartz ICP accelerometers are installed on the bearing housing, that is, there are two accelerometers for each bearing on the x-axis and y-axis; all failures occur after the bearing exceeds its design life, that is, after more than 100 million revolutions.

3. A rolling bearing life prediction method based on the similarity matching optimization theory according to claim 1, characterized in that: The implementation steps of dictionary set construction are as follows; Let the time domain representation of the collected vibration signal be x(t), and construct a dictionary set according to its fault impact characteristics; The bearing full-life trajectory signal is divided into two parts: the healthy part and the degradation part; among them, the bearing signal in the healthy state is mainly environmental noise without impact signals; the bearing often has fault impacts during the fault stage, and as the service life increases, the fault impacts become larger; the form of the impact is a periodic pulse based on the fault frequency, and the amplitude of the pulse changes randomly; The bearing impulse response signal is expressed as: In the formula, I represents the number of pulses; J represents the number of system modes; A ij is the amplitude of the j-th system frequency at the i-th pulse, and the specific value is a random value in [0, 0.5]; T is the theoretical period of the pulse; τ i is the difference between the theoretical period and the actual pulse time; ε j is the damping ratio corresponding to different modes; f dj is the system frequency corresponding to different modes; where the rotational speed is 2000 rpm, the bearing fault frequency is 236.4 Hz, the frequency components before and after the fault are 2000 Hz and 4000 Hz; the damping ratios in different modes are 0.1 and 0.05; the sampling rate is 3 kHz; The length of each sampling is 1 s; The degradation simulation signal consists of three parts: vibration response signal, environmental noise, and system noise; a double-exponential function is used to simulate the degradation trend in the construction of the simulation signal; The full-life simulation signal of the bearing is expressed as: In the formula, a1 and a2 are parameters reflecting the change in vibration amplitude; b1 and b2 are parameters reflecting the bearing degradation rate; η1 and η2 are environmental noise and system noise; the environmental noise and system noise take the signal-to-noise ratio of -150 db in the healthy state of the bearing, and λ reflects the rate at which the system noise increases with the increase in the degree of damage.

4. A rolling bearing life prediction method based on the similarity matching optimization theory according to claim 1, characterized in that: The steps for calculating the similarity between the test data and the sample dictionary set are as follows: S4.1 Determine the threshold T; find the degradation starting point of the bearing full-life cycle signal, screen out the bearing degradation information, and discard the healthy part of the bearing signal; S4.2 Obtain the degraded parts of the bearing test data and the sample data according to the threshold T, perform Gaussian function fitting on the degraded parts, retain the degradation trend of the signal, and eliminate the interference error caused by local fluctuations; f(x) is the Gaussian fitting function, and a1, b1, c1 are the fitting parameters; S4.3 Place the fitting functions of the test data and the sample data in the same domain and calculate the L2 norm of the two function spaces; (a1, b1, c1) and (a2, b2, c2) are the parameter vectors of the fitting results of the test data and the sample data respectively; the simplified result is as follows: π is taken as 3.14.

Images