Rolling bearing vibration behavior prediction method and system based on service process vibration evolution analysis

Abstract

The invention discloses a rolling bearing vibration behavior prediction method and system based on service process vibration evolution analysis. The method comprises the following steps: identifying service process evolution stage characteristics; carrying out correlation analysis on bearing contact surface degradation and vibration behavior evolution; and according to the obtained service process vibration behavior evolution model, constructing vibration behavior prediction based on an ARIMA-LSTM algorithm. Based on the evolution law of the bearing vibration behavior in the service process, an H1 for comprehensively representing the bearing vibration evolution behavior and a method suitable for predicting the vibration behaviors of the bearing in different stages are provided. Compared with the H1 directly obtained through machine learning, the H1 is realized on the basis of analyzing the staged characteristics of the vibration behavior evolution in the bearing service process, and the bearing evolution process is described more accurately. According to the vibration behavior prediction method provided by the invention, the problems of poor prediction accuracy and low applicability caused by different evolution characteristics in different stages of the bearing service process are well solved, and the method has a high engineering practical value.

Description

Technical Field

[0001] The present invention belongs to the technical field of bearing vibration prediction, and relates to a rolling bearing vibration behavior prediction method and system, and in particular to a rolling bearing vibration behavior prediction method and system based on vibration evolution. Background Art

[0002] As one of the core components supporting rotating equipment, rolling bearings often suffer different types of damage to the bearing raceways as contact surfaces under complex loads such as offset radial force and dynamic radial excitation. These damages make the bearing vibration changes more complicated, directly affecting the dynamic operating performance such as the bearing rotation accuracy, and even causing destructive accidents in the equipment system equipped with the bearings. In order to avoid sudden equipment accidents caused by bearing vibration changes, the demand for bearing vibration behavior prediction technology is urgent. As the basic support for predictive maintenance, bearing vibration behavior prediction can improve the safety and stability of rotating equipment operation.

[0003] At present, bearing vibration behavior prediction can be divided into two categories according to the technical route: time series data-driven and mechanical model-driven. It is generally believed that fault prediction methods mainly include three categories

[197] : 1) Failure mechanism method, such as mechanical models such as the Paris model and the Forman model to characterize the damage degradation process; 2) Data-driven methods can be divided into two categories: machine learning and statistical data-driven, (1) shallow machine learning such as neural networks, support vector machines, extreme learning machines and deep learning methods such as convolutional neural networks and deep autoencoder networks, (2) failure statistics such as exponential distribution and Weibull distribution and degradation statistics such as Winner process, Gamma process, random regression; 3) Mechanism model and data hybrid driven methods, generally in two specific forms, the combination of data-driven strategy model and mechanism model, and the fusion of data model and mechanism-based decision layer.

[0004] Harris, TA, & Kotzalas, MN (2006). "Rolling Bearing Analysis." studied the dynamic characteristics of rolling bearings in detail, especially the various loads that bearings are subject to during service. Their analysis model can predict the vibration behavior of bearings under normal conditions, but for complex load conditions and the evolution of raceway damage, the prediction accuracy of the model is low, and it is difficult to accurately reflect the actual service status of rolling bearings. The limitation of this study is that it fails to effectively consider the impact of dynamic damage expansion on vibration behavior, so it is difficult to meet the needs of high-precision vibration prediction.

[0005] "Patel, V., & Darpe, AK (2009). "Dynamic Modeling of Bearing for Prediction of Faults in Rolling Elements." Journal of Sound and Vibration." The study proposed a bearing fault prediction method based on dynamic modeling, focusing on analyzing the contact characteristics between rolling elements and raceways to identify potential failure modes. However, the model is not accurate enough in predicting faults under complex working conditions (such as multiple impact loads and cyclic excitations), especially in predicting the cumulative effect of damage. Since the model is inadequate in considering the dynamic changes during the damage expansion process, it is impossible to accurately evaluate the vibration evolution behavior of the bearing after long-term service.

[0006] "Patent US6782289B2: "Bearing Vibration Analysis and Fault Prediction System," Inventor: Jones, M. (2004)." The patent introduces a system for bearing vibration analysis and fault prediction, which uses sensors to collect vibration signals and analyzes them through characteristic frequencies. However, the system can only identify certain known fault modes, and there are technical limitations on dynamic feature recognition and long-term prediction during the evolution of bearing raceway damage, especially in the service process. It is impossible to adjust the model parameters in real time to improve the accuracy of the prediction. Therefore, the system shows a large prediction error in practical applications, especially in complex load environments.

