Method and system for constructing dynamic interactive twin model of rolling bearings throughout their life cycle

Abstract

The present invention discloses a method and system for constructing a dynamic interactive twin model of a rolling bearing throughout its life cycle, comprising the following steps: signal acquisition, time and frequency domain feature extraction and analysis, digital twin modeling and analysis, parameter optimization mechanism, and generation of twin data. The present invention constructs a dynamic digital twin model with a model parameter update mechanism based on the parameters and operating conditions of the actual bearing. The model can interact with the bearing measurement signal in real time to update its parameters, as well as simulate the evolution of roller relative slip and bearing surface morphology throughout its life cycle, revealing the mapping mechanism between fault evolution and vibration response. Through the parameter update mechanism, the model realizes real-time interaction with the actual bearing, simulates the evolution of surface morphology in different degradation stages, and can generate sufficient life cycle twin data to effectively compensate for the problem of insufficient actual data.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a method and system for constructing a dynamic interactive twin model of a rolling bearing throughout its life cycle. Background Art

[0002] Rolling bearings are precision mechanical components that convert sliding friction between a rotating shaft and its seat into rolling friction, thereby reducing friction losses and ensuring the smooth operation and long-term stability of mechanical equipment. However, rolling bearings operate under complex and variable operating conditions, making it difficult to obtain bearing vibration signals under certain operating conditions. This results in a shortage of valid data samples, and the lack of fault data samples under these conditions in bearing fault databases makes it difficult to supplement and improve bearing fault categories, making it impossible to build effective fault diagnosis models.

[0003] Therefore, a dynamic digital twin model with a model parameter update mechanism was constructed based on the parameters and operating conditions of the actual bearing. It interacts with the bearing measurement signals in real time to update its parameters, simulates the evolution of roller relative slip and bearing surface morphology throughout its life cycle, and reveals the mapping mechanism between fault evolution and vibration response.

[0004] Real-time monitoring and fault diagnosis of rolling bearings have long been a key area of ​​research. Effective condition detection and diagnosis, addressing the lack of labeled sample data under different operating conditions and the lack of sample data in certain specific conditions, is crucial for reducing economic costs, maximizing bearing life, improving production efficiency, and preventing safety incidents.

[0005] The core of digital twin technology is the high-precision simulation of the physical world. By complementing it with big data analytics, it digitizes actual projects, equipment, or systems, enabling real-time interaction with the real world. Through unified data access, real-world equipment data is assigned to the model and continuously updated, mapping the state of the physical entity in real time and reflecting its entire lifecycle. Optimized twin data enables comprehensive, multi-dimensional digital management and simulation analysis of physical objects, providing accurate data and intelligent decision-making support, thereby optimizing production efficiency, improving product quality, and reducing costs.

[0006] Currently, direct monitoring data from the operating status of industrial equipment is commonly used. However, potential problems often arise: During bearing operation, effective vibration signals cannot be collected under certain operating conditions, resulting in missing fault data samples under these conditions in the bearing fault database, making it difficult to supplement and improve bearing fault categories; data affected by noise or other external factors makes analysis and prediction difficult; and data collected by sensors contains bearing fault and degradation information, requiring effective methods to extract more feature quantities from a multidimensional perspective to represent more feature information. This requires a deep understanding of the equipment status, increasing the difficulty of subsequently identifying the bearing operating status. Summary of the Invention

[0007] In response to the shortcomings of the existing technology, the present invention provides a method and system for constructing a dynamic interactive twin model of a rolling bearing throughout its life cycle, which is used to effectively compensate for the problem of insufficient actual data and to more accurately establish a bearing diagnostic model.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0009] A method for constructing a dynamic interactive twin model of a rolling bearing throughout its life cycle comprises the following steps:

[0010] S1. Signal acquisition: Acquire the real-time operating parameters of the rolling bearing and obtain the vibration acceleration data of the rolling bearing throughout its life cycle;

[0011] S2. Time and frequency domain feature extraction and analysis: Extract the time domain, frequency domain, and time-frequency domain characteristic parameters of the rolling bearing, calculate the characteristic values, connect the characteristic values ​​of the data sets of the entire life cycle into a characteristic curve in chronological order, set the alarm threshold analysis in combination with the normal distribution, use the envelope entropy of the inherent mode function as the evaluation standard of the variational mode decomposition effect, use the genetic algorithm to optimize the penalty factor α, and adaptively select the optimal parameter. After the signal is decomposed, the decomposition level K is obtained according to the spectral correlation coefficient method, and the time and frequency domain features are extracted;

[0012] S3, Digital Twin Modeling and Analysis: Draw a three-dimensional model of the rolling bearing and implement digital twin modeling of the rolling bearing using finite element analysis software. Using the characteristic parameters extracted in S2 as input, generate n groups of dynamic parameter samples uniformly distributed within the parameter variation range. Perform parametric sweeps and modal analysis on each group, calculate the first four natural frequencies corresponding to the n groups, merge the input and output parameters, and generate the input-output data set of the deep neural network.

