A digital twin guided early fault signal denoising method for rolling bearings

Abstract

The application relates to the technical field of signal denoising, and discloses a digital twin guided early fault signal denoising method for rolling bearings, which comprises the following steps: step S1, inputting the running conditions of an experiment table and bearing parameters into a bearing dynamics equation, combining with historical data of real equipment bearings, constructing a digital twin model of the rolling bearing, and generating a simulation vibration signal under a healthy state; and step S2, obtaining a small amount of measured signals to update the digital twin model; wherein the small amount of real-time monitoring data at the initial stage of operation is combined with the simulation vibration signal generated under the healthy state. The digital twin guided early fault signal denoising method for rolling bearings generates high-quality simulation data, for the first time applies the digital twin technology to the early fault signal denoising of the rolling bearing, and verifies the feasibility of generating simulation data highly consistent with actual operation by using the digital twin for signal processing under a complex industrial environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of signal denoising, in particular to a digital twin guided early fault signal denoising method for rolling bearings. BACKGROUND

[0002] In modern manufacturing, bearings, as a key component of rotating machinery, are widely used in various equipment and systems, such as automobile engines, aerospace equipment, industrial robots, and energy production devices. The stable operation of bearings directly affects the performance, efficiency, and safety of the entire mechanical system. Therefore, early fault diagnosis of bearings is of great practical significance and economic value to ensure their safe and reliable service.

[0003] However, the factory environment is complex and variable, and the vibration signals generated during the operation of bearings are significantly affected by various environmental factors. Firstly, various vibration sources will introduce interference signals. Secondly, factors such as friction will generate noise. These influences often cause the amplitude and frequency of the vibration signal to deviate from the ideal state, posing significant challenges to fault detection.

[0004] Therefore, it is necessary to provide a digital twin guided early fault signal denoising method for rolling bearings. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a digital twin guided early fault signal denoising method for rolling bearings, which has the advantages of effectively eliminating the noise of early fault signals and highlighting the fault features, solving the problems raised in the background art.

[0006] The present application provides the following technical solution: a digital twin guided early fault signal denoising method for rolling bearings, comprising the following steps:

[0007] Step S1, input the operating conditions of the test bench and the bearing parameters into the bearing dynamics equation and combine the real equipment bearing historical data to construct a rolling bearing digital twin model and generate a simulated vibration signal in a healthy state;

[0008] Step S2, update the digital twin model with a small amount of measured signals; wherein, combine a small amount of real-time monitoring data at the initial stage of operation with the generated simulated vibration signal in a healthy state, update the digital twin model by using cosine similarity, and inject early weak fault features in the updated model;

[0009] Step S3, construct a denoising algorithm based on an adversarial strategy, input the measured signals collected in real time into the generator to generate a preliminary denoising signal, then input it into the discriminator, and at the same time input the generated noise-free simulated signal containing early weak fault features as the discrimination signal to correct the denoising result;

[0010] Step S4, the value of the loss function gradually decreases with training and tends to be stable, and finally outputs the final denoising result.

[0011] Preferably, in the step S1, a bearing dynamics equation is constructed and solved in combination with mechanical knowledge to generate a bearing simulation signal; wherein the bearing dynamics equation is: .

[0012] Preferably, M, C and K respectively represent the total mass, damping coefficient and total contact stiffness of the system (the system includes rollers, bearings and bearing base); Z is the number of rolling bodies, and i is 1 to Z, and are the outer ring vibration accelerations in x direction and y direction at time t respectively; and are the load components respectively; (wherein 1.5 represents the load deformation of the ball bearing) is the contact deformation between the i-th rolling body and the raceway, which can be specifically represented as: .

[0013] Preferably, x and y are the outer ring vibration deformations in x direction and y direction at time t; c is the radial clearance of the rolling bearing; is the angular position of the i-th rolling body at time t, is the displacement excitation varying with the fault position, when the components of the rolling bearing are fault-free, ; λ is the effective contact coefficient of the rolling body, which can be represented as: .

[0014] Preferably, in the step S2, the cosine similarity is represented as:

[0015] , the represents the measured signal collected from the equipment, represents the simulation signal under the same working condition as .

[0016] Preferably, in the step S2, when injecting early weak fault features into the updated digital twin model, the way is to change , which can be specifically represented as:

[0017] inner ring fault ( ),

[0018]

[0019] outer ring fault ( ),

[0020]

[0021] Inner and outer ring compound fault,

[0022] .