[0007] In view of the shortcomings of the above studies and patents, the present invention proposes a rolling bearing vibration behavior prediction method and system based on the vibration evolution analysis of the service process. This method introduces the dynamic stiffness model and the vibration characteristics of periodic excitation, combined with the autoregressive integrated moving average model (ARIMA)-long short-term memory (LSTM) machine learning method, to achieve high-precision prediction of bearing vibration behavior based on the raceway damage expansion effect. Compared with the above method, the present invention can:

[0008] (1) More accurately reflect the actual vibration state of rolling bearings under complex load environments.

[0009] (2) Real-time analysis of the impact of raceway damage on vibration characteristics to overcome the poor accuracy of traditional models in long-term service prediction.

[0010] (3) Effectively improve the ability to capture the evolution law of bearing vibration under different working conditions and achieve accurate prediction of the bearing health status. Summary of the invention

[0011] The purpose of the present invention is to address the deficiencies in the prior art. The present invention provides a bearing vibration behavior method and system based on the evolution of vibration behavior during the service process. The present invention analyzes the correlation between the contact surface and the bearing vibration evolution during the service process, constructs a multi-source data fusion evolution characterization model considering the Granger causal effect according to the vibration evolution law, and proposes a bearing vibration behavior prediction based on the ARIMA-LSTM algorithm suitable for predicting different evolution stages.

[0012] The technical solution of the present invention:

[0013] A method for predicting the vibration behavior of a rolling bearing based on the vibration evolution analysis during service comprises the following steps:

[0014] S1: Identify the characteristics of the evolution stage of the service process and construct a multi-source vibration data fusion dataset;

[0015] S101: Extraction of vibration characteristics during bearing service: Construct a long-span bearing service process test plan that takes into account the variability of the actual service environment of the bearing, including offset radial force and dynamic radial excitation;

[0016] S1011: Determine the bearing speed, loading force amplitude and frequency through a long-term bearing service process test program;

[0017] S1012: Collect the original data of bearing acceleration, inner ring displacement and cage displacement during the bearing service process;

[0018] S102: extracting time domain eigenvalues ​​and frequency domain eigenvalues ​​from the acquired original data of bearing acceleration, inner ring displacement, and cage displacement, and constructing a sample data set;

[0019] S1021: Perform time domain and frequency domain analysis on the original data of bearing acceleration, inner ring displacement and cage displacement respectively;

[0020] S1022: Extract time domain eigenvalues ​​and frequency domain eigenvalues ​​and build a sample database; the formula for calculating the root mean square and peak value of the time domain eigenvalues ​​is as follows:

[0021]

[0022] Where N is the total number of data points in the sample database, x i Represents the value of each data point in the sample database;

[0023] S103: Characterizing the bearing vibration evolution law during the service process: According to the sample data set of step S102, determine the service process characterization parameter set; and according to the causal relationship algorithm, determine the combination scheme of multi-source feature combination to characterize the bearing vibration evolution during the service process, and construct a multi-source vibration data fusion data set;

[0024] Among them, the multi-source feature combinations are acceleration and inner ring displacement, acceleration and cage displacement.

[0025] S1031: Based on the sample data set of S102, the sample database obtained in S1022 is cleaned, denoised and smoothed to remove noise and abnormal data; at the same time, data standardization or normalization is performed to make the processed data have a consistent scale;

[0026] S1032: Based on the causal relationship algorithm, the Granger causal effect between different vibration behaviors during the service process of the bearing is analyzed, and a combination scheme of multi-source feature combination to characterize the vibration evolution during the service process of the bearing is determined; the Granger causal relationship calculation method is as follows:

[0027]

[0028] In the formula, n is the number of samples in the sample database, and To determine whether there is a Granger causal effect coefficient, η 1t and η 2t is the error term; x t and t are two sets of time series data, x t-i and t-i They are the past values ​​of two sets of time series data;

[0029] Time series y t is the past value y t-i and another time series x t The lag term x t-i Jointly decided, if Not all 0, it is considered that x t For t There is a Granger causal effect;

[0030] S1033: Store the data set with Granger causality in the database; use the SQL structured database to label each data sample with the label information of the corresponding operating condition, damage stage, and degradation degree, and construct a multi-source vibration data fusion data set to facilitate subsequent model training and verification;

[0031] S2: Correlation analysis between bearing contact surface degradation and vibration behavior evolution and construction of a staged evolution model of bearing vibration behavior;