[0013] S4. Parameter optimization mechanism: Determine the optimization variables and use the value range of each parameter as a constraint condition. Combine the natural frequency prediction value of the twin model and the experimental value of the physical entity as the target value to construct the objective function. Establish a dynamic parameter optimization model for virtual-real mapping. Use the particle swarm optimization algorithm, use the relevant parameters of the virtual model as the optimization variables, and use the minimum relative error between the theoretical natural frequency calculated by the finite element model and the corresponding experimental value as the objective function to optimize and determine the relevant parameters of the twin physical model.

[0014] S5. Twin data: Input the initialized geometric parameters into the full life cycle dynamic twin model to generate highly applicable full life cycle vibration acceleration twin data.

[0015] Preferably, in S2, in order to study the running state of the rolling bearing, important features of the time domain, frequency domain and time-frequency domain are extracted. Considering the collection of the vibration acceleration signal of the rolling bearing, in the process of extracting the time domain features, the characteristic curve is used in combination with the normal distribution for statistical analysis. It is assumed that a group of bearings collects N groups of experimental data from normal to failure, and the length of each group of experimental data is M. The characteristic index of each group of experimental data is calculated respectively, and the N groups of characteristic indexes are combined into a vector L. The mean μ and variance σ are calculated. The statistic U is known to be:

[0016]

[0017] The statistic U can be used as the offset of the amplitude in the degradation stage. According to Chebyshev's inequality, in any data set, the probability that a data point is n standard deviations away from the mean of the current data point is very small, that is:

[0018]

[0019] Based on the characteristics of the normal distribution, approximately 90% of the eigenvalues ​​are selected to be within the range of ±c standard deviations, avoiding the problem of subjective threshold selection. The first 10% of normal samples of the bearing are taken, and the value of 3σ+u is calculated as the threshold line. If n consecutive eigenvalues ​​exceed the selected fault warning value, it is considered that the bearing degradation stage has changed. The value of n is set by the characterization ability of the actual bearing degradation process.

[0020] The characteristic values ​​of each set of experimental data are calculated, and the characteristic values ​​of the whole life data sets are connected in chronological order to form a vibration characteristic curve. The important time domain characteristic parameters that can characterize the bearing degradation characteristics are analyzed and extracted.

[0021] Preferably, in said S2, a single time domain analysis cannot determine the fault type of the rolling bearing, and important frequency domain features are extracted from the full life vibration trend curve for analysis.

[0022] Preferably, in S2, it is difficult to obtain sufficient information by simply extracting time and frequency domain features. Time and frequency domain analysis is an effective means of processing non-stationary fault signals. Since the bearing life cycle data is collected in a strong background noise environment, in order to reduce the interference of noise and restore the original characteristics of the signal, a process based on genetic algorithm (GA) to optimize the parameter α is selected and applied to the VMD algorithm to improve the accuracy and stability of signal decomposition;

[0023] Step 1: Set the parameters of the GA algorithm. The penalty factor α ranges from [100, 5000]. Based on the algorithm accuracy and speed, the population size is set to 20. The variables are encoded in 20-bit binary format. The number of iterations is set to 20. The optional parameter for copying the generation gap population is set to 0.95. Initialize the crossover and mutation probabilities and assign them to each population. The parameter range and initial value can be dynamically adjusted according to the specific application scenario and signal characteristics.

[0024] Step 2: Initialize the parameter α, loop through the individuals in the population, perform VMD decomposition on the signal, and calculate the envelope entropy of each IMF as the fitness function;

[0025] Step 3: Genetic algorithm loop, starting from zero and looping genetic algebra, optimizing individuals through selection, crossover and mutation operations, selecting to retain or eliminate individuals according to the value of the fitness function, and taking the value with the minimum envelope entropy as the result of the fitness function F(x), that is:

[0026] F(x)=min{E p1 , E p2 ,…,E pi}(i=1,...,K)

[0027] Step 4: Compare the best individuals of the source population and the target population. If they are the same, the best individual of the source population is retained. Otherwise, it is replaced by the best individual of the target population. The selected individual is placed in the new population to obtain the optimal fitness value of this iteration.

