Digital twin bearing fault diagnosis method based on virtual and real dynamic interaction

Abstract

The invention belongs to the technical field of a computer system based on a specific calculation model, relates to a virtual and real dynamic interaction-based digital twin bearing fault diagnosis method, and aims to solve the problem that a bearing system in large rotary mechanical equipment is large in fault hidden danger and avoid serious consequences caused by faults. A generally used bearing is in a high-precision standard, so that system signal data under bearing fault operation is not easy to collect in a factory, and the unbalanced data sample distribution directly leads to finiteness of fault diagnosis in a real scene only based on a big data driving mode classification method, so that the system signal data cannot be easily collected in the factory. The method has good scene working condition self-adaption, can be applied to fault diagnosis of a bearing system in different industrial scenes, fully excavates limited industrial data information and rich physical priori knowledge, and ensures safe and reliable operation of an important rotating mechanical system through a digital twinning technology.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer systems based on specific computing models, and in particular relates to a digital twin bearing fault diagnosis method based on dynamic interaction between virtual and reality. Background Art

[0002] With the advent of the Industrial 4.0 era, intelligent manufacturing and equipment health management have become important directions for industrial development. In rotating machinery equipment, bearings are key components. Bearing failures often lead to equipment downtime and huge economic losses. Therefore, bearing fault diagnosis technology has broad application prospects and research value in the industry.

[0003] Most of the existing big data-driven methods collect a large amount of equipment operation data, use machine learning and data mining technology to analyze and model the sensor signal data, and thus predict and detect potential faults. However, this method relies on a large number of fault data samples, and in actual industrial applications, it is very difficult to obtain enough fault samples, resulting in an imbalanced data sample problem, which limits the application effect of big data-driven methods in practice. However, the physical model-driven method is based on the physical characteristics and operation mechanism of the equipment, and simulates and analyzes by building a mathematical model. This method can provide an in-depth understanding of equipment failures, but building a high-precision physical model often requires complex calculations and expertise, and it is difficult to update and adjust in real time in actual applications.

[0004] In order to overcome the limitations of the above methods, digital twin technology came into being. Digital twin technology builds a virtual model of physical equipment and combines it with real-time sensor data to achieve real-time monitoring and prediction of equipment status. The digital twin bearing fault diagnosis method based on dynamic interaction between virtual and real life combines the accuracy of physical models and the real-time nature of big data analysis, and can effectively perform bearing fault diagnosis under zero fault samples.

[0005] However, current digital twin technology faces multiple challenges in building high-precision physical models and real-time updates. First, building high-precision physical models requires a deep understanding of the working principles and operating mechanisms of the equipment, which usually involves complex mathematical and physical modeling processes and requires a lot of computing resources. Second, real-time updates of models to reflect the latest status of the equipment require efficient data processing and analysis capabilities, which are difficult to achieve in actual industrial environments, especially when the data volume is large and the real-time requirements are high. In addition, traditional digital twin methods have limited effects when processing zero-fault samples, and it is difficult to make full use of limited fault data for effective fault diagnosis. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides a digital twin bearing fault diagnosis method based on dynamic interaction between virtual and reality to solve the problems described in the above background technology.

[0007] The purpose and efficacy of the digital twin bearing fault diagnosis method based on virtual and real dynamic interaction of the present invention are achieved by the following specific technical means:

[0008] S1: Construct a virtual physical model of the bearing and simulate and generate virtual signals;

[0009] S2: Install multi-source sensors on actual industrial equipment to collect real-time sound and vibration signal data of rotating machinery systems under normal operation;

[0010] S3: preprocessing the virtual and real signals obtained in step S1 and step S2, and performing time-frequency domain feature analysis on the preprocessed virtual and real signals;

[0011] S4: Evaluate the correlation error between the virtual signal and the real signal according to the correlation coefficient formula;

[0012] S5: The proxy model is constructed by the GRP method, and the proxy model is trained with the data generated by simulation to approximately describe the signal simulation behavior of the virtual model. Finally, the fitting accuracy of the proxy model is evaluated by MSE or R2.

[0013] S6: Taking minimizing the correlation error and the surrogate model fitting error as the optimization objective function, the firefly algorithm (FA) is combined to iteratively optimize the hyperparameters of the combination model and the dynamic model of the virtual model;

[0014] S7: The optimized hyperparameters obtained in step S6 and the optimized and updated virtual model can be regarded as a physical model that can simulate high-fidelity fault signals. The simulation generates high-fidelity fault signals and fault-free signals of normal operation of equipment in industrial scenarios. Through feature matching and pattern recognition methods, it can be determined whether there is a fault in the bearing of the rotating machinery.

