Rotary machinery residual life twin prediction method based on physical co-evolution

Abstract

The invention discloses a rotary machine residual life twin prediction method based on physical co-evolution, and the method comprises the steps: collecting a state signal of a bearing through a vibration sensor, and obtaining the operation condition parameters of the bearing; evaluating the bearing state by using an initial degradation evaluation method, detecting the bearing state in combination with an anomaly detection algorithm, and when a detection index value is greater than a preset evaluation threshold value, entering a degradation stage of the bearing; establishing a bearing dynamic model and a digital state virtual model, and updating state information of the digital state virtual model by using state parameters obtained by the bearing dynamic model; and finally predicting the residual service life of the bearing by combining the bearing state information extracted by the physical information neural network with the residual service life prediction network. According to the invention, real-time monitoring and safety and reliability evaluation of the mechanical state are realized, so that the intelligence, self-adaptability and result interpretability of mechanical equipment state evaluation are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mechanical equipment state evaluation, in particular to a twin method for predicting the residual life of rotating machinery based on physical co-evolution. BACKGROUND

[0002] Current mechanical equipment residual life prediction methods mainly include: 1. Based on mathematical probability distribution, that is, by measuring the geometric space distance of state parameter probability distribution of rotating machinery at different times to judge whether the mechanical state is normal; 2. Use the related mathematical and physical equations of rotating machinery to establish a dynamic model, and evaluate the use reliability of the rotating machinery by simulating the state parameters of the rotating machinery through the dynamic model; 3. Based on data-driven method, that is, using intelligent algorithm combined with a large amount of labeled full-life state data to establish a reliability evaluation model.

[0003] However, method 1 assumes that the state parameters of the rotating machinery follow a specific probability distribution, but in actual application, the state change of the mechanical system can be very complex and difficult to accurately describe with a simple probability distribution model; At the same time, the probability distribution at a single time may ignore the dynamic change information in the time series, resulting in lag or deviation in the judgment of the mechanical state. The accuracy of the model established by method 2 depends on the accurate parameter input, but in actual application, many parameters of the mechanical system are difficult to accurately obtain through conventional means; At the same time, affected by the complexity of the machinery, the modeling process is easily limited by the simplifying assumptions, and cannot fully reflect the nonlinear, time-varying characteristics and complex interactions in the actual working condition. Method 3 is highly dependent on the quality and quantity of data, and is prone to overfitting when the amount of data is small or the data characteristics are insufficient, which cannot adapt to the changes of the actual working condition, resulting in distorted evaluation results. The safety and reliability of rotating machinery in use is closely related to its running state, and the running state can be directly reflected by the real-time collected state parameters. The state change of the rotating machinery in the running process is essentially the external manifestation of its internal dynamics characteristics, so a virtual state twin model that can map the internal dynamics characteristics of the rotating machinery can be established, and the real-time collected state parameters of the machinery are interacted with the twin state information of the virtual model in real time, so as to realize the accurate characterization and safety and reliability prediction of the rotating machinery state. SUMMARY

[0004] The purpose of the present application is to provide a twin method for predicting the residual life of rotating machinery based on physical co-evolution, which realizes real-time monitoring and safety and reliability evaluation of the mechanical state, thereby effectively improving the intelligence, adaptability and result interpretability of the mechanical equipment state evaluation.

[0005] To achieve the above purpose, the present application provides the following scheme: A twin method for predicting the residual life of rotating machinery based on physical co-evolution, comprising the following steps: S1, collecting the state signal of the bearing between the vibration sensors, and obtaining the operating condition parameters of the bearing; S2, evaluating the bearing state by using an initial degradation evaluation method, that is, calculating the root mean square error between the fitting distributions as a degradation detection index, combining an anomaly detection algorithm to detect the bearing state, and when the detection index value is greater than a preset evaluation threshold, the bearing enters a degradation stage; S3, establishing a bearing dynamics model based on a bearing dynamics equation, simultaneously establishing a digital state virtual model based on a physical information neural network, and then updating the state information of the digital state virtual model by using the state parameters obtained from the bearing dynamics model, to realize real-time interaction between virtual reality and the physical system; S4, establishing a remaining useful life prediction network, and finally predicting the remaining useful life of the bearing by combining the bearing state information extracted by the physical information neural network with the remaining useful life prediction network, and when the remaining useful life of the bearing reaches a failure threshold, the bearing is maintained.