[0028] Step 5: Loop through steps 2 to 4 until the maximum number of iterations is reached, the optimization stops, and the global optimal individual penalty factor α is obtained;

[0029] Step 6: Substitute the parameter α optimized by the GA algorithm into the VMD algorithm to calculate the spectral correlation coefficient between the IMF component and the original signal;

[0030] Step 7: When the minimum spectral correlation coefficient is less than the set threshold, K = K-1; otherwise, K = K+1, and continue to perform steps 4 to 6 until the maximum number of decomposition levels is reached, and then stop decomposition;

[0031] The improved VMD method can more accurately capture the time-varying characteristics of the signal and better describe the changing laws of the signal; optimizing the parameter α through the genetic algorithm can effectively avoid the offset distortion of the center frequency of the IMF component, thereby improving the accuracy and stability of the signal decomposition; using the spectral correlation coefficient method to determine the number of decomposition layers K can avoid the influence of under-decomposition and over-decomposition on the robustness of the signal, thereby obtaining more accurate decomposition results; the extracted time-frequency domain features are more sensitive to early bearing faults than the time domain and frequency domain features.

[0032] Preferably, in said S3, a three-dimensional model of the rolling bearing in a normal state is drawn using software, and the physical parameters of the rolling bearing required to establish the model include bearing pitch diameter, ball diameter, number of balls, and contact angle;

[0033] Import the 3D model into the finite element analysis software, perform material settings, contact mode settings, mesh generation, and apply loads and constraints to establish a twin finite element model of the rolling bearing. Use the optimized Latin hypercube experimental design method to generate n groups of randomly ordered dynamic parameter sample sequences that are evenly distributed within the parameter variation range. Using the parametric sweep function of the finite element analysis software, extract each group and substitute them into the finite element model for modal analysis, solving for the first four natural frequencies corresponding to the n groups under different input parameters.

[0034] The input and output parameters are merged to generate an input-output dataset (i.e., twin dataset) for training the deep neural network. A deep neural network model is constructed, and the deep neural network is trained using the twin dataset. m groups of test sample datasets of non-training sample points are randomly generated to test the accuracy of the trained deep network model. The global mean absolute error and global mean absolute percentage error of the test set are used as evaluation indicators to evaluate the performance of deep neural network models at different layers. Finally, the deep neural network model with the best performance is selected as the twin finite element model of the rolling bearing to achieve digital mirroring of the solid finite element model.

[0035] Preferably, in S4, the parameter updating method takes into account the signal characteristics and similarities of multiple signal sources such as the time domain and the frequency domain. In order to obtain the best parameters and ensure a good match between the simulated data and the actual data, the parameters of the dynamic twin model are updated using an optimization algorithm. During the calibration process, multiple parameters need to be determined and updated, and these parameters need to match the physical measurements in many aspects. The mathematical expression of the objective function in the model updating process is:

[0036]

[0037] is the updated parameter value after model optimization, TX 孪生 ,TY 真实Represent the time domain waveforms of the twin signal and the actual signal, TX 孪生 ,TY 真实 Represent the frequency domain waveforms of the twin signal and the actual signal after Fourier transform, W 孪生 , W 真实 Represent the time-frequency domain waveforms of the twin signal and the actual signal respectively, correlation represents the Pearson correlation coefficient, They are the peak values ​​of the two signals, respectively. With the goal of minimizing the three functions, the parameters of the twin model are optimized to achieve an accurate match with the actual data;

[0038] The multi-objective particle swarm optimization (MOPSO) is integrated into the model to dynamically update the parameters of the twin model according to the actual signal, thereby showing a powerful exploration capability and achieving rapid convergence. Its expression is as follows:

[0039] The update formula for particle velocity is:

[0040]

[0041] represent the velocity vector and position vector of particle i in the kth iteration, respectively. is the historical optimal position of particle i in the kth iteration, is the historical optimal position of the population, r1 and r2 are random numbers in the range [0,1], which can increase the randomness of the search, c1 and c2 are learning factors, and ω is the inertia weight. As the number of iterations increases, the ω inertia weight continues to decrease, making PSO have stronger global convergence ability in the early stage and stronger local convergence ability in the later stage;

[0042]

