A method and system for constructing an angular domain model of a fault rolling bearing based on a physical information neural network

Abstract

The application discloses a kind of based on physical information neural network's fault rolling bearing angular domain model construction method and system, method includes: the instantaneous angular displacement sequence and the encoder instantaneous angular velocity signal collected to physical information neural network is input, the initial value of setting dynamics parameter;Bearing dynamics model is embedded in the training process of network, solve the angular variable vibration displacement of fault bearing three degrees of freedom angular domain dynamics equation group, calculate disturbance torque;Physical loss is constructed;The residual of actual IAS is used as true value loss, and the overall loss is calculated in combination with physical loss;The accurate estimation value of dynamics parameter is obtained by network, it is input into the IAS response under different fault types and working conditions in fault bearing dynamics model;The expansion of the sample of fault diagnosis data set sample is completed using the generated sample obtained. It is used to improve the imbalance intelligent diagnosis problem that normal sample amount is far more than fault sample under actual working condition, and the IAS signal sample of fault bearing is supplemented.

Description

Technical Field

[0001] This invention relates to the field of bearing fault diagnosis technology, specifically to a method and system for constructing a digital twin model of a faulty rolling bearing angular domain based on a physical information neural network. Background Technology

[0002] As a core component of transmission systems, the health status of rolling bearings directly determines the operational reliability of critical equipment. For example, in the joint units of industrial robots, even slight wear can lead to a significant decrease in motion control accuracy. Therefore, real-time monitoring and fault detection of rolling bearings are crucial for ensuring the normal operation of equipment. Existing condition monitoring and fault detection methods are mainly divided into signal processing methods and deep learning-based artificial intelligence methods; however, intelligent fault diagnosis methods typically require massive amounts of fault samples. Sample augmentation technology based on digital twins, utilizing both dynamic models and measured data, has gained increasing attention. This method requires only a small number of measured signal samples to generate a large number of twin fault samples, reducing experimental costs. Therefore, a method for augmenting bearing fault IAS signal samples has been proposed.

[0003] Recently, Physical Information Neural Networks (PINNs) have provided a new approach for intelligent bearing fault diagnosis by embedding physical laws into the loss function. Some of these methods have already been successfully applied. For example, Qin et al. proposed a digital twin method for bearing fault vibration signals based on inverse PINN by constraining the true value errors in the frequency domain of measured and simulated signals. Compared to vibration signals, IAS signals have advantages such as non-invasive measurement, no periodic calibration, and direct correlation with dynamics, and have been widely used in fault diagnosis. However, research on twin methods for bearing IAS fault samples has not yet been carried out. On the other hand, the above methods mainly use physical information, such as frequency domain information, to solve the signal distribution problem, rather than coupling the physical laws of the dynamic model with PINN to solve the dynamic response and estimate unknown dynamic parameters. Therefore, it is worthwhile to explore a method that couples the differential equations of the fault dynamic model with PINN to construct a digital twin model. Summary of the Invention

[0004] The purpose of this invention is to provide a method for constructing a digital twin model of a faulty rolling bearing angular domain based on a physical information neural network, addressing the aforementioned problems. This method improves the unbalanced intelligent diagnosis problem where the number of normal samples far exceeds the number of faulty samples under actual working conditions, and supplements the IAS signal samples of the faulty bearing.

[0005] The technical solution of the present invention is as follows:

[0006] A method for constructing a digital twin model of a faulty rolling bearing in the angular domain based on a physical information neural network includes the following steps:

[0007] The instantaneous angular displacement (IAD) sequence and the acquired encoder instantaneous angular velocity (IAS) signal are input into the physical information neural network; the initial values ​​of the dynamic parameters are set according to expert experience, and the dynamic parameters are implicitly encoded into the learnable parameters in the network.

[0008] A dynamic model of the angular domain of the faulty bearing is constructed. Based on the structural parameters and working conditions of the faulty bearing, the three-degree-of-freedom angular domain dynamic equation of the rolling bearing is solved by the fourth-order Runge-Kutta algorithm and tensor calculation to obtain the angular variable vibration displacement of the inner ring and calculate the angular variable disturbance torque.