[0006] Preferably, in S2, the initial degradation evaluation method for evaluating the bearing state specifically includes: First, the state signal collected at different times t is fitted with a Gaussian distribution , and the RMS between the state distributions at different times is calculated as a degradation index, that is, , wherein is the state distribution at time t+1, is the root mean square error between the state distribution at the current time t and the state distribution at the next time t+1; the real-time is detected by using an anomaly detection algorithm, and when is greater than a detection threshold, the time t is the initial degradation point of the bearing.

[0007] Preferably, the real-time is detected by using an anomaly detection algorithm, and the detection formula is as follows: (3) , wherein is the median absolute deviation, is the median of the observed RMS, and when , the bearing starts to degrade.

[0008] Preferably, in S3, it further includes: In the bearing dynamics modeling, first, the real-time vibration signal collected is used to obtain the operating condition parameters of the bearing; then the vibration signal is subjected to Fourier transform and displacement conversion, to obtain the bearing characteristic frequency and the vibration displacement According to the characteristic frequency, the real-time state of the bearing is judged, and the vibration displacement is converted into the fault crack depth; finally, the obtained bearing state and fault crack depth are combined with the rolling bearing motion differential equation and the PINN network to perform twinning on the state signal of the bearing to obtain a twinned state signal .

[0009] Preferably, in S3, a bearing dynamics model is established based on a bearing dynamics equation, and specifically includes: In a fixed coordinate system, without considering the friction force, considering the radial load and the contact force, the motion differential equation of the inner ring of the bearing is: (4) Without considering the friction force, considering the contact force, the motion differential equation of the outer ring of the bearing is: (5) The torsional vibration equation of the bearing is as follows: (6) Wherein, Fox is the x-direction Hertz contact force of the outer ring of the bearing, Fix is the x-direction Hertz contact force of the inner ring of the bearing, Foy is the y-direction Hertz contact force of the outer ring of the bearing, Fiy is the y-direction Hertz contact force of the inner ring of the bearing, is the x-direction acceleration of the inner ring of the bearing, is the x-direction velocity of the inner ring of the bearing, is the x-direction displacement of the inner ring of the bearing, is the y-direction acceleration of the inner ring of the bearing, is the y-direction velocity, is the y-direction displacement of the inner ring of the bearing, is the damping coefficient of the inner ring of the bearing, c o is the damping coefficient of the outer ring of the bearing, m i is the mass of the inner ring of the bearing, m o is the mass of the outer ring of the bearing, k o is the contact stiffness of the outer ring of the bearing, k i is the contact stiffness of the inner ring of the bearing, J i is the moment of inertia of the inner ring of the bearing, J o is the moment of inertia of the outer ring of the bearing, is the angular acceleration of the inner ring, is the angular velocity of the inner ring, is the angular displacement of the inner ring, is the angular acceleration of the outer ring, is the angular velocity of the outer ring, is the angular displacement of the outer ring.

[0010] Preferably, in S3-S4, the state parameter obtained by using the bearing dynamics model updates the state information of the digital state virtual model, specifically including: The network parameters of the digital state virtual model are updated using the following loss function and back propagation algorithm to obtain the rolling bearing state twin model: The loss function of the rolling bearing state twin model is: (10) The state data loss is: (11) Wherein, respectively represent the supervision learning loss of the inner ring and the outer ring, is the inner ring dynamics constraint loss, is the outer ring dynamics constraint loss, is the twin state signal loss, is the inner ring initial condition loss, is the outer ring initial condition loss, is the outer ring fixed time displacement boundary condition loss, is the force boundary condition loss, is the predicted remaining useful life value loss, is the partial differential loss coefficient, is the initial condition loss coefficient, is the boundary condition loss coefficient, is the total length of the state signal, is the running time of the nth state signal point; at the same time, the network parameters of the remaining useful life prediction network are updated in the updating process.