[0043] ω max ,ω min Indicates the maximum and minimum values ​​of the inertia weight, while iter and iter max Indicates the current number of iterations and the maximum number of iterations;

[0044] The particle position update formula is:

[0045]

[0046] The MOPSO multi-objective optimization algorithm is used to optimize the parameter group in the existing multi-objective optimization algorithm based on Pareto advantage;

[0047] Pareto dominance is introduced into particle swarm optimization to solve multi-objective optimization problems. MOPSO aims to find a set of Pareto optimal solutions, where no solution dominates another solution among multiple objectives. The following operations are performed to determine the optimal value from multiple solutions: 1) Pareto optimal solutions are calculated and stored in an archive; 2) a roulette wheel selection method is used to determine the most advantageous solution, which effectively balances all objectives.

[0048] At the same time, the present invention also provides a system or electronic device for constructing a dynamic interactive twin model of a rolling bearing throughout its entire life cycle, which is characterized by being used to implement the above-mentioned method for constructing a dynamic interactive twin model of a rolling bearing throughout its entire life cycle, thereby achieving automatic modeling and more accurate modeling.

[0049] At the same time, the present invention also provides a storage medium on which a computer program is stored, which, when executed, is used to implement the above-mentioned dynamic interactive twin model method of rolling bearings throughout their life cycle.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] Based on the parameters and operating conditions of actual bearings, the present invention constructs a dynamic digital twin model with a model parameter update mechanism. Unlike traditional dynamic models, it can interact with bearing measurement signals in real time to update its parameters, as well as simulate the evolution of roller relative slip and bearing surface morphology throughout its life cycle. It reveals the mapping mechanism between fault evolution and vibration response, realizes real-time interaction with actual bearings through the parameter update mechanism, simulates the evolution of surface morphology in different degradation stages, and can generate sufficient life cycle twin data to effectively compensate for the problem of insufficient actual data. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flow chart of the method for constructing the present invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] like Figure 1 As shown, the present invention provides a technical solution: a method for constructing a dynamic interactive twin model of a rolling bearing throughout its life cycle, comprising the following steps:

[0055] S1. Signal acquisition: Acquire the real-time operating parameters of the rolling bearing and obtain the vibration acceleration data of the rolling bearing throughout its life cycle;

[0056] S2. Time and frequency domain feature extraction and analysis: Extract the time domain, frequency domain, and time-frequency domain characteristic parameters of the rolling bearing, calculate the characteristic values, connect the characteristic values ​​of the data sets of the entire life cycle into a characteristic curve in chronological order, set the alarm threshold analysis in combination with the normal distribution, use the envelope entropy of the inherent mode function (IMF) as the evaluation standard of the variational mode decomposition (VMD) effect, use the genetic algorithm (GA) to optimize the penalty factor α, and adaptively select the optimal parameter. After the signal is decomposed, the decomposition level K is obtained according to the spectral correlation coefficient method, and the time and frequency domain features are extracted;

[0057] S3. Digital twin modeling and analysis: Draw a three-dimensional model of the rolling bearing and implement digital twin modeling of the rolling bearing using finite element analysis software. Using the characteristic parameters extracted in S2 as input, generate n groups of dynamic parameter samples uniformly distributed within the parameter variation range. Perform parametric sweeps and modal analysis on each group, calculate the first four natural frequencies corresponding to the n groups, merge the input and output parameters (the first four natural frequencies), and generate the input-output dataset (twin dataset) of the deep neural network.

[0058] S4. Parameter optimization mechanism: Determine the optimization variables and use the value range of each parameter as a constraint condition. Combine the natural frequency prediction value of the twin model and the experimental value of the physical entity as the target value to construct the objective function. Establish a dynamic parameter optimization model for virtual-real mapping. Use the particle swarm optimization algorithm, use the relevant parameters of the virtual model as the optimization variables, and use the minimum relative error between the theoretical natural frequency calculated by the finite element model and the corresponding experimental value as the objective function to optimize and determine the relevant parameters of the twin physical model.

[0059] S5. Twin data: Input the initialized geometric parameters into the full life cycle dynamic twin model to generate highly applicable full life cycle vibration acceleration twin data.