[0032] S201: Quantitatively analyze the contact surface degradation stage and divide the bearing vibration evolution stage;

[0033] S2011: Quantitative analysis of the degradation stages of the contact surface at two levels: size and quantity;

[0034] S2012: Determine the three evolution stages of defect-free, defect initiation and defect expansion based on contact surface degradation;

[0035] S202: The bearing inner ring displacement and bearing acceleration are fused by a multi-source data fusion method based on the least square method to construct a vibration behavior evolution model;

[0036] S2021: Combining the causal relationship between different vibration data at different operation stages, the single vibration behavior and the associated data features with Granger causal effect are input into the multi-source data fusion model; the expression of the multi-source data fusion model is as follows:

[0037] z(k)=C(k)x(k)+w(k)(3)

[0038] In the formula,

[0039] z(k)=[z 1 (k) T z 2 (k) T ...z N (k) T ] T is the fused observation data;

[0040] C(k)=[C 1 (k) T C 2 (k) T ...C N (k) T ] T is the state transfer matrix;

[0041] x(k)=[x 1 (k) T x 2 (k) T ...x N (k) T ] T is the system state at time k;

[0042] w(k)=[w 1 (k) T w 2 (k) T ... N (k) T ] T To measure noise;

[0043] The estimated value of x(k) and the corresponding estimation error covariance matrix are obtained by the least squares method:

[0044]

[0045] Where R(k) is the covariance of the measurement noise w(k), P(k) is the estimated error covariance matrix of x(k), is the estimated value of x(k);

[0046] The multi-sensor feature information fusion equation is established based on the idea of ​​least squares method:

[0047]

[0048] In the formula, is the fused data, q is the number of data to be fused, W i is the weighting factor, X i is the data to be fused;

[0049] S2022: Conduct correlation analysis with contact surface degradation. Based on the correlation between degradation characteristics and vibration behavior, divide the operating state of the bearing into three degradation stages: no damage, damage initiation, and initiation and extension. Construct a staged evolution model of bearing vibration behavior.

[0050] S3: Based on the phased evolution model of bearing vibration behavior obtained in S2, a vibration behavior prediction based on the ARIMA-LSTM algorithm is constructed;

[0051] S301: ARIMS-LSTM algorithm is used to predict the bearing vibration behavior at different operating stages; the key calculation formula of ARIMS-LSTM algorithm is as follows:

[0052] Is to predict the time series X through the ARIMA model t The prediction result at time t is calculated as follows:

[0053]

[0054] Is to predict the time series X through the LSTM model t The prediction result at time t is calculated as follows:

[0055]

[0056] e LSTM (t) is the residual correction performed by the LSTM model by learning the prediction error of the ARIMA model. The calculation method is as follows:

[0057]

[0058] Final prediction results It is the sum of the ARIMA model prediction value and the LSTM model residual prediction value, calculated as follows:

[0059]

[0060] S302: Select MSE, RMSE and MAE as indicators for evaluating the difference between the predicted value and the true value, and optimize and adjust the prediction model; the calculation formulas of MSE, RMSE and MAE are as follows:

[0061]

[0062] In the formula, m is the number of samples, y is j is the true value of the jth sample, is the predicted value of the jth sample.

[0063] A system for predicting rolling bearing vibration behavior based on service process vibration evolution analysis includes a control module, a drive module, a test acquisition module and a calculation analysis module, wherein the test acquisition module is connected to the calculation analysis module; the control module is connected to the drive module and the test acquisition module respectively;

[0064] The calculation and analysis module includes an evolution feature calculation module, an evolution process stage analysis module, and a vibration behavior prediction module which are connected in sequence; the control module controls the operation of the drive module and sends signals to the test acquisition module to collect bearing acceleration, inner ring displacement, and retainer displacement signals, which are sent to the calculation and analysis module to be calculated and analyzed in sequence through the evolution feature calculation module, the evolution process stage analysis module, and the vibration behavior prediction module.

[0065] The evolution feature calculation module performs time domain and frequency domain analysis on the bearing acceleration, inner ring displacement and cage displacement, extracts feature values, and obtains feature information for characterizing the bearing vibration evolution.

[0066] The evolution process phase analysis module adopts the least square method multi-source data fusion method, combined with the contact surface degradation analysis, to divide the evolution process into stages.

[0067] The vibration behavior prediction module uses the ARIMA-LSTM algorithm to predict vibration data such as acceleration and inner ring displacement at different operating stages.