[0011] The application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the physical co-evolution-based rotating machinery remaining life prediction twin method according to any one of the above.

[0012] According to the embodiments of the application, the following technical effects are achieved: (1) The application provides a physical co-evolution-based rotating machinery reliability prediction and evaluation method, which can realize real-time monitoring and safe and reliable evaluation of the mechanical state without relying on a large number of fault samples, high-quality data and expert knowledge for parameter calibration and threshold setting, thereby effectively improving the intelligence, self-adaptability and result interpretability of the mechanical equipment state evaluation, and providing more accurate and reliable support for fault diagnosis and maintenance decision of industrial equipment. (2) This invention provides an algorithm for predicting the remaining service life of rotating machinery. In the case of insufficient offline tag data, low digital quality, and no expert knowledge, the algorithm uses a small amount of historical data and real-time status data to update the parameters of the state twin model and predict the remaining effective service life of the equipment. This improves the working condition adaptability and intelligence level of the reliability assessment method and enhances the effectiveness of machinery maintenance decisions. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a diagram showing the main framework of the remaining useful life prediction algorithm used in the method of this invention. Figure 2 A flowchart illustrating a twin method for predicting the remaining life of rotating machinery based on physical co-evolution provided by the present invention; Figure 3 Flowchart of the dynamic state twinning algorithm of this invention; Figure 4 This is a block diagram of the collaborative evolution digital twin prediction algorithm of the present invention; Figure 5 This is a block diagram of the training architecture for state twinning and lifetime prediction in this invention. Figure 6 The image shows the normal state signal and signal spectrum of the twin under the 1800 rpm operating condition in this embodiment of the invention. Figure 7 The image shows the twin outer ring fault status signal and signal spectrum under 1800 rpm operating conditions in this embodiment of the invention. Figure 8 This is a diagram showing the twin inner ring fault status signal and signal spectrum under 1800 rpm operating conditions in an embodiment of the present invention. Figure 9 The image shows the twin inner ring fault status signal and signal spectrum under 2400 rpm operating conditions in this embodiment of the invention. Figure 10 The image shows the twin outer ring fault status signal and signal spectrum diagram under the 2400 rpm operating condition in this embodiment of the invention. Figure 11 The image shows the twin inner ring fault status signal and signal spectrum under 2400 rpm operating conditions in this embodiment of the invention. Wherein, (a) is the horizontal vibration signal, (b) is the spectrum of the horizontal vibration signal, (c) is the vertical vibration signal, and (d) is the spectrum of the vertical vibration signal. Figure 12 This is a diagram showing the initial degradation classification and remaining life prediction results of bearing 1 in an embodiment of the present invention; Figure 13 This is a diagram showing the initial degradation classification and remaining life prediction results of bearing 2 in an embodiment of the present invention; Figure 14 This is a diagram showing the initial degradation classification and remaining life prediction results of bearing 3 in an embodiment of the present invention; Figure 15 This is a diagram showing the initial degradation classification and remaining life prediction results of bearing 4 in an embodiment of the present invention; (a) represents the full life cycle signal, (b) represents the initial degradation point assessment, and (c) represents the remaining useful life assessment. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] Example 1 like Figure 1 As shown, the present invention provides a twin method for predicting the remaining life of rotating machinery based on physical co-evolution, comprising the following steps: S1. Collect bearing status signals through vibration sensors and obtain bearing operating parameters; S2. The bearing condition is evaluated using the initial degradation assessment method, that is, the root mean square error between the fitted distributions is calculated as the degradation detection index, and the bearing condition is detected by combining the anomaly detection algorithm. When the detection index value is greater than the preset evaluation threshold, the bearing enters the degradation stage. S3. Establish a bearing dynamics model based on the bearing dynamics equation, and at the same time establish a digital state virtual model based on the physical information neural network. Then, use the state parameters obtained from the bearing dynamics model to update the state information of the digital state virtual model to realize real-time interaction between virtual reality and physical system. S4. Establish a remaining service life prediction network. Finally, combine the bearing status information extracted by the physical information neural network with the remaining service life prediction network to predict the remaining service life of the bearing. When the remaining service life of the bearing reaches the failure threshold, condition maintenance is performed on the bearing.