[0015] Preferably, the virtual physical model of the bearing is constructed in step S1, and a virtual signal is generated by simulation, wherein the detailed virtual signal generation principle includes steps such as vibration impact signal, attenuated oscillation signal, alternating load and amplitude, and the process of generating sound signal:

[0016] 1. First generate a random vibration impact signal U impulse :

[0017]

[0018] Where N is the number of sample points to be simulated, δ(■) is the Dirac function, which is mainly used to describe the instantaneous impact occurring at a specific time point, and T impulse is the period of vibration shock;

[0019] 2. Design the attenuated oscillation signal s t :

[0020] s t (t) = exp(-Bt)·cos(2πf n t)

[0021] Where B is the attenuation coefficient, f n is the natural frequency.

[0022] 3. Determine the load M:

[0023] M=m·g·cos(2πFreq # t+Ψ0)

[0024] Where m is mass, g is the acceleration due to gravity, and Freq # is the system frequency, including the response frequency under no fault and different fault types, and Ψ0 is the initial phase. Among them, the inner and outer ring fault frequencies Freq io The calculation formula is as follows:

[0025]

[0026] Among them, rev is the rotating machine speed (unit: rpm), d is the diameter of the bearing rolling element, D m is the pitch diameter, α is the contact angle, and z is the number of rolling elements. The rolling element failure frequency Freq rolI The calculation formula is as follows:

[0027]

[0028] In addition, the failure frequency of the inner and outer cages Freq cage The calculation formula is as follows:

[0029]

[0030] 4. Determine the amplitude A m :

[0031] A m =τ·M

[0032] Where τ is the proportionality coefficient.

[0033] 5. Alternating load T: Alternating load T represents the load caused by system vibration and can be expressed as:

[0034] Ψ=2π(f0-Freq # )t+Ψ1, T

[0035] =m·e·ω 2 ·cos(Ψ)

[0036] Where f0 is the fundamental frequency of the system, Ψ1 is the phase offset, and e is the eccentricity;

[0037] 6. Alternating amplitude A t , impact amplitude A impulse And the total amplitude A:

[0038] A t =τ·T

[0039] A=A m +A t

[0040] A impulse =A·U impulse

[0041] 7. Finally, the virtual model is simulated to generate a virtual signal S v :

[0042] S v (t) = A m ·U(t)·e -βπ cos2πf n (t)

[0043] 8. From a practical perspective, the sound signal is mainly generated by the sound waves generated by the vibration impact of the rotating mechanical equipment combined with the on-site noise. In order to simulate the sound signal, we considered the vibration signal S v (t) The process of being converted into a sound wave signal through some medium, and the sound signal contains environmental noise, which needs to be modeled. The sound wave signal S a (t) can be expressed as the attenuation and modulation form of the vibration signal:

[0044]

[0045] in, is the sound attenuation coefficient, f a is the sound frequency, is the phase. The environmental noise N(t) can be represented by white noise or other random noise models:

[0046]

[0047] Where K is the number of noise frequency components, A k is the amplitude of the kth noise frequency component, f kis the kth noise frequency, is the phase of the kth noise frequency component;

[0048] 9. Simulated virtual sound signal S S (t) can be expressed as the superposition of the sound wave signal converted from the vibration signal and the environmental noise:

[0049] S S (t) = S a (t)+N(t)

[0050] Preferably, in step S2, multi-source sensors are installed on actual industrial equipment to collect sound and vibration signal data of the rotating mechanical system under normal operation in real time. From the perspective of industrial testing, this step is crucial and involves installing high-precision multi-source sensors at key parts of the equipment. The sensors can capture subtle changes in the equipment under normal operating conditions. The data acquisition system of the sensors needs to have high frequency and high resolution to ensure that the acquired data is detailed and accurate. By real-time monitoring and recording the operating status of the mechanical system, the data provides reliable baseline data for subsequent fault diagnosis and health monitoring, and helps identify the normal behavior patterns of the equipment under different operating conditions.

[0051] Preferably, in step S3, the virtual and real signals obtained in steps S1 and S2 are preprocessed, and their time-frequency domain characteristics are analyzed, wherein in the step of preprocessing the signal, the signal is filtered and normalized.