[0009] To construct the physical loss, we take the differential equation of the torsional direction in the dynamic model of the three-degree-of-freedom faulted bearing and calculate the residual. At the same time, we embed the dynamic parameters to be learned into the differential equation of the loss and iteratively calculate it.

[0010] The residuals of the measured IAS are used as the true loss, and the total loss is calculated by combining them with the physical loss. Weight coefficients are set to make the IAS response of the network regression close to the real working conditions, while also ensuring that the network regression process satisfies the actual laws in the physical loss.

[0011] The total loss is reduced by optimizing the algorithm, the network is trained until convergence, and the backpropagation algorithm is used to drive iterative optimization of the dynamic parameters.

[0012] Accurate estimates of dynamic parameters are obtained through the network and input into the dynamic model of the faulty bearing to obtain the IAS response under different fault types and operating conditions.

[0013] Use the obtained generated samples to expand the fault diagnosis dataset samples.

[0014] The above method solves the problem of large errors caused by the reliance on expert experience in setting traditional bearing dynamic parameters. By optimizing both the physical loss function and the true loss function, and combining them with the backpropagation algorithm, high-precision iterative estimation of dynamic parameters is achieved. The accurate solution of dynamic parameters significantly improves the accuracy of the dynamic model of the faulty bearing angular domain, laying a physical foundation for subsequent fault simulation.

[0015] Furthermore, the construction of the angular domain dynamic model of the faulty bearing specifically includes:

[0016] In the proposed model This indicates the rotational degrees of freedom of the inner ring. The number of rolling elements. For the first The angular position of a rolling element, when the rolling element is in pure rolling motion and has no relative slippage with the raceway, is expressed as follows:

[0017] ,

[0018] in, For the first The initial position of each rolling element To maintain the angular frequency of the cage;

[0019] The outer ring of the partial fault rectangle has a length and width of [missing information]. H , L To reduce the complexity of fault modeling, the displacement excitation caused by local defects in the bearing is simplified to the increment of the bearing's internal clearance when the rolling element passes over the fault. Therefore, the additional radial displacement can be expressed as:

[0020] ,

[0021] in, This refers to the radial clearance of the bearing. This is a function to determine whether the rolling element has entered a fault state; the corresponding circumferential angle for the fault is determined by... L Calculations are performed when the rolling element enters the fault region. Other positions are β j =0;

[0022] The corner domain model considers the rolling friction between the rolling element and the inner raceway. When the rolling element is rolling purely, it is subjected to the normal forces of the inner and outer raceways. and tangential force Function: To shift the point of application of the normal force generated by rolling friction forward in the direction of relative motion of the rolling elements. , ,in , Let be the radius of the rolling element, and be the th . The expression for the normal force on a rolling element is:

[0023] ,

[0024] in, k b The equivalent stiffness for lateral vibration of the bearing; the inner ring is subjected to the first The tangential force of each rolling element is expressed as follows:

[0025] ,

[0026] The disturbance torque on the inner ring is:

[0027] ,

[0028] Among them, R i The inner radius;

[0029] Therefore, the established set of differential equations for the angular domain dynamics of a three-degree-of-freedom bearing outer ring fault is as follows:

[0030] ,

[0031] in, The lumped mass of the bearing-rotor system; Let the system's moment of inertia be denoted by . Input torque; It is the bearing's lateral vibration damping; It is the bearing torsional vibration damping; It is the lateral vibration stiffness; It is torsional stiffness; , They are respectively , Axial load; This refers to the instantaneous angular velocity response of the bearing. The function that determines whether a rolling element is in contact with the raceway is expressed as:

[0032] ,

[0033] The inner ring angular vibration displacement was obtained using the fourth-order Runge-Kutta method.

[0034] Calculate angular torque ,in x and y for X and Y The inner ring angular vibration displacement in the axial direction.

[0035] Furthermore, the physical loss is characterized by:

[0036] ,

[0037] in, is the length of the instantaneous angular displacement sequence.

[0038] Furthermore, the truth loss is characterized by:

[0039] ,

[0040] in, The length of the measured IAS signal sequence; These are the weights in a neural network; The IAS signal representing the neural network fit; The measured IAS signal is used; the total loss function is:

[0041] ,

[0042] in, and It is a weighted hyperparameter used to balance the interaction between the two types of losses.