[0023] Preferably, the and respectively represent the fault angle positions of the outer ring and the inner ring, L represents the fault width of each component of the rolling bearing, , and respectively represent the rolling body radius, the outer ring radius and the inner ring radius, (.) represents the remainder function.

[0024] Preferably, in the step S3, the optimization objective function of the denoising algorithm of the countermeasure strategy can be represented as: .

[0025] Preferably, the , and respectively represent the probability distribution of the measured signal, the simulation signal and , wherein can be represented as , and ε is a random number, ; , and respectively represent the probability value of the measured signal, the simulation signal and ; E represents expectation; is a constant, and the default value is 10; represents the two-norm.

[0026] Compared with the prior art, the present application has the following beneficial effects:

[0027] 1. The digital twin guided rolling bearing early fault signal denoising method applies digital twin technology to rolling bearing fault signal denoising, verifies the feasibility of using digital twin to generate simulation data highly consistent with actual operation for signal processing in complex industrial environments, improves denoising precision and robustness, and the proposed method uses digital twin to accurately simulate bearing operation state, effectively reduces the influence of environmental noise and interference on fault signals, makes the denoised signals closer to the real fault characteristics, and enhances the reliability of fault detection.

[0028] 2、The early fault signal denoising method of the digital twin guided rolling bearing realizes real-time fault diagnosis based on digital twin, realizes real-time synchronization of the running state of the physical bearing and the digital model by closely combining the digital twin model with the fault detection system, ensures timely capture and denoising processing of the fault signal, enables the fault detection system to quickly respond, meets the strict requirements of industrial sites on real-time monitoring and early warning, enhances the adaptability and robustness of the system, and dynamically adapts to different working environments and operating conditions through continuous updating of the digital twin model, realizes real-time monitoring of the bearing state and signal optimization, and ensures that the fault detection capability remains efficient and stable in a variable industrial environment. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0030] Figure 1 The denoising method flowchart of the present application;

[0031] Figure 2 The digital twin model updating architecture diagram of the present application;

[0032] Figure 3 The countermeasure algorithm structure diagram of the present application;

[0033] Figure 4 The three index test result diagram of the present application. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0035] Please refer to Figure 1 , Figure 2 , Figure 3 and Figure 4 , a digital twin guided rolling bearing early fault signal denoising method, comprising the following steps:

[0036] Step S1, input the running condition of the test bench and the bearing parameters into the bearing dynamics equation and combine the real equipment bearing historical data to construct a rolling bearing digital twin model, and generate a simulated vibration signal in a healthy state;

[0037] Step S2, a small amount of measured signal is obtained to update the digital twin model; wherein the early-stage real-time monitoring data is combined with the simulation vibration signal in the generated healthy state, the digital twin model is updated by using the cosine similarity, and the early-stage weak fault feature is injected into the updated model;

[0038] Step S3, a denoising algorithm based on an antagonistic strategy is constructed, the real-time collected measured signal is input into the generator to generate a preliminary denoised signal, then the preliminary denoised signal is input into the discriminator, and the simulation signal without noise and containing the early-stage weak fault feature is used as a discrimination signal to correct the denoising result;

[0039] Step S4, the value of the loss function gradually decreases and tends to be stable with training, and finally the final denoising result is output.

[0040] As a preferred technical solution of the application, in step S1, the bearing dynamics equation is constructed and solved in combination with mechanical knowledge to generate the bearing simulation signal; wherein the bearing dynamics equation is: , M, C and K represent the total mass, the damping coefficient and the total contact stiffness of the system (the system includes the roller, the bearing and the bearing base) respectively; Z is the number of rolling elements, i is 1 to Z, and are the outer ring vibration accelerations in the x direction and the y direction at time t respectively; and are the load components; (the load deformation of the ball bearing is represented by 1.5) is the contact deformation between the i th rolling element and the raceway, which can be specifically represented as: , x and y are the outer ring vibration deformations in the x direction and the y direction at time t; c is the radial clearance of the rolling bearing; is the angular position of the i th rolling element at time t, is the displacement excitation varying with the fault position, when the components of the rolling bearing are fault-free, ; λ is the effective contact coefficient of the rolling element, which can be represented as: , =2kg, =200Ns / m, =8, =0, =12,000N, = .

[0041] As a preferred technical solution of the application, in step S2, the cosine similarity is represented as: , represents the measured signal collected from the equipment, represents the simulation signal in the generated healthy state. The simulation signal under the same working condition is considered >0.6 represents that the model is reasonable.