[0052] 1. Butterworth low-pass filter: The filter removes high-frequency noise from the signal, making the signal smoother and facilitating subsequent analysis;

[0053]

[0054] Where n represents the order of the filter, f represents the frequency component of the signal, and f c represents the cutoff frequency of the Butterworth low-pass filter, and Represent the filtered virtual signal and real signal respectively;

[0055] 2. Normalized signal: Through normalization processing, the amplitude of the signal is made within a standard range (usually between 0 and 1) to facilitate comparison and analysis between different signals;

[0056]

[0057] in, and represent the virtual signal and real signal after normalization, respectively. and Respectively represent the minimum values ​​of the virtual signal and the real signal after filtering, and Respectively represent the maximum values ​​of the virtual signal and the real signal after filtering;

[0058] Preferably, in step S4, the correlation coefficient formula evaluates the time-frequency correlation error RE between the virtual signal and the real signal f and frequency domain correlation error RE f The calculation formula can be found in the claims section.

[0059] Preferably, in step S5, a proxy model is constructed by a GRP method, and the proxy model is trained by using data generated by simulation to approximately describe the signal simulation behavior of the virtual model. The detailed steps are as follows:

[0060] 1. Select the input parameter set and the corresponding output response Y = y1, y2, ..., y n , where x i is the i-th input parameter vector, y i is the corresponding model output;

[0061] 2. Choose a suitable covariance function k(x,x'), such as the commonly used radial basis function (RBF) kernel:

[0062]

[0063] in, is the signal variance and l is the scale parameter of length.

[0064] 3. For a given input x * , the mean and covariance of its output are:

[0065]

[0066] where k*=[k(x,x1),k(x,x2),…,k(x*,x n )] T is the covariance vector between the new input and the training data, k is the covariance matrix between the training data, I is the identity matrix, is the noise variance.

[0067] Preferably, in step S6, the hyperparameters of the combination model and the dynamic model of the virtual model are iteratively optimized by minimizing the correlation error and the proxy model fitting error as the optimization objective function, combined with the firefly algorithm (FA), as shown in the pseudo code of Algorithm 1 for details;

[0068]

[0069]

[0070] The dynamic interaction between virtual and reality is also the process of iterative optimization of the hyperparameters of the virtual model and adaptation to the current working conditions. First, in order to avoid excessive redundancy of parameters that need to be optimized, the model is simplified by ignoring unimportant parameters to reduce the computational cost. Therefore, it is necessary to evaluate the sensitivity of the model output to changes in input parameters to determine which hyperparameters have the most significant impact on the model. In this step, we first define the sensitivity problem and perform Sobol sensitivity analysis to calculate the first-order sensitivity index S i and the overall sensitivity index S T .

[0071] 1. Define sensitive issues:

[0072] Y=f(θ),θ∈[θ min ,θ max ]

[0073] Where Y is the model output and θ is the input parameter vector, whose value is in the interval [θ min ,θ max ]Inside.

[0074] 2. Perform Sobol sensitivity analysis:

[0075]

[0076] Among them, S i reflects a single input parameter θ i The contribution to the variance of the model output Y is expressed as i ], Var(E[Y|θ i ]) is the variance of this expected value, Var(Y) is the total variance of Y;

[0077] Then FA is used to optimize the parameters of the geometric model and the dynamic model. In this step, the number of fireflies n and the maximum number of iterations max are set. Iter , light absorption coefficient γ, attraction β0, step size factor α, initialize the firefly position θ i ∈[θ min ,θ max ] and its intensity I i :

[0078] I i =f(θ i )

[0079] Among them, f(θ i ) represents the objective function or fitness function of the optimization process. The greater its intensity, the more superior the firefly position. This objective function can be defined as:

[0080] f(θ i )=ω1RE f +ω2MSE-ω3R 2

[0081] Among them, ω1, ω2, and ω3 are weighted coefficients of correlation error, mean square error, and determination coefficient, respectively, which are used to balance the influence of correlation error, mean square error, and determination coefficient. In the iterative optimization of the FA main loop, for each firefly i, the distance r between it and all other fireflies j is calculated. ij :

[0082] r ij =|θ i -θ j |

[0083] If I j >I i , then firefly i is attracted by firefly j and moves according to the following update formula:

[0084]

[0085] Among them, β0 is the base value of attraction, γ is the light absorption coefficient, which controls the speed at which attraction decays with distance, α is the random step factor, and rand is a random number between [0,1] used to simulate random motion and update the position θ of firefly i. i , making sure it is within range until the light intensity of the firefly I i Until the conditions are met.