[0018] The method of this invention specifically includes: 1. State Twin Remaining Useful Life Prediction Algorithm Figure 1 The main framework of the remaining service life prediction algorithm involves directly acquiring and storing the bearing's state signals using vibration sensors. An initial degradation assessment method is then used to evaluate the bearing's condition, specifically by calculating the root mean square error (RMS) between fitted distributions as a degradation detection index. This is combined with an anomaly detection algorithm to detect the bearing's condition; when the detection index value exceeds an evaluation threshold, the bearing enters the degradation stage. Finally, a bearing dynamics model and a physical information neural network interact to generate a virtual digital model of the bearing's real-time state. The bearing state information extracted from the physical information network is then combined with a life prediction network to predict the bearing's remaining service life. When the bearing's remaining service life reaches the failure threshold, condition-based maintenance is performed.

[0019] Figure 2 The main process for predicting the remaining service life of bearings includes initial degradation point assessment and remaining service life prediction. In the initial degradation point assessment, the state information collected at different times t is first analyzed. Perform Gaussian distribution fitting The degradation index is calculated by measuring the RMS of the state distributions at different time points. Using anomaly detection models For real time When testing is conducted, If the value exceeds the detection threshold, then time t is considered the initial degradation point of the bearing. After the initial degradation point assessment is completed, the bearing enters the degradation stage, at which point the remaining service life algorithm is used to predict the remaining service life. First, a physical information neural network is used. Using status signals Comparison of bearing dynamics model with operating condition parameters By performing differential calculations, the twin state of the bearing is obtained. Then the twin state The bearing state characteristics are derived by inputting the data into the physical information model and utilizing the information interaction between the dynamic model and the physical information network. Then the extracted state features Input into the remaining useful life prediction network To predict the remaining service life of the bearing Finally, based on the predicted lifespan value With failure threshold The algorithm determines whether a bearing is nearing failure; when the failure threshold is exceeded, the bearing fails, and the entire bearing life is predicted. The overall algorithm can be divided into three parts: (1) abnormal state detection; (2) dynamic modeling; and (3) solving the physical information neural network and predicting the remaining service life. The algorithm flow for solving the dynamic model is as follows: Figure 3 As shown, the remaining useful life prediction process is as follows: Figure 4 As shown.

[0020] The detailed process for initial degradation point assessment and remaining useful life prediction is as follows: (1) Abnormal state detection Using the fitted Gaussian distribution of the vibration signal Combined with anomaly detection algorithms abnormality , It can be derived from the average of neighboring data. and standard deviation To measure. According to the 3-sigma criterion, if for data at time node t... exist So, the current time value of the task If an anomaly is found, the current time t is considered the fault point in the fault diagnosis.

[0021] (1) Assuming there existed before the current observation point 1 RMS data point, and sort these observations to obtain ,when When it is an odd number The median was obtained as ,when When it is even The median is .

[0022] Using normalized median absolute deviation in anomaly detection to replace standard deviation .

[0023] (2) The outlier detection formula is shown below: (3) During state detection, the RMS (Resolution Mean Square) between the state distribution at the current time t and the state distribution at the next time t+1 is used as the anomaly judgment value. The bearing begins to degrade.

[0024] (2) Bearing dynamics modeling In dynamic modeling, the bearing's operating parameters are first obtained based on the acquired real-time vibration signals. Then, the vibration signals are analyzed... Fourier transform and shift transform are performed to obtain the bearing characteristic frequencies. With vibration displacement The real-time state of the bearing is determined based on its characteristic frequency, and converted into fault crack depth based on vibration displacement. Finally, the obtained bearing state and fault crack depth are combined with the rolling bearing motion differential equation and the bearing state signal from the PINN network to generate a twinned state signal. .