[0068] Beneficial effects of the present invention: The present invention proposes a method for using multi-source data fusion to characterize the phased evolution of the vibration behavior of the bearing during service and an ARIMA-LSTM prediction model for the vibration behavior of the bearing in different operating stages for vibration data with Granger causal effects. Compared with the characterization of the vibration evolution of the bearing during service based on a single vibration physical quantity, the vibration evolution characterization method of the present invention using multi-source data fusion covers a variety of vibration behaviors, characterizes the phased characteristics of vibration evolution more accurately, and is more applicable to harsh service processes such as complex loads. Secondly, the ARIMA-LSTM prediction model proposed in the present invention can process both linear and nonlinear bearing data at the same time, and the LSTM model corrects the residual by learning the prediction error of the ARIMA model, thereby improving the accuracy of the prediction model and its robustness in the face of complex service environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 The figure is a flow chart of the method of the present invention.

[0070] Figure 2 The following are the result diagrams, where (a) is the bearing acceleration evolution curve; (b) is the bearing inner ring displacement evolution curve; (c) is the bearing cage displacement evolution curve.

[0071] Figure 3 The Granger causality diagram between the vibration behavior of the bearing during service.

[0072] Figure 4 The curves of vibration data and damage size change during service, where (a) is the bearing acceleration and damage size evolution curve; (b) is the bearing inner ring displacement and damage size evolution curve; (c) is the bearing cage displacement and damage size evolution curve.

[0073] Figure 5 This is a multi-source data fusion flow chart.

[0074] Figure 6 The following is a characterization diagram of the bearing vibration evolution stage after the bearing vibration data is fused, where (a) is the fusion curve of acceleration and inner ring displacement; (b) is the fusion curve of acceleration and cage displacement.

[0075] Figure 7 These are the bearing acceleration prediction curves, where (a) is the bearing acceleration prediction curve in the defect-free stage using the ARIMA-LSTM algorithm; (b) is the bearing acceleration prediction curve in the defect initiation stage using the ARIMA-LSTM algorithm; and (c) is the bearing acceleration prediction curve in the defect expansion stage using the ARIMA-LSTM algorithm.

[0076] Figure 8These are the displacement prediction curves of the inner ring of the bearing, where (a) is the displacement prediction curve of the inner ring of the bearing in the defect-free stage using the ARIMA-LSTM algorithm; (b) is the displacement prediction curve of the inner ring of the bearing in the defect initiation stage using the ARIMA-LSTM algorithm; and (c) is the displacement prediction curve of the inner ring of the bearing in the defect expansion stage using the ARIMA-LSTM algorithm.

[0077] Fig. 9 The displacement prediction curves of the bearing cage are shown in Figure 2, where (a) is the displacement prediction curve of the bearing cage in the defect-free stage of the ARIMA-LSTM algorithm; (b) is the displacement prediction curve of the bearing cage in the defect initiation stage of the ARIMA-LSTM algorithm; and (c) is the displacement prediction curve of the bearing cage in the defect expansion stage of the ARIMA-LSTM algorithm.

[0078] Fig.10 The bar graphs are the prediction errors of bearing acceleration, inner ring displacement and cage displacement, where (a) is the prediction result analysis of bearing acceleration at different operating stages; (b) is the prediction result analysis of inner ring displacement acceleration at different operating stages; (c) is the prediction result analysis of cage displacement acceleration at different operating stages.

[0079] Fig.11 This is the schematic diagram of the system calculation and analysis module.

[0080] Fig.12 Schematic diagram of the testing device of the present invention. DETAILED DESCRIPTION

[0081] The specific implementation of the present invention is further described below in conjunction with the accompanying drawings and technical solutions.

[0082] Example

[0083] This example takes the data set collected by the NU2230 cylindrical roller bearing on the vertical bearing test bench as an example. The test conditions are as follows: the bias radial force is 10 kN, the test speed is 1200 r / min, the dynamic radial excitation is loaded for 1 minute every 0.5 hours, the sampling frequency of the acceleration and displacement sensors is 5120 Hz, the test lasts for 2500 minutes, and 1 s of data is collected every 25 minutes. The collected 5.12×10 5 The data of each point are downsampled, and the bias radial force, acceleration under dynamic radial excitation, inner ring displacement and cage data are downsampled, cleaned, denoised and smoothed to remove noise and abnormal data respectively. Figure 2 As shown in (a), (b) and (c).

[0084] The Granger causality effect relationships between the vibration data of the bearing service process obtained by the Granger causality test algorithm are as follows: Figure 3 As shown, on this basis, a multi-source vibration data fusion dataset in different operation stages was constructed.