[0086] Preferably, in step S7, the parameters obtained by optimizing the firefly algorithm (FA), the updated virtual model is regarded as a physical model capable of simulating high-fidelity fault signals, the simulation signal generated using the optimized parameters is highly similar to the real fault signal, and the high-fidelity fault signal generated by the simulation and the fault-free signal of normal operation of the equipment in the industrial scenario are combined, and the characteristic parameters between the two, such as spectral characteristics, time domain characteristics, etc., are compared using feature matching and pattern recognition methods, so as to effectively determine whether there is a fault in the bearing in the rotating machinery. Through this process, not only the accuracy and reliability of fault diagnosis are improved, but also timely and accurate reference basis can be provided for equipment maintenance.

[0087] Beneficial effects:

[0088] 1. By building a high-fidelity virtual physical model of the bearing, combined with real-time multi-channel sound and vibration sensor data, signal preprocessing, feature extraction and time-frequency domain analysis are performed. By establishing a proxy model to replace the complex virtual signal simulation model, dynamic interaction with real scene signals is efficiently achieved. The firefly algorithm (FA) is used to iteratively optimize the parameters of the virtual model to minimize the correlation error between the virtual signal and the real signal and the fitting accuracy of the proxy model, and finally achieve high-fidelity fault signal simulation and fault diagnosis;

[0089] 2. By combining high-fidelity virtual physical model simulation signals and big data sensor signals, high-precision prediction and diagnosis of equipment is achieved under zero-failure sample data;

[0090] 3. Based on the hybrid technology of bearing virtual physical model construction and big data monitoring, a rolling bearing diagnosis method with zero fault samples is designed to be applied to products such as digital health management, intelligent operation and maintenance, and decision-making of rolling bearing systems in real factories. The digital diagnosis method first adopts nonlinear dynamic bearing virtual signal simulation modeling, and uses multi-channel, symmetrically arranged sound and advanced sensors to collect real signal data in industrial scenarios in real time. The time-frequency domain feature analysis is performed after preprocessing the virtual and real signals to evaluate their correlation errors. In order to replace the complex virtual signal simulation model and realize effective dynamic interaction with the signals in the real scenario, an agent model is established to approximately describe the signal simulation behavior of the original virtual model, and high-fidelity simulation behavior is achieved by evaluating the fitting accuracy of the regression prediction. Taking minimizing the correlation error between the time-frequency domain characteristics of the virtual signal and the real-scene signal and the fitting accuracy of the best proxy model as the optimization objective function, the FA (firefly algorithm) is combined to iteratively optimize the hyperparameters of the virtual model, realizing the real-time optimization of the geometric parameters and dynamic parameters of the virtual model. Finally, the invented dynamic interaction method enables the virtual signal simulation model to fully explore the physical characteristics of the signal that operates normally and without faults in the real scene, so as to obtain high-fidelity virtual signals of inner ring, outer ring, roller and cage faults.

[0091] 4. When facing the bearing system in large-scale rotating mechanical equipment, its potential faults are great. In order to avoid the serious consequences caused by failures, the bearings generally used are of high-precision standards. Therefore, it is not easy to collect system signal data under bearing failure operation in the factory. This unbalanced data sample distribution directly leads to the limitations of fault diagnosis in real scenarios based solely on big data-driven pattern classification methods. Therefore, this method has good scene condition adaptability and can be applied to bearing system fault diagnosis in different industrial scenarios. It fully mines limited industrial data information and rich physical prior knowledge, and ensures the safe and reliable operation of important rotating mechanical systems through digital twin technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 This is a flow chart of the virtual signal regression prediction method based on the Gaussian regression GPR agent model.

[0093] Figure 2 It is the process of optimizing the parameters of geometric and dynamic models based on the Firefly Algorithm (FA). DETAILED DESCRIPTION

[0094] The present invention will be described below with reference to examples; however, the present invention is not limited to the following examples.

[0095] In addition, in the following description, unless otherwise specified, % refers to m / m mass percentage. All reagents, raw materials and instruments used in the present invention are well known in the art and can be purchased from the market, but do not limit the implementation of the present invention. Other reagents and equipment well known in the art can be applied to the implementation of the following embodiments of the present invention.