[0043] This application also includes a faulty rolling bearing angular domain system based on a physical information neural network, comprising:

[0044] The instantaneous angular displacement (IAD) sequence and the acquired encoder instantaneous angular velocity (IAS) signal are input into the physical information neural network; the initial values ​​of the dynamic parameters are set, and the dynamic parameters are implicitly encoded into the learnable parameters in the network;

[0045] The module for constructing the angular domain dynamic model of the faulty bearing: Based on the structural parameters and working conditions of the faulty bearing, the three-degree-of-freedom angular domain dynamic equation of the rolling bearing is solved by using the fourth-order Runge-Kutta algorithm and tensor calculation to obtain the angular variable vibration displacement of the inner ring and calculate the angular variable disturbance torque;

[0046] Physical loss construction module: Take the differential equation of the torsional direction in the dynamic model of the three-degree-of-freedom fault bearing and calculate the residual. At the same time, embed the dynamic parameters to be learned into the differential equation of the loss for iterative calculation.

[0047] The truth loss construction module uses the residuals of the measured IAS as the truth loss, combines the physical loss to calculate the overall loss, and sets weight coefficients to make the IAS response of the network regression closer to the real working conditions.

[0048] Optimization module: Reduces the total loss through optimization algorithms, trains the network until convergence, and uses the backpropagation algorithm to drive iterative optimization of dynamic parameters;

[0049] Output module: Obtains accurate estimates of dynamic parameters through the network, inputs them into the faulty bearing dynamic model to obtain IAS responses under different fault types and operating conditions; uses the obtained generated samples to expand the fault diagnosis dataset samples.

[0050] Furthermore, the faulty bearing angular domain dynamic model construction module specifically includes:

[0051] In the proposed model This indicates the rotational degrees of freedom of the inner ring. The number of rolling elements. For the first The angular position of a rolling element, when the rolling element is in pure rolling motion and has no relative slippage with the raceway, is expressed as follows:

[0052] ,

[0053] in, For the first The initial position of each rolling element To maintain the angular frequency of the cage;

[0054] The outer ring of the partial fault rectangle has a length and width of [missing information]. H , L This means that the displacement excitation caused by local defects in the bearing can be simplified to the increment of the bearing's internal clearance when the rolling element passes through the fault. Therefore, the additional radial displacement can be expressed as:

[0055] ,

[0056] in, This refers to the radial clearance of the bearing. This is a function to determine whether the rolling element has entered a fault state; the corresponding circumferential angle for the fault is determined by... L Calculations are performed when the rolling element enters the fault region. Other positions are β j =0;

[0057] When the rolling element is rolling purely, it is subjected to normal forces from the inner and outer raceways. and tangential force Function: To shift the point of application of the normal force generated by rolling friction forward in the direction of relative motion of the rolling elements. , ,in , Let be the radius of the rolling element, and be the th . The expression for the normal force on each rolling element is:

[0058] ,

[0059] in, k b The equivalent stiffness for lateral vibration of the bearing; the inner ring is subjected to the first The tangential force of each rolling element is expressed as follows:

[0060] ,

[0061] The disturbance torque on the inner ring is:

[0062] ,

[0063] Among them, R i The inner radius;

[0064] Therefore, the established set of differential equations for the angular domain dynamics of a three-degree-of-freedom bearing outer ring fault is as follows:

[0065] ,

[0066] in, The lumped mass of the bearing-rotor system; Let the system's moment of inertia be denoted by . Input torque; It is the bearing's lateral vibration damping; It is the bearing torsional vibration damping; It is the lateral vibration stiffness; It is torsional stiffness; , They are respectively , Axial load; This refers to the instantaneous angular velocity response of the bearing. The function that determines whether a rolling element is in contact with the raceway is expressed as:

[0067] ,

[0068] The inner ring angular vibration displacement was obtained using the fourth-order Runge-Kutta method.

[0069] Calculate angular torque ,in x and y for X and Y The inner ring angular vibration displacement in the axial direction.

[0070] Furthermore, the physical loss construction module is as follows:

[0071] ,

[0072] in, is the length of the instantaneous angular displacement sequence.

[0073] Furthermore, the truth loss construction module is as follows:

[0074] ,

[0075] in, The length of the measured IAS signal sequence; These are the weights in a neural network; The IAS signal representing the neural network fit; The measured IAS signal is used; the total loss function is:

[0076] ,

[0077] in, and It is a weighted hyperparameter used to balance the interaction between the two types of losses.