[0085] The deterioration of the bearing contact surface and the evolution of vibration behavior were analyzed. The acceleration, inner ring displacement, cage displacement and damage size expansion changes were shown in Figure 2. Figure 4 As shown in (a), (b) and (c), the operating state of the bearing is divided into three degradation stages: no damage, damage initiation and initiation and extension.

[0086] use Figure 5 The multi-source data fusion method shown in the figure performs multi-source data fusion on the bearing acceleration and inner ring displacement, and the acceleration and cage displacement. The fusion results are shown in Figure 2. Figure 6 As shown in (a) and (b), the stage characteristics of the vibration evolution of the bearing during service are analyzed, including no damage (T1), damage initiation (T2) and damage extension (T3).

[0087] The bearing acceleration data in three stages are predicted based on the ARIMA-LSTM algorithm. The prediction results are as follows: Figure 7 As shown in (a), (b) and (c).

[0088] The bearing inner ring displacement data in three stages is predicted based on the ARIMA-LSTM algorithm. The prediction results are as follows: Figure 8 As shown in (a), (b) and (c).

[0089] The bearing cage displacement data in three stages are predicted based on the ARIMA-LSTM algorithm. The prediction results are as follows: Fig. 9 As shown in (a), (b) and (c).

[0090] The prediction error MSE, RMSE and MAE values ​​of bearing acceleration, inner ring displacement and cage displacement at different operating stages are shown in the figure. Fig.10 As shown in (a), (b) and (c).

[0091] According to the present invention, a rolling bearing vibration behavior prediction method and system based on the vibration evolution analysis during the service process, as an important data support for predictive maintenance of bearings, increases the predictability and efficiency of bearing operation and maintenance, thereby maximizing the reliability and availability of equipment, and effectively supplements the inefficiency and high cost of periodic maintenance and the high software and hardware dependence and high analytical capability requirements of real-time monitoring, forming a comprehensive equipment maintenance strategy to support the stable operation of high-end rotating equipment.