[0096] Embodiment 1:

[0097] This embodiment provides a digital twin bearing fault diagnosis method based on virtual and real dynamic interaction, and the specific steps are as follows:

[0098] S1: Construct the simulation model V(θ) of the initial virtual physical signal, initialize the parameters θ of the virtual model, and simulate and generate the initial virtual signal;

[0099] S2: Through multi-channel, symmetrically arranged sound and vibration sensors, multi-source, multi-modal sound and vibration signals in real scenes are collected in real time;

[0100] S3: Preprocessing the initial virtual signal in step S1 and the real scene signal in step S2 simultaneously and extracting time-frequency domain features;

[0101] S4: Calculate the correlation error RE between the time-frequency features of the virtual and real signals extracted in step S3 t Correlation error RE of frequency domain characteristics f , and set the thresholds of the two correlation errors;

[0102] S5: In order to make the dynamic interaction between virtual and real signals more efficient, the GPR model based on reduced-order agent is used to approximate regression fitting of the constructed complex virtual model, so as to reduce the computational burden in the FA optimization process in the subsequent S6 step;

[0103] S6: In order to avoid excessive redundancy of parameters that need to be optimized in the virtual model during the interaction process, Sobol sensitivity analysis is used and the first-order sensitivity index S is calculated. i and the overall sensitivity index S TThe most significant parameters are analyzed and these important parameters are selected as the parameters θ that need to be optimized in the virtual model;

[0104] S7: By setting the objective function f(θ i )=ω1RE f +ω2MSE-ω3R 2 , FA is used to iteratively optimize the parameters θ that need to be optimized, and finally the optimized parameters θ are used * Update virtual signal S v (t);

[0105] S8: In step S2, sufficient normal operating fault-free sample data is obtained, combined with step S7 to obtain high-fidelity simulated virtual signals of inner ring, outer ring, roller and cage faults. The digitally enabled twin data set can be used to solve the lack of fault sample data in real scenarios, and different types of operating faults can be accurately predicted by constructing feature matching and pattern recognition methods of fault signals.

[0106] Embodiment 2:

[0107] For step S5, combine Figure 1 The virtual signal regression prediction method based on the Gaussian process regression (GPR) agent model can be divided into the following steps:

[0108] First, import the input data X = {x1, x2, ..., x n} and the corresponding output data Y={y1,y2,…,y n} training data set, where data X is the signal characteristic factors collected from the real system or experiment, such as the mean and variance of the actual signal, which affects various parameters of the virtual signal and is used to train the independent variable of the GPR model, and data Y is the corresponding virtual signal or the target characteristic factors predicted by the model, such as the mean and variance of the virtual signal, which is used to train the dependent variable of the GPR model;

[0109] Next, we construct a Gaussian process regression model f(x) and calculate the fitness accuracy of the regression. We select a suitable kernel function k(x,x′) as the covariance function to measure the similarity between data points. For a given input x*, we calculate the output mean μ(x) after Gaussian regression. * ) and covariance σ(x * ), use the real input data X to train the proxy regression model to obtain the fitted proxy model;

[0110] After completing the model training, the proxy model is used to perform regression prediction on the new input data to generate virtual signal data. In order to evaluate the model performance, the regression error is calculated using MSE to determine the accuracy of the prediction. At the same time, the determination coefficient R is used2 Calculate the fitting accuracy of the model and further evaluate the effect of the model. The detailed pseudo code involved in this method is described in Algorithm 2:

[0111]

[0112]

[0113] Embodiment three:

[0114] In step S6, the virtual model parameters that need to be optimized are obtained through Sobol sensitivity analysis. These parameters can be divided into geometric model and dynamic model parameters, which characterize the physical characteristics of the simulated virtual signal.

[0115] Embodiment 4:

[0116] Step S7: Combine Figure 2 FA is an intelligent optimization algorithm based on the luminous behavior of fireflies in nature. It is often used to solve nonlinear, multi-peak and high-dimensional optimization problems. When optimizing the parameters of the virtual model, the process of FA is as follows:

[0117] First, FA needs to define an objective function f(θ), whose parameter θ needs to be optimized. Next, select a certain number of fireflies n and set the maximum number of iterations max iter , random step coefficient α, base value of attraction coefficient β0 and absorption coefficient γ, as well as upper and lower bounds L and U of the parameters, then randomly initialize the position θ of each firefly within the bounds of the parameters i , and calculate its initial fitness value f i =f(θ i ).