[0078] This application also includes a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the method for constructing a fault rolling bearing angular domain model based on a physical information neural network.

[0079] This application also includes a storage medium storing a computer program, which, when executed by a processor, implements the steps in the method for constructing a fault rolling bearing angular domain model based on a physical information neural network.

[0080] Compared with existing technologies, the advantages of this invention are:

[0081] The accurate solution of bearing dynamic parameters (currently only approximate estimates based on expert experience) has been achieved, thus improving the accuracy and reliability of the dynamic model; the function of generating fault samples across operating conditions has been successfully developed, effectively alleviating the problem of imbalance between healthy and faulty samples that is common in the dataset; the need for actual experiments has been greatly reduced, the time and economic costs of sample acquisition have been significantly reduced, and the efficiency of research and economic production has been improved. Attached Figure Description

[0082] Figure 1 Schematic diagram for modeling the lumped mass of a bearing.

[0083] Figure 2 A simplified diagram of the fault dimensions for the dynamic model.

[0084] Figure 3 This is a schematic diagram of the overall structure of the physical information neural network in this application. Detailed Implementation

[0085] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0086] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0087] Please see Figure 1-3 A method for constructing a faulty rolling bearing angular domain model based on a physical information neural network, such as... Figure 1 , Figure 2 and Figure 3 As shown, it includes the following steps:

[0088] Input the structural parameters of the rolling bearing. The bearing model used in the experiment is SKF6205. The specific structural parameters can be obtained from the bearing's design parameters. The measured IAS signal acquired from the IAS measurement test bench is input into the physical information neural network. Based on expert experience, key dynamic parameters are set: torsional stiffness is set to 0 N•m / rad, and torsional damping is set to 0.015 N•s•m / rad.

[0089] A three-degree-of-freedom angular domain dynamic model of a faulty bearing is constructed. Based on the structural parameters and working conditions of the faulty bearing, the three-degree-of-freedom angular domain dynamic equation of the rolling bearing is solved using the fourth-order Runge-Kutta algorithm and tensor calculation to obtain the angular variable vibration displacement of the inner ring and calculate the angular variable disturbance torque.

[0090] The construction of a three-degree-of-freedom fault bearing angular domain dynamic model specifically includes:

[0091] In the proposed model This indicates the rotational degrees of freedom of the inner ring. The number of rolling elements. For the first The angular position of a rolling element, when the rolling element is in pure rolling motion and has no relative slippage with the raceway, is expressed as follows:

[0092] ,

[0093] in, For the first The initial position of each rolling element To maintain the angular frequency of the cage;

[0094] The outer ring of the partial fault rectangle has a length and width of [missing information]. H , L This means that the displacement excitation caused by local defects in the bearing can be simplified to the increment of the bearing's internal clearance when the rolling element passes through the fault. Therefore, the additional radial displacement can be expressed as:

[0095] ,

[0096] in, This refers to the radial clearance of the bearing. This is a function to determine whether the rolling element has entered a fault state; the corresponding circumferential angle for the fault is determined by... L Calculations are performed when the rolling element enters the fault region. Other positions are βj =0;

[0097] When the rolling element is rolling purely, it is subjected to normal forces from the inner and outer raceways. and tangential force Function: To shift the point of application of the normal force generated by rolling friction forward in the direction of relative motion of the rolling elements. , ,in , Let be the radius of the rolling element, and be the th . The expression for the normal force on each rolling element is:

[0098] ,

[0099] in, k b The equivalent stiffness for lateral vibration of the bearing; the inner ring is subjected to the first The tangential force of each rolling element is expressed as follows:

[0100] ,

[0101] The disturbance torque on the inner ring is:

[0102] ,

[0103] Among them, R i The inner radius;

[0104] Therefore, the established set of differential equations for the angular domain dynamics of a three-degree-of-freedom bearing outer ring fault is as follows:

[0105] ,

[0106] in, The lumped mass of the bearing-rotor system; Let the system's moment of inertia be denoted by . Input torque; It is the bearing's lateral vibration damping; It is the bearing torsional vibration damping; It is the lateral vibration stiffness; It is torsional stiffness; , They are respectively , Axial load; This refers to the instantaneous angular velocity response of the bearing. The function that determines whether a rolling element is in contact with the raceway is expressed as:

[0107] ,

[0108] The inner ring angular vibration displacement was obtained using the fourth-order Runge-Kutta method.