Claims

1. A method for predicting the vibration behavior of rolling bearings based on the vibration evolution analysis during service, characterized in that: The following steps are involved: S1: Identify the characteristics of the evolution stage of the service process and construct a multi-source vibration data fusion dataset; S101: Extraction of vibration characteristics during bearing service: Construct a long-span bearing service process test plan that takes into account the variability of the actual service environment of the bearing, including offset radial force and dynamic radial excitation; S1011: Determine the bearing speed, loading force amplitude and frequency through a long-term bearing service process test program; S1012: Collect the original data of bearing acceleration, inner ring displacement and cage displacement during the bearing service process; S102: extracting time domain eigenvalues ​​and frequency domain eigenvalues ​​from the acquired original data of bearing acceleration, inner ring displacement, and cage displacement, and constructing a sample data set; S1021: Perform time domain and frequency domain analysis on the original data of bearing acceleration, inner ring displacement and cage displacement respectively; S1022: Extract time domain eigenvalues ​​and frequency domain eigenvalues ​​and build a sample database; the formula for calculating the root mean square and peak value of the time domain eigenvalues ​​is as follows: Where N is the total number of data points in the sample database, x i Represents the value of each data point in the sample database; S103: Characterizing the bearing vibration evolution law during the service process: According to the sample data set of step S102, determine the service process characterization parameter set; and according to the causal relationship algorithm, determine the combination scheme of multi-source feature combination to characterize the bearing vibration evolution during the service process, and construct a multi-source vibration data fusion data set; Among them, the multi-source feature combinations are acceleration and inner ring displacement, acceleration and cage displacement; S1031: Based on the sample data set of S102, the sample database obtained in S1022 is cleaned, denoised and smoothed to remove noise and abnormal data; at the same time, data standardization or normalization is performed to make the processed data have a consistent scale; S1032: Based on the causal relationship algorithm, the Granger causal effect between different vibration behaviors during the service process of the bearing is analyzed, and a combination scheme of multi-source feature combination to characterize the vibration evolution of the bearing during service process is determined; the Granger causal relationship calculation method is as follows: In the formula, n is the number of samples in the sample database, and To determine whether there is a Granger causal effect coefficient, η 1t and η 2t is the error term; x t and t are two sets of time series data, x t-i and t-i They are the past values ​​of two sets of time series data; Time series y t is the past value y t-i and another time series x t The lag term x t-i Jointly decided, if Not all 0, it is considered that x t For t There is a Granger causal effect; S1033: Store the data set with Granger causality in the database; use the SQL structured database to label each data sample with the label information of the corresponding operating condition, damage stage, and degradation degree, and construct a multi-source vibration data fusion data set to facilitate subsequent model training and verification; S2: Correlation analysis between bearing contact surface degradation and vibration behavior evolution and construction of a staged evolution model of bearing vibration behavior; S201: Quantitatively analyze the contact surface degradation stage and divide the bearing vibration evolution stage; S2011: Quantitative analysis of the degradation stages of the contact surface at two levels: size and quantity; S2012: Determine the three evolution stages of defect-free, defect initiation and defect expansion based on contact surface degradation; S202: The bearing inner ring displacement and bearing acceleration are fused by a multi-source data fusion method based on the least square method to construct a vibration behavior evolution model; S2021: Combining the causal relationship between different vibration data at different operation stages, the single vibration behavior and the associated data features with Granger causal effect are input into the multi-source data fusion model; the expression of the multi-source data fusion model is as follows: z(k)=C(k)x(k)+w(k)(3) In the formula, z(k)=[z1(k) T z2(k) T ...z N (k) T ] T is the fused observation data; C(k)=[C1(k) T C2(k) T …C N (k) T ] T is the state transfer matrix; x(k)=[x1(k) T x2(k) T ...x N (k) T ] T is the system state at time k; w(k)=[w1(k) T w2(k) T ... N (k) T ] T To measure noise; The estimated value of x(k) and the corresponding estimation error covariance matrix are obtained by the least squares method: Where R(k) is the covariance of the measurement noise w(k), P(k) is the estimated error covariance matrix of x(k), is the estimated value of x(k); The multi-sensor feature information fusion equation is established based on the idea of ​​least squares method: In the formula, is the fused data, q is the number of data to be fused, W i is the weighting factor, X i is the data to be fused; S2022: Conduct correlation analysis with contact surface degradation. Based on the correlation between degradation characteristics and vibration behavior, divide the operating state of the bearing into three degradation stages: no damage, damage initiation, and initiation and extension. Construct a staged evolution model of bearing vibration behavior. S3: Based on the phased evolution model of bearing vibration behavior obtained in S2, a vibration behavior prediction based on the ARIMA-LSTM algorithm is constructed; S301: ARIMS-LSTM algorithm is used to predict the bearing vibration behavior at different operating stages; the key calculation formula of ARIMS-LSTM algorithm is as follows: Is to predict the time series X through the ARIMA model t The prediction result at time t is calculated as follows: Is to predict the time series X through the LSTM model t The prediction result at time t is calculated as follows: e LSTM (t) is the residual correction performed by the LSTM model by learning the prediction error of the ARIMA model. The calculation method is as follows: Final prediction results It is the sum of the ARIMA model prediction value and the LSTM model residual prediction value, calculated as follows: S302: Select MSE, RMSE and MAE as indicators for evaluating the difference between the predicted value and the true value, and optimize and adjust the prediction model; the calculation formulas of MSE, RMSE and MAE are as follows: In the formula, m is the number of samples, y is j is the true value of the jth sample, is the predicted value of the jth sample.

2. A system for predicting the vibration behavior of rolling bearings based on the analysis of vibration evolution during service, characterized in that: The system includes a control module, a drive module, a test acquisition module and a calculation and analysis module. The test acquisition module is connected to the calculation and analysis module; the control module is connected to the drive module and the test acquisition module respectively; The calculation and analysis module includes an evolution feature calculation module, an evolution process stage analysis module, and a vibration behavior prediction module which are connected in sequence; the control module controls the operation of the drive module and sends signals to the test acquisition module to collect bearing acceleration, inner ring displacement, and retainer displacement signals, which are sent to the calculation and analysis module to be calculated and analyzed in sequence through the evolution feature calculation module, the evolution process stage analysis module, and the vibration behavior prediction module.

3. The system according to claim 2, characterized in that The evolution feature calculation module performs time domain and frequency domain analysis on the bearing acceleration, inner ring displacement and cage displacement, extracts feature values, and obtains feature information for characterizing the bearing vibration evolution.

4. The system according to claim 2, characterized in that The evolution process phase analysis module adopts the least square method multi-source data fusion method, combined with the contact surface degradation analysis, to divide the evolution process into stages.

5. The system according to claim 2, characterized in that The vibration behavior prediction module uses the ARIMA-LSTM algorithm to predict the vibration data of acceleration and inner ring displacement at different operating stages.

Images