[0025] A bearing dynamic model is established using the bearing's differential equation of motion. In a fixed coordinate system, neglecting friction and considering only radial load and contact force, the differential equation of motion for the bearing's inner ring is: (4) Neglecting friction and considering only contact forces, the differential equation of motion for the outer ring of the bearing is: (5) The equation for the torsional vibration of the bearing is as follows: (6) in, Fox This refers to the Hertzian contact force in the x-direction of the bearing outer ring. Fix This represents the Hertzian contact force in the x-direction of the bearing inner ring. Foy This represents the Hertzian contact force in the y-direction of the bearing outer ring. Fiy This represents the Hertzian contact force in the y-direction of the bearing inner ring. The acceleration in the x-direction of the inner ring of the bearing is... The velocity of the inner ring of the bearing in the x direction is... This represents the displacement of the bearing inner ring in the x-direction. The acceleration in the y-direction of the inner ring of the bearing is... The velocity is in the y-direction. This represents the displacement of the bearing's inner ring in the y-direction. This is the damping coefficient of the bearing inner ring. c o This is the damping coefficient of the bearing outer ring. m i For the quality of the bearing inner ring, m o For the quality of the bearing outer ring, k o This refers to the contact stiffness of the bearing outer ring. k i This refers to the contact stiffness of the bearing inner ring. J i This represents the moment of inertia of the inner ring of the bearing. J oThis refers to the rotational inertia of the outer ring of the bearing. For inner angular acceleration, The inner angular velocity, For inner angular displacement, For outer angular acceleration, The outer angular velocity, This represents the outer angular displacement.

[0026] (3) Solving physical information neural networks and predicting remaining useful life Based on bearing dynamics formulas and physical information neural network theory, the torsional vibration equation and motion differential equation of rolling bearings are used as physical constraints for the PINN, forcing the neural network to output the physical information of the bearing. Combined with the bearing's vibration signal, the network parameters are optimized through supervised learning. The network input is time. The collected vibration signals Bearing parameters (mass) Moment of inertia Damping coefficient Stiffness (and the structural parameters of the bearing), the output is the x and y displacements of the inner and outer rings, as well as the angular displacements of the inner and outer rings, respectively. .

[0027] Based on the bearing dynamics equations, the motion differential operator for the inner ring of the bearing is: (7) The outer-circle motion differential operator is: (8) The differential operator for torsional vibration is: (9) The model loss function is: (10) The state data loss is: (11) These represent the supervised learning losses of the inner and outer circles, respectively.

[0028] Dynamic constraint loss: (12) (13) (14) in, For the inner circle dynamic constraint loss, For the constraint loss of the outer ring dynamics, For twin-state signal loss, For the motion differential operator of the inner ring of the bearing. For the outer-circle motion differential operator, For torsional vibration differential operators.

[0029] The initial conditional loss is: (15) Displacement boundary condition loss when the outer ring is fixed: (16) Force boundary condition loss: (17) in, For the initial condition loss of the inner circle, Loss due to initial conditions of the outer ring. This refers to the displacement boundary condition loss when the outer ring is fixed. Force boundary condition loss, To predict the loss of remaining useful life, The partial differential loss coefficient, The initial condition loss coefficient, The boundary condition loss coefficient, This refers to the bearing operating time.

[0030] The PINN network parameters are updated using the above loss function and backpropagation algorithm to obtain a rolling bearing state twin model. Simultaneously, the network parameters of the rolling bearing lifetime prediction network are updated during the update process, specifically calculating the remaining lifetime value. Compared with predicted lifespan value loss The training framework for state twin and lifetime prediction models is as follows: Figure 5 As shown.

[0031] (18) 2. Beneficial effects: Based on the bearing's characteristic frequencies, calculations show that at 1800 rpm (30 Hz), the outer ring fault frequency is 92.49 Hz and the inner ring fault frequency is 147.51 Hz; at 2400 rpm (40 Hz), the outer ring fault frequency is 123.32 Hz and the inner ring fault frequency is 196.68 Hz. Using the proposed twin prediction method to solve the dynamic model, the state data distributions at 1800 rpm and 2400 rpm are as follows: Figures 6-11As shown in (a)-(d), at 1800 rpm, the twin state signal frequency is 30 Hz, the outer ring fault frequency is 92 Hz, and the inner ring fault frequency is 147 Hz. At 2400 rpm, the twin state signal frequency is 40 Hz, the outer ring fault frequency is 123 Hz, and the inner ring fault frequency is 197 Hz. The twin signal fault characteristics are close to the theoretical fault frequencies, indicating the effectiveness of the proposed method, which can twinnize characteristics identical to those of the actual state.