[0109] Calculate angular torque ,in x and y for X and Y The inner ring angular vibration displacement in the axial direction.

[0110] In the three-degree-of-freedom dynamic model, the differential equations in the rotational degree-of-freedom direction are embedded with physical losses. The instantaneous angular displacement sequence, angular disturbance torque, and dynamic parameters to be estimated are input, and the 2-norm is calculated to obtain the specific physical losses.

[0111] The physical loss is:

[0112] ,

[0113] in, is the length of the instantaneous angular displacement sequence.

[0114] The true value loss is calculated by taking the mean square error between the measured IAS signal and the IAS obtained from the network solution. The true value loss is:

[0115] ,

[0116] in, The length of the measured IAS signal sequence; These are the weights in a neural network; The IAS signal representing the neural network fit; The measured IAS signal is used; the total loss function is:

[0117] ,

[0118] in, and These are the weight hyperparameters used to balance the interaction between the two types of losses. In the total loss function, the first term constrains the consistency between the neural network output and the dynamic model, ensuring that the neural network learns the correct physical laws; the second term constrains the consistency between the neural network output and the measured data, ensuring the network's prediction accuracy and preserving as much working condition information as possible. In the physical constraint loss function, learnable dynamic parameters are introduced. and This allows neural networks to not only optimize data fitting errors during training, but also guide the estimation of dynamic parameters through angular domain differential equations. This setup enables the neural network to gradually approximate true physical parameters while ensuring adherence to physical laws.

[0119] Table 1 shows the neural network structure parameters in the proposed method.

[0120] Tab.1 Proposed neural network structure parameters

[0121]

[0122] The neural network was trained until convergence by calculating physical loss and ground value loss. The hardware used for computation was an Intel i5-12600 CPU with 16 GB of RAM and an NVIDIA 4060 GPU with 8 GB of RAM. The network architecture was built and trained using Python 3.11 and Pytoch 2.5.1. The structure and parameters of the FCN used are shown in Table 1. Each training sample contained 10,000 points. In the proposed method, the PINN training setting was Epoches = 500,000. The Adam optimizer was used, with a learning rate of 0.001 and first-moment decay estimation. β =1=0.9, the second moment estimates the decay rate β2=0.999, and convergence is accelerated by adaptive learning rate adjustment. and The parameters are implicitly encoded into the network weights as learnable parameters and updated synchronously through backpropagation.

[0123] The exact solution of the angular domain dynamic parameters of the bearing is obtained by using a physical information neural network. This solution is then input into the angular domain dynamic model of the faulty bearing. Different fault parameters and operating condition parameters are set, and the IAS response under different fault conditions is obtained by using the fourth-order Runge-Kutta algorithm.

[0124] The accurate solution of bearing dynamic parameters (currently only approximate estimates based on expert experience) has been achieved, thus improving the accuracy and reliability of the dynamic model; the function of generating fault samples across operating conditions has been successfully developed, effectively alleviating the problem of imbalance between healthy and faulty samples that is common in the dataset; the need for actual experiments has been greatly reduced, the time and economic costs of sample acquisition have been significantly reduced, and the efficiency of research and economic production has been improved.

[0125] This application also includes a faulty rolling bearing angular domain system based on a physical information neural network, comprising:

[0126] The instantaneous angular displacement (IAD) sequence and the acquired encoder instantaneous angular velocity (IAS) signal are input into the physical information neural network; the initial values ​​of the dynamic parameters are set, and the dynamic parameters are implicitly encoded into the learnable parameters in the network;

[0127] The module for constructing the angular domain dynamic model of the faulty bearing: Based on the structural parameters and working conditions of the faulty bearing, the three-degree-of-freedom angular domain dynamic equation of the rolling bearing is solved by using the fourth-order Runge-Kutta algorithm and tensor calculation to obtain the angular variable vibration displacement of the inner ring and calculate the angular variable disturbance torque;

[0128] Physical loss construction module: Take the differential equation of the torsional direction in the dynamic model of the three-degree-of-freedom fault bearing and calculate the residual. At the same time, embed the dynamic parameters to be learned into the differential equation of the loss for iterative calculation.