[0060] In S2, in order to study the running state of rolling bearings, important features of time domain, frequency domain and time-frequency domain are extracted. Considering the collection of vibration acceleration signals of rolling bearings, in the process of extracting time domain features, characteristic curves are combined with normal distribution for statistical analysis. Assuming that a group of bearings collect N groups of experimental data from normal to failure, the length of each group of experimental data is M. The characteristic index of each group of experimental data is calculated respectively, and the N groups of characteristic indexes are combined into a vector L. The mean μ and variance σ are calculated. According to the normal distribution characteristics, the statistic U is:

[0061]

[0062] The statistic U can be used as the offset of the amplitude in the degradation stage. According to Chebyshev's inequality, in any data set, the probability that a data point is n standard deviations away from the mean of the current data point is very small, that is:

[0063]

[0064] Based on the characteristics of the normal distribution, approximately 90% of the eigenvalues ​​are selected to be within the range of ±c standard deviations, avoiding the problem of subjective threshold selection. The first 10% of normal samples of the bearing are taken, and the value of 3σ+u is calculated as the threshold line. If n consecutive eigenvalues ​​exceed the selected fault warning value, it is considered that the bearing degradation stage has changed. The value of n is set by the characterization ability of the actual bearing degradation process.

[0065] The characteristic values ​​of each set of experimental data are calculated, and the characteristic values ​​of the whole life data sets are connected in chronological order to form a vibration characteristic curve. The important time domain characteristic parameters that can characterize the bearing degradation characteristics are analyzed and extracted.

[0066] In S2, single time domain analysis cannot determine the fault type of rolling bearings, and important frequency domain features are extracted from the full life vibration trend curve for analysis.

[0067] In S2, simply extracting time and frequency domain features is insufficient to obtain sufficient information. Time and frequency domain analysis is an effective method for processing non-stationary fault signals. Bearing life cycle data is collected in a strong background noise environment. To reduce noise interference and restore the original signal characteristics, a genetic algorithm (GA)-based optimization process for parameter α is selected and applied to the VMD algorithm to improve the accuracy and stability of signal decomposition.

[0068] Step 1: Set the parameters of the GA algorithm. The penalty factor α ranges from [100, 5000]. Based on the algorithm accuracy and speed, the population size is set to 20. The variables are encoded in 20-bit binary format. The number of iterations is set to 20. The optional parameter for copying the generation gap population is set to 0.95. Initialize the crossover and mutation probabilities and assign them to each population. The parameter range and initial value can be dynamically adjusted according to the specific application scenario and signal characteristics.

[0069] Step 2: Initialize the parameter α, loop through the individuals in the population, perform VMD decomposition on the signal, and calculate the envelope entropy of each IMF as the fitness function;

[0070] Step 3: Genetic algorithm loop, starting from zero and looping genetic algebra, optimizing individuals through selection, crossover and mutation operations, selecting to retain or eliminate individuals according to the value of the fitness function, and taking the value with the minimum envelope entropy as the result of the fitness function F(x), that is:

[0071] F(x)=min{Ep1 , E p2 ..., E pi}(i=1,...,K)

[0072] Step 4: Compare the best individuals of the source population and the target population. If they are the same, the best individual of the source population is retained. Otherwise, it is replaced by the best individual of the target population. The selected individual is placed in the new population to obtain the optimal fitness value of this iteration.

[0073] Step 5: Loop through steps 2 to 4 until the maximum number of iterations is reached, the optimization stops, and the global optimal individual penalty factor α is obtained;

[0074] Step 6: Substitute the parameter α optimized by the GA algorithm into the VMD algorithm to calculate the spectral correlation coefficient between the IMF component and the original signal;

[0075] Step 7: When the minimum spectral correlation coefficient is less than the set threshold, K = K-1; otherwise, K = K+1, and continue to perform steps 4 to 6 until the maximum number of decomposition levels is reached, and then stop decomposition;

[0076] The improved VMD method can more accurately capture the time-varying characteristics of the signal and better describe the changing laws of the signal; optimizing the parameter α through the genetic algorithm can effectively avoid the offset distortion of the center frequency of the IMF component, thereby improving the accuracy and stability of the signal decomposition; using the spectral correlation coefficient method to determine the number of decomposition layers K can avoid the influence of under-decomposition and over-decomposition on the robustness of the signal, thereby obtaining more accurate decomposition results; the extracted time-frequency domain features are more sensitive to early bearing faults than the time domain and frequency domain features.