[0032] The remaining service life value of the bearing is predicted by the twin prediction network as follows: Figures 12-15 As shown in (a)-(c), the obtained prediction evaluation indicators are shown in Table 1 (Remaining Life Prediction Evaluation Indicators). RMSE is the square root of the mean of the squares of the differences between the predicted and actual values, MAE is the mean of the absolute differences between the predicted and actual values, and R² is the coefficient of determination used to measure the prediction accuracy of the method. The closer RMSE and MAE are to 0, the more accurate the prediction performance. 2 The closer to 1, the more accurate the prediction performance.

[0033] Table 1

[0034] As shown by the initial degradation segmentation curve, the designed evaluation algorithm can accurately segment the normal degradation stage of the bearing based on changes in the bearing signal state. The remaining life prediction curves for different bearings show that the proposed prediction algorithm can accurately assess the remaining life of the bearing. The trends between the predicted and actual life values ​​are very close. Before the bearing completely degrades and fails (failure threshold set to 0.2), maintenance and replacement can be performed based on the life warning value. Table 1 shows that the proposed prediction algorithm predicts RMSE values ​​for different bearings that are generally less than 0.05, MAE values ​​that are generally less than 0.04, and R² values ​​that are generally greater than 0.9. These indicators demonstrate that the algorithm can accurately predict the remaining life of the bearing and determine the time of complete bearing failure. These results prove that the proposed algorithm can improve the intelligence and accuracy of mechanical equipment operation and maintenance, and reduce the failure rate and maintenance costs of mechanical equipment.

[0035] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a twin method for predicting the remaining life of rotating machinery based on physical co-evolution as described above.

[0036] This invention combines real-time state data of rotating machinery, anomaly detection algorithms, physical dynamics models, and Physical Information Neural Networks (PINN) to establish a state twin remaining service life prediction model. The model consists of two stages: initial degradation point identification and remaining service life prediction. In the initial degradation identification stage, the root mean square error (RMS) of the Gaussian fitting distribution of real-time state data and the anomaly detection algorithm are used to divide the bearing into normal and degradation stages, obtaining the initial degradation point. In the remaining service life prediction stage, the automatic differentiation process of the backpropagation algorithm interacts with the mechanical dynamics model. First, a state dynamics simulation model is established using the physical equations of motion of the rotating machinery, and a digital state virtual model is established based on PINN. Then, the state information of the virtual digital model is updated using the real-time state parameters of the rotating machinery and the state parameters obtained from the dynamics model, realizing real-time interaction between virtual reality and the physical system. Finally, the remaining service life value is evaluated by extracting state features through PINN and using the remaining service life prediction network to assess the remaining service life of the rotating machinery.

[0037] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0038] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A twin method for predicting the remaining life of rotating machinery based on physical co-evolution, characterized in that, Includes the following steps: S1. Collect bearing status signals through vibration sensors and obtain bearing operating parameters; S2. The bearing condition is evaluated using the initial degradation assessment method, that is, the root mean square error between the fitted distributions is calculated as the degradation detection index, and the bearing condition is detected by combining the anomaly detection algorithm. When the detection index value is greater than the preset evaluation threshold, the bearing enters the degradation stage. S3. Establish a bearing dynamics model based on the bearing dynamics equation, and at the same time establish a digital state virtual model based on the physical information neural network. Then, use the state parameters obtained from the bearing dynamics model to update the state information of the digital state virtual model to realize real-time interaction between virtual reality and physical system. S4. Establish a remaining service life prediction network. Finally, combine the bearing status information extracted by the physical information neural network with the remaining service life prediction network to predict the remaining service life of the bearing. When the remaining service life of the bearing reaches the failure threshold, condition maintenance is performed on the bearing.