[0129] The truth loss construction module uses the residuals of the measured IAS as the truth loss, combines the physical loss to calculate the overall loss, and sets weight coefficients to make the IAS response of the network regression closer to the real working conditions.

[0130] Optimization module: Reduces the total loss through optimization algorithms, trains the network until convergence, and uses the backpropagation algorithm to drive iterative optimization of dynamic parameters;

[0131] Output module: Obtains accurate estimates of dynamic parameters through the network, inputs them into the faulty bearing dynamic model to obtain IAS responses under different fault types and operating conditions; uses the obtained generated samples to expand the fault diagnosis dataset samples.

[0132] The module for constructing the angular domain dynamic model of the faulty bearing specifically includes:

[0133] In the proposed model This indicates the rotational degrees of freedom of the inner ring. The number of rolling elements. For the first The angular position of a rolling element, when the rolling element is in pure rolling motion and has no relative slippage with the raceway, is expressed as follows:

[0134] ,

[0135] in, For the first The initial position of each rolling element To maintain the angular frequency of the cage;

[0136] The outer ring of the partial fault rectangle has a length and width of [missing information]. H , L This means that the displacement excitation caused by local defects in the bearing can be simplified to the increment of the bearing's internal clearance when the rolling element passes through the fault. Therefore, the additional radial displacement can be expressed as:

[0137] ,

[0138] in, This refers to the radial clearance of the bearing. This is a function to determine whether the rolling element has entered a fault state; the corresponding circumferential angle for the fault is determined by... L Calculations are performed when the rolling element enters the fault region. Other positions are β j =0;

[0139] When the rolling element is rolling purely, it is subjected to normal forces from the inner and outer raceways. and tangential force Function: To shift the point of application of the normal force generated by rolling friction forward in the direction of relative motion of the rolling elements. , ,in , Let be the radius of the rolling element, and be the th . The expression for the normal force on each rolling element is:

[0140] ,

[0141] in, k b The equivalent stiffness for lateral vibration of the bearing; the inner ring is subjected to the first The tangential force of each rolling element is expressed as follows:

[0142] ,

[0143] The disturbance torque on the inner ring is:

[0144] ,

[0145] Among them, R i The inner radius;

[0146] Therefore, the established set of differential equations for the angular domain dynamics of a three-degree-of-freedom bearing outer ring fault is as follows:

[0147] ,

[0148] in, The lumped mass of the bearing-rotor system; Let the system's moment of inertia be denoted by . Input torque; It is the bearing's lateral vibration damping; It is the bearing torsional vibration damping; It is the lateral vibration stiffness; It is torsional stiffness; , They are respectively , Axial load; This refers to the instantaneous angular velocity response of the bearing. The function that determines whether a rolling element is in contact with the raceway is expressed as:

[0149] ,

[0150] The inner ring angular vibration displacement was obtained using the fourth-order Runge-Kutta method.

[0151] Calculate angular torque ,in x and y for X and Y The inner ring angular vibration displacement in the axial direction.

[0152] The physical loss construction module is:

[0153] ,

[0154] in, is the length of the instantaneous angular displacement sequence.

[0155] The truth loss construction module is:

[0156] ,

[0157] in, The length of the measured IAS signal sequence; These are the weights in a neural network; The IAS signal representing the neural network fit; The measured IAS signal is used; the total loss function is:

[0158] ,

[0159] in, and It is a weighted hyperparameter used to balance the interaction between the two types of losses.

[0160] This application also includes a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement steps such as those in a method for constructing a faulty rolling bearing angular domain model based on a physical information neural network.

[0161] This application also includes a storage medium storing a computer program, which, when executed by a processor, implements the steps in a method for constructing a fault rolling bearing angular domain model based on a physical information neural network.

[0162] The embodiments described above merely illustrate specific implementation methods of this application, and while the descriptions are detailed and specific, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the technical solution of this application, and these modifications and improvements all fall within the scope of protection of this application.