[0077] In S3, software is used to draw a three-dimensional model of the rolling bearing in a normal state, and the physical parameters of the rolling bearing required to establish the model include bearing pitch diameter, ball diameter, number of balls, and contact angle;

[0078] Import the 3D model into the finite element analysis software, perform material settings, contact mode settings, mesh generation, and apply loads and constraints to establish a twin finite element model of the rolling bearing. Use the optimized Latin hypercube experimental design method to generate n groups of randomly ordered dynamic parameter sample sequences that are evenly distributed within the parameter variation range. Using the parametric sweep function of the finite element analysis software, extract each group and substitute them into the finite element model for modal analysis, solving for the first four natural frequencies corresponding to the n groups under different input parameters.

[0079] The input and output parameters are merged to generate an input-output dataset (i.e., twin dataset) for training the deep neural network. A deep neural network model is constructed, and the deep neural network is trained using the twin dataset. m groups of test sample datasets of non-training sample points are randomly generated to test the accuracy of the trained deep network model. The global mean absolute error and global mean absolute percentage error of the test set are used as evaluation indicators to evaluate the performance of deep neural network models at different layers. Finally, the deep neural network model with the best performance is selected as the twin finite element model of the rolling bearing to achieve digital mirroring of the solid finite element model.

[0080] In S4, the parameter update method considers the signal characteristics and similarities of multiple signal sources such as time domain and frequency domain. In order to obtain the best parameters and ensure a good match between simulated data and actual data, the optimization algorithm is used to update the parameters of the dynamic twin model. During the calibration process, multiple parameters need to be determined and updated, and these parameters need to match the physical measurements in many aspects. The mathematical expression of the objective function in the model update process is:

[0081]

[0082] is the updated parameter value after model optimization, TX 孪生 ,TY 真实 Represents the time domain waveforms of the twin signal and the actual signal, FX 孪生 , FY 真实 Represent the frequency domain waveforms of the twin signal and the actual signal after Fourier transform, W 孪生 , W 真实 Represent the time-frequency domain waveforms of the twin signal and the actual signal respectively, correlation represents the Pearson correlation coefficient, They are the peak values ​​of the two signals, respectively. With the goal of minimizing the three functions, the parameters of the twin model are optimized to achieve an accurate match with the actual data;

[0083] The multi-objective particle swarm optimization (MOPSO) is integrated into the model to dynamically update the parameters of the twin model according to the actual signal, thereby showing a powerful exploration capability and achieving rapid convergence. Its expression is as follows:

[0084] The update formula for particle velocity is:

[0085]

[0086] represent the velocity vector and position vector of particle i in the kth iteration, respectively. is the historical optimal position of particle i in the kth iteration, is the historical optimal position of the population, r1 and r2 are random numbers in the range [0,1], which can increase the randomness of the search, c1 and c2 are learning factors, and ω is the inertia weight. As the number of iterations increases, the ω inertia weight continues to decrease, making PSO have stronger global convergence ability in the early stage and stronger local convergence ability in the later stage;

[0087] ω max ,ω min Indicates the maximum and minimum values ​​of the inertia weight, while iter and iter max Indicates the current number of iterations and the maximum number of iterations;

[0088] The particle position update formula is:

[0089]

[0090] The MOPSO multi-objective optimization algorithm is used to optimize the parameter group in the existing multi-objective optimization algorithm based on Pareto advantage;

[0091] Pareto dominance is introduced into particle swarm optimization to solve multi-objective optimization problems. MOPSO aims to find a set of Pareto optimal solutions, where no solution dominates another solution among multiple objectives. The following operations are performed to determine the optimal value from multiple solutions: 1) Pareto optimal solutions are calculated and stored in an archive; 2) a roulette wheel selection method is used to determine the most advantageous solution, which effectively balances all objectives.

[0092] Working principle: Based on the parameters and operating conditions of actual bearings, a dynamic digital twin model with a model parameter update mechanism is constructed. It interacts with the bearing measurement signals in real time to update its parameters, simulates the evolution of roller relative slip and bearing surface morphology throughout its life cycle, and realizes real-time interaction with the actual bearing through the parameter update mechanism. It simulates the evolution of surface morphology in different degradation stages, generates sufficient life cycle twin data, and effectively compensates for the problem of insufficient actual data.