2. The twin method for predicting the remaining life of rotating machinery based on physical co-evolution as described in claim 1, characterized in that, In step S2, the bearing condition is evaluated using an initial degradation assessment method, specifically including: First, the state signals collected at different times t are analyzed. Perform Gaussian distribution fitting The degradation index is calculated by measuring the RMS of the state distributions at different time points. ,in, The state distribution at time t+1 The root mean square error between the state distribution at the current time t and the state distribution at the next time t+1 is used; an anomaly detection algorithm is employed to analyze the real-time... When testing is conducted, If the value exceeds the detection threshold, then time t is the initial degradation point of the bearing.

3. The twin method for predicting the remaining life of rotating machinery based on physical co-evolution as described in claim 1, characterized in that, Using anomaly detection algorithms for real-time The detection process is performed using the following formula: (3) in, This represents the absolute deviation of the median. To observe the median of RMS, when The bearing begins to degrade.

4. The twin method for predicting the remaining life of rotating machinery based on physical co-evolution as described in claim 1, characterized in that, S3 also includes: In bearing dynamics modeling, the bearing's operating parameters are first obtained based on the acquired real-time vibration signals; then the vibration signals are analyzed... Fourier transform and shift transform are performed to obtain the bearing characteristic frequencies. With vibration displacement The real-time state of the bearing is determined based on its characteristic frequency, and then converted into the fault crack depth based on the vibration displacement. Finally, the obtained bearing state and fault crack depth are combined with the differential equation of motion of the rolling bearing and the PINN network to generate a twinned state signal of the bearing. .

5. The twin method for predicting the remaining life of rotating machinery based on physical co-evolution as described in claim 4, characterized in that, In S3, a bearing dynamics model is established based on the bearing dynamics equations, specifically including: In a fixed coordinate system, neglecting friction but considering radial load and contact force, the differential equation of motion for the inner ring of the bearing is: (4) Neglecting friction but considering contact forces, the differential equation of motion for the outer ring of the bearing is: (5) The equation for the torsional vibration of the bearing is as follows: (6) in, Fox This refers to the Hertzian contact force in the x-direction of the bearing outer ring. Fix This represents the Hertzian contact force in the x-direction of the bearing inner ring. Foy This represents the Hertzian contact force in the y-direction of the bearing outer ring. Fiy This represents the Hertzian contact force in the y-direction of the bearing inner ring. The acceleration in the x-direction of the inner ring of the bearing is... The velocity of the inner ring of the bearing in the x direction is... This represents the displacement of the bearing inner ring in the x-direction. The acceleration in the y-direction of the inner ring of the bearing is... The velocity is in the y-direction. This represents the displacement of the bearing's inner ring in the y-direction. This is the damping coefficient of the bearing inner ring. c o This is the damping coefficient of the bearing outer ring. m i For the quality of the bearing inner ring, m o For the quality of the bearing outer ring, k o This refers to the contact stiffness of the bearing outer ring. k i This refers to the contact stiffness of the bearing inner ring. J i This represents the moment of inertia of the inner ring of the bearing. J o This refers to the rotational inertia of the outer ring of the bearing. For inner angular acceleration, The inner angular velocity, For inner angular displacement, For outer angular acceleration, The outer angular velocity, This represents the outer angular displacement.

6. The twin method for predicting the remaining life of rotating machinery based on physical co-evolution as described in claim 1, characterized in that, In steps S3-S4, updating the state information of the digital state virtual model using the state parameters obtained from the bearing dynamics model specifically includes: The network parameters of the digital state virtual model are updated using the following loss function and backpropagation algorithm to obtain the rolling bearing state twin model: The loss function for the rolling bearing state twin model is: (10) The state data loss is: (11) in, These represent the supervised learning losses of the inner and outer circles, respectively. For the inner circle dynamic constraint loss, For the constraint loss of the outer ring dynamics, For twin-state signal loss, For the initial condition loss of the inner circle, Loss due to initial conditions of the outer ring. This refers to the displacement boundary condition loss when the outer ring is fixed. Force boundary condition loss, To predict the loss of remaining useful life, The partial differential loss coefficient, The initial condition loss coefficient, The boundary condition loss coefficient, The total length of the state signal. The runtime of the nth state signal point is given; simultaneously, the network parameters of the remaining lifetime prediction network are updated during the update process.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a twin method for predicting the remaining life of rotating machinery based on physical co-evolution as described in any one of claims 1 to 6.

Images