Claims

1. A method for constructing a faulty rolling bearing angular domain model based on a physical information neural network, characterized in that, Includes the following steps: The instantaneous angular displacement (IAD) sequence and the acquired encoder instantaneous angular velocity (IAS) signal are input into the physical information neural network; the initial values ​​of the dynamic parameters are set, and the dynamic parameters are implicitly encoded into the learnable parameters in the network; A dynamic model of the angular domain of the faulty bearing is constructed. Based on the structural parameters and working conditions of the faulty bearing, the three-degree-of-freedom angular domain dynamic equation of the rolling bearing is solved by the fourth-order Runge-Kutta algorithm and tensor calculation to obtain the angular variable vibration displacement of the inner ring and calculate the angular variable disturbance torque. To construct the physical loss, we take the differential equation of the torsional direction in the dynamic model of the three-degree-of-freedom faulted bearing and calculate the residual. At the same time, we embed the dynamic parameters to be learned into the differential equation of the loss and iteratively calculate it. The residuals of the measured IAS are used as the true loss, and the total loss is calculated by combining the physical loss. Weighting coefficients are set to make the IAS response of the network regression closer to the real working conditions. The total loss is reduced by optimizing the algorithm, the network is trained until convergence, and the backpropagation algorithm is used to drive iterative optimization of the dynamic parameters. Accurate estimates of dynamic parameters are obtained through the network and input into the dynamic model of the faulty bearing to obtain the IAS response under different fault types and operating conditions. Use the obtained generated samples to expand the fault diagnosis dataset samples; The specific steps involved in constructing the angular domain dynamic model of the faulty bearing are as follows: In the proposed model This indicates the rotational degrees of freedom of the inner ring. The number of rolling elements. For the first The angular position of a rolling element, when the rolling element is in pure rolling motion and has no relative slippage with the raceway, is expressed as follows: ， in, For the first The initial position of each rolling element To maintain the angular frequency of the cage; The outer ring of the partial fault rectangle has a length and width of [missing information]. H , L This means that the displacement excitation caused by local defects in the bearing can be simplified to the increment of the bearing's internal clearance when the rolling element passes through the fault. Therefore, the additional radial displacement can be expressed as: ， in, This refers to the radial clearance of the bearing. This is a function to determine whether the rolling element has entered a fault state; the corresponding circumferential angle for the fault is determined by... L Calculations are performed when the rolling element enters the fault region. Other positions are β j =0; When the rolling element is rolling purely, it is subjected to normal forces from the inner and outer raceways. and tangential force Function: To shift the point of application of the normal force generated by rolling friction forward in the direction of relative motion of the rolling elements. , ,in , Let be the radius of the rolling element, and be the th . The expression for the normal force on a rolling element is: ， in, k b The equivalent stiffness for lateral vibration of the bearing; the inner ring is subjected to the first The tangential force of each rolling element is expressed as follows: ， The disturbance torque on the inner ring is: ， Among them, R i The inner radius; Therefore, the established set of differential equations for the angular domain dynamics of a three-degree-of-freedom bearing outer ring fault is as follows: ， in, The lumped mass of the bearing-rotor system; Let the system's moment of inertia be denoted by . Input torque; It is the bearing's lateral vibration damping; It is the bearing torsional vibration damping; It is the lateral vibration stiffness; It is torsional stiffness; , They are respectively , Axial load; This refers to the instantaneous angular velocity response of the bearing. The function that determines whether a rolling element is in contact with the raceway is expressed as: ， The inner ring angular vibration displacement was obtained using the fourth-order Runge-Kutta method. Calculate angular torque ,in x and y for X and Y The inner ring angular vibration displacement in the axial direction.

2. The method for constructing a faulty rolling bearing angular domain model based on a physical information neural network according to claim 1, characterized in that, The physical loss is: ， in, is the length of the instantaneous angular displacement sequence.

3. The method for constructing a faulty rolling bearing angular domain model based on a physical information neural network according to claim 2, characterized in that, The truth loss is: ， in, The length of the measured IAS signal sequence; These are the weights in a neural network; The IAS signal representing the neural network fit; The measured IAS signal is used; the total loss function is: ， in, and It is a weighted hyperparameter used to balance the interaction between the two types of losses.