[0093] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0094] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a dynamic interactive twin model of a rolling bearing throughout its life cycle, characterized by: The following steps are involved: S1. Signal acquisition: Acquire the real-time operating parameters of the rolling bearing and obtain the vibration acceleration data of the rolling bearing throughout its life cycle; S2. Time-Frequency Domain Feature Extraction and Decomposition: Extract the time, frequency, and time-frequency domain characteristic parameters of the rolling bearing, calculate the eigenvalues, and connect the eigenvalues ​​of the full life cycle data sets in chronological order to form a characteristic curve. Set the alarm threshold analysis based on the normal distribution. Use the envelope entropy of the intrinsic mode function as the evaluation criterion for the variational mode decomposition effect. Utilize the genetic algorithm to optimize the penalty factor α, and adaptively select the optimal parameter. After the signal is decomposed, the decomposition level K is determined according to the spectral correlation coefficient method to extract the time-frequency domain features. S3, Digital Twin Modeling and Analysis: Draw a three-dimensional model of the rolling bearing and implement digital twin modeling of the rolling bearing using finite element analysis software. Using the characteristic parameters extracted in S2 as input, generate n groups of dynamic parameter samples uniformly distributed within the parameter variation range. Perform parametric sweeps and modal analysis on each group, calculate the first four natural frequencies corresponding to the n groups, merge the input and output parameters, and generate the input-output data set of the deep neural network. S4. Parameter optimization mechanism: Determine the optimization variables and use the value range of each parameter as a constraint condition. Combine the natural frequency prediction value of the twin model and the experimental value of the physical entity as the target value to construct the objective function. Establish a dynamic parameter optimization model for virtual-real mapping. Use the particle swarm optimization algorithm, use the relevant parameters of the virtual model as the optimization variables, and use the minimum relative error between the theoretical natural frequency calculated by the finite element model and the corresponding experimental value as the objective function to optimize and determine the relevant parameters of the twin physical model. S5. Twin data generation: Input the initialized geometric parameters into the full life cycle dynamic twin model to generate highly applicable full life cycle vibration acceleration twin data.

2. The method for constructing a dynamic interactive twin model of a rolling bearing throughout its life cycle according to claim 1, characterized in that: In S2, in the process of extracting time domain features, a characteristic curve is used in combination with a normal distribution for statistical analysis. Assume that a group of bearings collects N sets of experimental data from normal to failure, and the length of each set of experimental data is M. The characteristic index of each set of experimental data is calculated respectively, and the N sets of characteristic indexes are combined into a vector L. The mean μ and variance σ are calculated. The statistic U is known from the normal distribution characteristics as follows: The statistic U is used as the amplitude shift in the degradation phase; in any data set, the probability that a data point is n standard deviations away from the mean of the current data point is: Based on the characteristics of the normal distribution, approximately 90% of the eigenvalues ​​are selected to be within the range of ±c standard deviations. The top 10% of normal samples of the bearing are taken, and the value of 3σ+u is calculated as the threshold line. If n consecutive eigenvalues ​​exceed the selected fault warning value, it is considered that the bearing degradation stage has changed. The value of n is set by the characterization capability of the actual bearing degradation process. The characteristic values ​​of each set of experimental data are calculated, and the characteristic values ​​of the whole life data sets are connected in chronological order to form a vibration characteristic curve. The important time domain characteristic parameters that can characterize the bearing degradation characteristics are analyzed and extracted.

3. The method for constructing a dynamic interactive twin model of a rolling bearing throughout its life cycle according to claim 2, characterized in that: In S2, important frequency domain features are extracted from the full life vibration trend curve for analysis.

4. The method for constructing a dynamic interactive twin model of a rolling bearing throughout its life cycle according to claim 3, characterized in that: The time-frequency domain feature extraction and decomposition process in step S2 is as follows: Step 1: Set the parameters of the GA algorithm. The penalty factor α ranges from [100, 5000]. Based on the algorithm accuracy and speed, the population size is set to 20. The variables are encoded in 20-bit binary format. The number of iterations is set to 20. The optional parameter for copying the generation gap population is set to 0.

95. Initialize the crossover and mutation probabilities and assign them to each population. The parameter range and initial value can be dynamically adjusted according to the specific application scenario and signal characteristics. Step 2: Initialize the parameter α, loop through the individuals in the population, perform VMD decomposition on the signal, and calculate the envelope entropy of each IMF as the fitness function; Step 3: Genetic algorithm loop, starting from zero and looping genetic algebra, optimizing individuals through selection, crossover and mutation operations, selecting to retain or eliminate individuals according to the value of the fitness function, and taking the value with the minimum envelope entropy as the result of the fitness function F(x), that is: F(x)=min{E p1 ,E p2 ,…,E pi }(i=1,…,K); Step 4: Compare the best individuals of the source population and the target population. If they are the same, the best individual of the source population is retained. Otherwise, it is replaced by the best individual of the target population. The selected individual is placed in the new population to obtain the optimal fitness value of this iteration. Step 5: Loop through steps 2 to 4 until the maximum number of iterations is reached, the optimization stops, and the global optimal individual penalty factor α is obtained; Step 6: Substitute the parameter α optimized by the GA algorithm into the VMD algorithm to calculate the spectral correlation coefficient between the IMF component and the original signal; Step 7: When the minimum spectral correlation coefficient is less than the set threshold, K = K-1; Otherwise, K=K+1, and continue to execute steps 4-6 until the maximum number of decomposition levels is reached, and then stop decomposition.