[0118] In each iteration, each firefly i is compared with other fireflies j in turn, such as j's fitness f j Better than i's adaptation f i , then calculate the distance r between the two ij =|θ i -θ j |, based on this distance, calculate the attraction β ij Then, according to the attraction and random step size δ, the position θ of firefly i is updated i =θ i +β ij (θ j -θ i )+α(rand()-0.5), ensure that the updated position is within the boundary [L,U], and re-evaluate the fitness value f i；

[0119] After each iteration, record the best parameter θ in the current iteration* and its fitness value f * 'After many iterations, the algorithm returns the global optimal parameter θ * , to optimize the virtual model parameters. FA can effectively jump out of the local optimum and find the global optimal solution by simulating the mutual attraction and random search between fireflies. The detailed pseudo code involved in this method is described in Algorithm 3:

[0120]

[0121]

[0122] Embodiment five:

[0123] In step S8, high-fidelity fault signals are generated through online simulation and real-time acquisition of fault-free signals of normal operation of equipment in industrial scenarios, so as to realize dynamic interaction and real-time twinning and realize online fault diagnosis. First, the sensor signals of normal operation of rotating machinery system are collected. Combined with the dynamic interaction method, the simulation software can synchronously generate high-fidelity virtual signals under various bearing fault conditions online. Then, the time-frequency domain features of these signals are extracted and matched and pattern recognized. A big data driven model containing dynamic twin attributes can be trained. By deploying this trained data-driven model, combined with real-time monitoring of rotating machinery system signals under normal operation, high-precision real-time fault detection and diagnosis can be realized through dynamic interaction between virtual signals and real signals, ensuring the normal operation and preventive maintenance of equipment.

[0124] Technical personnel should note that although the present invention has been described according to the above specific implementation methods, the inventive concept of the present invention is not limited to this invention, and any modification using the inventive concept will be included in the scope of protection of this patent.

Claims

1. A digital twin bearing fault diagnosis method based on dynamic interaction between virtual and real world, characterized in that: The digital twin bearing fault diagnosis method based on virtual and real dynamic interaction includes the following steps: S1. Construct the initial virtual physical model of the bearing and simulate and generate digital virtual signals; S2, using multi-channel symmetrically arranged sound and vibration sensors to collect real signal data of rotating mechanical systems in industrial scenarios in real time; S3, performing time-frequency domain feature analysis on the virtual signal and the real signal after preprocessing; S4, evaluating the correlation coefficient and error between the virtual signal and the real signal; S5. Establish a proxy model to approximately describe the signal simulation behavior of the virtual model with lower complexity and computational cost; S6. Use mean square error and determination coefficient to evaluate the regression fitting fitness accuracy of the surrogate model; S7, designing the optimization objective function to minimize the correlation error between the time-frequency domain characteristics of the virtual signal and the real scene signal and the fitting accuracy of the best proxy model; S8, combining the firefly algorithm (FA) to iteratively optimize the hyperparameters of the virtual model to achieve dynamic interaction between the virtual model simulation signal and the real signal; S9. Finally, the optimized virtual model is used to simulate and generate high-fidelity fault signals. Through feature matching and pattern recognition methods, it is determined whether the bearing is faulty. This can be applied to digital health management, intelligent operation and maintenance, and decision support of rolling bearing systems in actual factories.

2. According to the digital twin bearing fault diagnosis method based on virtual and real dynamic interaction according to claim 1, wherein the step S1 constructs a virtual physical model of the bearing and simulates and generates a digital virtual signal, and the constructed virtual physical model of the bearing V(θ) can be regarded as a geometric model G(θ g ) and the kinetic model D(θ d ) combination: V(θ)=G(θ g )+D(θ d ) in, Geometric model G(θ g ) describes the geometric parameters of rolling bearings, including rolling element diameter, pitch circle diameter, contact angle, number of rolling elements, etc. These parameters are used to calculate the characteristic frequencies of the bearing during operation, such as outer ring fault frequency, inner ring fault frequency, and rolling element fault frequency. The geometric model parameters can be statistically calculated as follows: Kinetic model D(θ d ) generates simulated fault signals based on these structural parameters and other dynamic parameters. By adjusting the parameters of the geometric model and the dynamic model, simulated signals of different fault types and severity can be generated. The dynamic model parameters can be statistically calculated as follows:

3. According to the digital twin bearing fault diagnosis method based on dynamic interaction between virtual and reality according to claim 1, multi-source and multi-modal sensing technology is used to realize real-time acquisition of all and vibration signals. In this step, it is necessary to first reasonably select and install sensors. The installation position and layout of sensors directly affect the accuracy and representativeness of the data. Multi-channel symmetrically arranged sensors can capture the comprehensive vibration characteristics of the equipment during operation and reduce the data deviation caused by a single measuring point. Secondly, the sensors should cover the key parts of the equipment, which are where vibration and faults are most obvious and can provide the most valuable vibration signals. When collecting data, attention should be paid to the sampling rate and collection period to ensure the comprehensiveness of the data and the capture of high-frequency characteristics. Finally, data storage and transmission are the basis of data processing, and the integrity and timeliness of the data must be ensured. Through these detailed operations, high-quality real signal data can be obtained, which provides a reliable data basis for the subsequent dynamic interaction and fault diagnosis of virtual and real signals.

4. According to claim 1, the digital twin bearing fault diagnosis method based on dynamic interaction between virtual and reality: wherein the virtual signal and the real signal are preprocessed and then subjected to time-frequency domain feature analysis.

5. The digital twin bearing fault diagnosis method based on virtual and real dynamic interaction according to any one of claims 1 or 3: by evaluating the correlation error between the virtual signal and the real signal to bridge the interaction between the physical sensing and the generated signal, the correlation error RE in the time domain t Can be defined as: Where ρ is the virtual signal S v (t) and the real signal S r (t), E is the expectation operator, and are the means of the virtual signal and the real signal, respectively. and are the standard deviations of the virtual signal and the real signal, respectively. It can be seen that when ρ is closer to 1, the error is smaller and the signals are more similar. When ρ is closer to 0, the error is larger and the signals are less similar.

6. According to claim 1, the digital twin bearing fault diagnosis method based on dynamic interaction between virtual and reality: wherein in step S5, a proxy model is established to approximately describe the signal simulation behavior of the virtual model, and the proxy model based on RSP (reduced-order proxy mode) specifically adopts the GPR (Gaussian process regression) method.

7. The digital twin bearing fault diagnosis method based on virtual and real dynamic interaction according to claim 1: Step S6 uses the mean square error MSE and the determination coefficient R 2 To evaluate the regression fitting fitness accuracy of the surrogate model: in, y i is the real output, is the predicted output of the surrogate model, m is the number of samples of the test data, is the mean of the true output. MSE is used to evaluate the average error of the surrogate model on the test data. The smaller the value, the higher the model fitting accuracy. 2 It is used to evaluate the fit of the model. The closer its value is to 1, the better the model fits the data. A value of 1 indicates that the model fits the data perfectly.

8. According to claim 1, the digital twin bearing fault diagnosis method based on virtual and real dynamic interaction: wherein the optimization objective function is to minimize the correlation error between the time-frequency domain characteristics of the virtual signal and the real scene signal and the fitting accuracy of the best proxy model, the purpose is to dynamically self-correct the virtual signal through the signal in the real scene, so that the physical characteristics of the virtual signal more faithfully reflect the real response; The correlation error RE in the time domain is introduced above t , the following introduces the correlation error RE on the frequency domain characteristics f , first, perform FFT (Fast Fourier Transform) calculation on the virtual signal and the real signal: And calculate its relative error RE in the frequency domain f :

9. The digital twin bearing fault diagnosis method based on virtual and real dynamic interaction according to claim 1: the hyperparameters of the virtual model are iteratively optimized in combination with FA (firefly algorithm) to achieve continuous calibration and optimization of the virtual model simulation signal, such as RE # >0.95, the geometric parameters and dynamic parameters of the virtual model are adjusted θ←θ+Δθ, otherwise, the optimization iteration ends and a high-fidelity virtual signal simulation model is obtained.

10. According to claim 1, the digital twin bearing fault diagnosis method based on dynamic interaction between virtual and reality: the use of the optimized virtual model simulation to generate high-fidelity fault signals, combined with the existing pattern classification methods of feature engineering, can effectively determine whether there are operational faults in the bearing system in the industrial scenario.

Images