4. A fault rolling bearing angular domain system based on a physical information neural network, characterized in that, include: The instantaneous angular displacement (IAD) sequence and the acquired encoder instantaneous angular velocity (IAS) signal are input into the physical information neural network. Set initial values ​​for the dynamic parameters and implicitly encode the dynamic parameters into the learnable parameters in the network; The module for constructing the angular domain dynamic model of the faulty bearing: Based on the structural parameters and working conditions of the faulty bearing, the three-degree-of-freedom angular domain dynamic equation of the rolling bearing is solved by using the fourth-order Runge-Kutta algorithm and tensor calculation to obtain the angular variable vibration displacement of the inner ring and calculate the angular variable disturbance torque; Physical loss construction module: Take the differential equation of the torsional direction in the dynamic model of the three-degree-of-freedom fault bearing and calculate the residual. At the same time, embed the dynamic parameters to be learned into the differential equation of the loss for iterative calculation. The truth loss construction module uses the residuals of the measured IAS as the truth loss, combines the physical loss to calculate the overall loss, and sets weight coefficients to make the IAS response of the network regression closer to the real working conditions. Optimization module: Reduces the total loss through optimization algorithms, trains the network until convergence, and uses the backpropagation algorithm to drive iterative optimization of dynamic parameters; Output module: Obtains accurate estimates of dynamic parameters through the network, inputs them into the faulty bearing dynamic model to obtain IAS response under different fault types and operating conditions; uses the obtained generated samples to expand the fault diagnosis dataset samples; The module for constructing the angular domain dynamic model of the faulty bearing specifically includes: In the proposed model This indicates the rotational degrees of freedom of the inner ring. The number of rolling elements. For the first The angular position of a rolling element, when the rolling element is in pure rolling motion and has no relative slippage with the raceway, is expressed as follows: ， in, For the first The initial position of each rolling element To maintain the angular frequency of the cage; The outer ring of the partial fault rectangle has a length and width of [missing information]. H , L This means that the displacement excitation caused by local defects in the bearing can be simplified to the increment of the bearing's internal clearance when the rolling element passes through the fault. Therefore, the additional radial displacement can be expressed as: ， in, This refers to the radial clearance of the bearing. This is a function to determine whether the rolling element has entered a fault state; the corresponding circumferential angle for the fault is determined by... L Calculations are performed when the rolling element enters the fault region. Other positions are β j =0; When the rolling element is rolling purely, it is subjected to normal forces from the inner and outer raceways. and tangential force Function: To shift the point of application of the normal force generated by rolling friction forward in the direction of relative motion of the rolling elements. , ,in , Let be the radius of the rolling element, and be the th . The expression for the normal force on each rolling element is: ， in, k b The equivalent stiffness for lateral vibration of the bearing; the inner ring is subjected to the first The tangential force of each rolling element is expressed as follows: ， The disturbance torque on the inner ring is: ， Among them, R i The inner radius; Therefore, the established set of differential equations for the angular domain dynamics of a three-degree-of-freedom bearing outer ring fault is as follows: ， in, The lumped mass of the bearing-rotor system; Let the system's moment of inertia be denoted by . Input torque; It is the bearing's lateral vibration damping; It is the bearing torsional vibration damping; It is the lateral vibration stiffness; It is torsional stiffness; , They are respectively , Axial load; This refers to the instantaneous angular velocity response of the bearing. The function that determines whether a rolling element is in contact with the raceway is expressed as: ， The inner ring angular vibration displacement was obtained using the fourth-order Runge-Kutta method. Calculate angular torque ,in x and y for X and Y The inner ring angular vibration displacement in the axial direction.

5. A fault rolling bearing angular domain system based on a physical information neural network according to claim 4, characterized in that, The physical loss construction module is: ， in, is the length of the instantaneous angular displacement sequence.

6. A fault rolling bearing angular domain system based on a physical information neural network according to claim 5, characterized in that, The truth loss construction module is: ， in, The length of the measured IAS signal sequence; These are the weights in a neural network; The IAS signal representing the neural network fit; The measured IAS signal is used; the total loss function is: ， in, and It is a weighted hyperparameter used to balance the interaction between the two types of losses.

7. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the method for constructing a fault rolling bearing angular domain model based on a physical information neural network as described in any one of claims 1-3.

8. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps in the method for constructing a fault rolling bearing angular domain model based on a physical information neural network as described in any one of claims 1-3.

Images