5. The method for constructing a dynamic interactive twin model of a rolling bearing throughout its life cycle according to claim 1, characterized in that: In said S3, a three-dimensional model of the rolling bearing in a normal state is drawn, and the physical parameters of the rolling bearing required for establishing the model include bearing pitch diameter, ball diameter, number of balls, and contact angle; Import the 3D model into the finite element analysis software, perform material settings, contact mode settings, mesh division, and apply loads and constraints to establish a twin finite element model of the rolling bearing. Use the optimized Latin hypercube experimental design method to generate n groups of random, disordered, and evenly distributed dynamic parameter sample sequences within the parameter variation range. Use the parametric scanning function of the finite element analysis software to extract each group and substitute them into the finite element model. Modal analysis, solving the first four natural frequencies under n groups of different input parameters; The input and output parameters generated are merged to generate an input-output dataset or twin dataset for training the deep neural network. A deep neural network model is constructed, and the deep neural network is trained using the twin dataset. m groups of test sample datasets of non-training sample points are randomly generated. The accuracy of the trained deep network model is tested, and the global mean absolute error and global mean absolute percentage error of the test set are used as evaluation indicators to evaluate the performance of deep neural network models at different layers. Finally, the deep neural network model with the best performance is selected as the twin finite element model of the rolling bearing to achieve digital mirroring of the solid finite element model.

6. The method for constructing a dynamic interactive twin model of a rolling bearing throughout its life cycle according to claim 1, characterized in that: In S4, the optimization algorithm is used to update the parameters of the dynamic twin model. During the calibration process, multiple parameters need to be determined and updated. The mathematical expression of the objective function during the update process is: is the updated parameter value after model optimization, TX 孪生 ,TY 真实 Represents the time domain waveforms of the twin signal and the actual signal, FX 孪生 ,FY 真实 Represent the frequency domain waveforms of the twin signal and the actual signal after Fourier transform, W 孪生 ,W 真实 Represent the time-frequency domain waveforms of the twin signal and the actual signal respectively, correlation represents the Pearson correlation coefficient, They are the peak values ​​of the two signals. With the goal of minimizing the three functions, the parameters of the twin model are optimized to achieve an accurate match with the actual data; The multi-objective particle swarm optimization (MOPSO) is integrated into the model, and the parameters of the twin model are dynamically updated according to the actual signal. The expression is as follows: The update formula for particle velocity is: represent the velocity vector and position vector of particle i in the kth iteration, respectively. is the historical optimal position of particle i in the kth iteration, is the historical optimal position of the population, r1 and r2 are random numbers in the range [0,1], which can increase the randomness of the search, c1 and c2 are learning factors, and ω is the inertia weight; as the number of iterations increases, the ω inertia weight continues to decrease; ω max ,ω min Indicates the maximum and minimum values ​​of the inertia weight, while iter and iter max Indicates the current number of iterations and the maximum number of iterations; The particle position update formula is: The MOPSO multi-objective optimization algorithm is used to optimize the parameter group in the existing multi-objective optimization algorithm based on Pareto advantage.

7. A dynamic interactive twin model construction system for rolling bearings throughout their life cycle, characterized by A method for constructing a dynamic interactive twin model of a rolling bearing over its entire life cycle as described in any one of claims 1 to 6 above is provided to achieve automatic modeling and more accurate modeling.

8. An electronic device for constructing a dynamic interactive twin model of a rolling bearing throughout its life cycle, characterized in that A method for constructing a dynamic interactive twin model of a rolling bearing over its entire life cycle as described in any one of claims 1 to 6 above is provided to achieve automatic modeling and more accurate modeling.

9. A storage medium storing a computer program thereon, which, when executed, is used to implement the dynamic interactive twin model method of a full life cycle rolling bearing as described in any one of claims 1 to 6.

Images