[0042] As a preferred technical solution of the present application, in step S2, when injecting the early weak fault feature into the updated digital twin model, the mode is to change , which can be specifically represented as:

[0043] Inner ring fault ),

[0044]

[0045] Outer ring fault ),

[0046]

[0047] Inner and outer ring composite fault

[0048] , and are the fault angle positions of the outer ring and the inner ring respectively, L is the fault width of each part of the rolling bearing, , , and represent the radii of the rolling body, the outer ring and the inner ring respectively, (.) represents the remainder function, wherein =3.96, =19.9, =14.65, =0.1778mm.

[0049] As a preferred technical solution of the present application, in step S3, the training of the denoising algorithm of the countermeasure strategy involves three stages; in the first stage, the real-time collected measured signal is input into the generator to generate a preliminary denoised signal, and the noise-free simulation signal is used to train the discriminator; in the second stage, the preliminary denoised signal from the generator is input into the discriminator, so that the discriminator judges which signals are from the generator and which are noise-free simulation signals, and the distinguished signals from the generator are re-input into the generator for re-denoising and feature extraction; in the fourth stage, the work of the last stage is repeatedly performed until the discriminator cannot distinguish which signals are from the generator and which are noise-free simulation signals, wherein the optimization objective function can be represented as: .

[0050] As a preferred technical solution of the present application, , and represent the measured signal, the simulation signal and The probability distribution of the loss function value can be expressed as: P (L) = P (L | H) P (H) + P (L | F) P (F) wherein The probability distribution of the loss function value can be expressed as: P (L) = P (L | H) P (H) + P (L | F) P (F) wherein ε is a random number, ; , and represent the measured signal, the simulated signal and the probability value of the simulated signal, respectively; E represents expectation; is a constant, and the default value is 10; represents the two-norm.

[0051] As a preferred technical solution of the application, in step S4, as the loss function value tends to be stable, the training gradually converges, and the mean square error between the output denoising signal and the obvious signal of the fault feature gradually decreases and tends to be stable, finally outputting the final denoising result, and verifying the accuracy of the denoising result and the superiority of the method through time-domain-envelope analysis and signal feature index comparison on the denoising result.

[0052] The application uses a certain bearing test bench to verify the feasibility of the method proposed in the application:

[0053] According to the equipment parameters of a test bench, a bearing digital twin dynamics model is constructed and solved, and a PCB352C33 acceleration sensor is used to collect vibration data of a rotating machinery bearing. The rolling bearing is 6203-(SKF) and is divided into three categories, including three bearing states under two working conditions, specifically three working conditions of a rotating speed of 35 Hz and a load of 12 kN, a rotating speed of 37.5 Hz and a load of 11 kN, and a rotating speed of 30 Hz and a load of 0.5 kN. The bearing has three types of faults, i.e., an outer ring, an inner ring and a mixed inner and outer ring. The sampling frequency is 25.6 kHz, the sampling time is 1.28 s, data is measured once per minute as a sample, and the measured vibration signal type is acceleration. The sampling direction has 32678 sampling points per sampling point.

[0054] A small amount of measured data in a healthy state tested at the beginning is used to update the digital twin model. In the updated digital twin model, a fault feature is injected to generate a noise-free high-fidelity simulated signal in three types of fault states of an inner ring, an outer ring and a mixed inner and outer ring, and input into the adversarial strategy algorithm.

[0055] After denoising model training, the final denoising result is as follows: Figure 4 ​As shown, for the outer ring fault, the denoised signal has obvious repeated impacts excited by the early outer ring defect of the bearing, and the outer ring fault frequency and its harmonic spectrum peak are obvious, and due to the difference in centrifugal force in the actual system, we can also observe the part of the side frequency band near the outer ring fault frequency and its harmonic spectrum peak, for the inner ring fault, the denoised signal has obvious repeated impacts excited by the early inner ring defect of the bearing, and the inner ring fault frequency and its harmonic spectrum peak are obvious, and even due to the rotation modulation, there are more obvious side frequency bands near the inner ring fault frequency and its harmonic spectrum peak, for the inner and outer ring fault, the denoised signal retains the outer ring fault characteristics on the basis of the original signal while highlighting the inner ring fault characteristics.

[0056] In the index analysis, for the outer ring fault, the kurtosis of the denoised signal is improved by 479.42% compared with the original signal, the peak factor is improved by 170.62%, and the Gini index is improved by 165.00%. For the inner ring fault, the kurtosis of the denoised signal is improved by 242.80% compared with the original signal, the peak factor is improved by 138.25%, and the Gini index is improved by 114.93%. For the inner and outer ring fault, the kurtosis of the denoised signal is improved by 407.25% compared with the original signal, the peak factor is improved by 154.41%, and the Gini index is improved by 168.97%.

[0057] In the early weak fault scenario, the digital twin method can solve the problem that the signal collected in actual engineering contains strong noise, and the digital twin combined with the countermeasures is first applied to the field of signal denoising. In order to verify the effectiveness of the method proposed in the application, time domain-envelope analysis combined with three commonly used statistical indicators, namely kurtosis, peak factor and Gini index, is used to comprehensively evaluate the experimental effect provided in the embodiment, which is more robust than single method evaluation.

[0058] Although the embodiments of the present application have been shown and described, it can be understood by those skilled in the art that various changes, modifications, replacements and variations can be made to these embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for denoising early-stage fault signals of rolling bearings guided by digital twins, characterized in that: The following steps are involved: Step S1: Input the operating conditions and bearing parameters of the test bench into the bearing dynamics equation and combine them with the historical data of the real equipment bearing to build a digital twin model of the rolling bearing and generate a simulated vibration signal in a healthy state; Step S2: Acquire a small amount of measured signals to update the digital twin model. This involves combining a small amount of real-time monitoring data from the initial operation with simulated vibration signals generated in a healthy state. The digital twin model is updated using cosine similarity, and early, weak fault signatures are injected into the updated model. Step S3: Construct a denoising algorithm based on an adversarial strategy. Input the real-time measured signal into the generator to generate a preliminary denoised signal, which is then input into the discriminator. At the same time, the generated noise-free simulated signal containing early weak fault characteristics is used as the discrimination signal to correct the denoising result. In step S4, the value of the loss function gradually decreases and stabilizes during training, and the final denoising result is output.

2. The method for denoising early-stage rolling bearing fault signals guided by digital twins according to claim 1 is characterized in that: In step S1, the bearing dynamics equation is constructed and solved in combination with mechanical knowledge to generate a bearing simulation signal; wherein the bearing dynamics equation is: .

3. The method for denoising early-stage rolling bearing fault signals guided by digital twins according to claim 2 is characterized in that: The M, C and K represent the total mass, damping coefficient and total contact stiffness of the system respectively. The system includes rollers, bearings and bearing bases; Z is the number of rolling elements, i ranges from 1 to Z, and are the outer ring vibration accelerations in the x and y directions at time t respectively; and are the load components respectively; Where 1.5 represents the load deformation of the ball bearing, which is the contact deformation between the i-th rolling element and the raceway. It can be expressed as: .

4. The method for denoising rolling bearing early fault signals guided by digital twins according to claim 3 is characterized by: x and y are the vibration deformations of the outer ring in the x-direction and y-direction at time t; c is the radial clearance of the rolling bearing; is the angular position of the i-th rolling element at time t, The displacement excitation changes with the fault position. When the rolling bearing components are fault-free, ; λ is the effective contact coefficient of the rolling element, which can be expressed as: .

5. The method for denoising rolling bearing early fault signals guided by digital twins according to claim 1, characterized in that: In step S2, the cosine similarity is expressed as: , Indicates the measured signal collected from the device. Represents Simulation signals under the same working conditions.

6. The method for denoising early-stage rolling bearing fault signals guided by digital twins according to claim 1, characterized in that: In step S2, when injecting early weak fault features into the updated digital twin model, the method is to change , which can be specifically expressed as: Inner race fault , ， Inner and outer ring compound fault, 。 7. The method for denoising rolling bearing early fault signals guided by digital twins according to claim 6, characterized in that: described and are the fault angle positions of the outer ring and inner ring respectively, L is the fault width of each component of the rolling bearing, 、 and Represent the rolling element radius, outer ring radius and inner ring radius respectively. (.) represents the remainder function.

8. The method for denoising early-stage rolling bearing fault signals guided by digital twins according to claim 1, characterized in that: In step S3, the optimization objective function of the denoising algorithm of the adversarial strategy can be expressed as: 。 9. The method for denoising early-stage rolling bearing fault signals guided by digital twins according to claim 1, characterized in that: described 、 and Represent the measured signal, simulated signal and The probability distribution of It can be expressed as , ε is a random number, ; , and Represent the measured signal, simulated signal and The probability value of ; E represents the expectation; It is a constant, the default value is 10; Represents the second norm.

Images