Rolling bearing fault diagnosis method based on mechanism and data fusion

Abstract

The application relates to the field of aero-engine main bearing fault diagnosis, and discloses a rolling bearing fault diagnosis method based on mechanism and data fusion, which generates bearing fault state data by constructing a bearing dynamics model and comprehensively considering key influencing factors such as bearing fault size, running speed and bearing load; the difference between the bearing fault state data and bearing health state data is used as an optimization function; an improved eagle fish optimization algorithm is adopted to optimize the damping coefficient and stiffness of the bearing inner ring and the damping coefficient and stiffness of the bearing outer ring; finally, the output data of the corrected bearing dynamics model are analyzed by using a TCN-BiGRU-CA model to output a fault diagnosis result. The application can effectively solve the problem that the simulation data and the measured data are too different due to parameter simplification of the traditional dynamics model, and significantly improves the accuracy and reliability of aero-engine main bearing fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis of main bearings in aero-engines, and discloses a method for fault diagnosis of rolling bearings based on mechanism and data fusion. Background Technology

[0002] Bearings are critical components of aero-engines, making fault detection of them of great significance. Because aero-engines often operate in complex and harsh environments characterized by high temperature, high pressure, high speed, and strong vibration, main bearings are prone to pitting, spalling, and wear, seriously threatening operational safety and potentially causing substantial economic losses. Existing fault diagnosis models are based on training under fixed operating conditions, but these are often difficult to apply directly when operating conditions change.

[0003] In recent years, in the field of bearing dynamics, numerous scholars have studied the dynamic response of bearings with single inner and outer ring failures, elucidating the vibration mechanism of failed bearings from multiple perspectives and achieving many results. Taking rolling bearings as the research object, some have explained the coupling mechanism of vibration signals during failure by establishing a four-degree-of-freedom dynamic model of a failed ball bearing; others have established a six-degree-of-freedom bearing dynamic simulation model with local surface defects by considering high-speed effects, gyroscopic torque, and centrifugal force.

[0004] However, the aforementioned dynamic model neglects the influence of modeling parameters on the dynamic model. Specifically, the modeling parameters are obtained based on theoretical calculations and empirical estimates, without refining them using measured data. This will lead to significant differences between the simulation data and the measured data, reducing the accuracy of the simulation data. Summary of the Invention

[0005] The purpose of this invention is to provide a rolling bearing fault diagnosis method based on mechanism and data fusion, which can effectively solve the problem of large differences between simulation data and measured data caused by parameter simplification in traditional dynamic models, and significantly improve the accuracy and reliability of fault diagnosis of aero-engine main bearings.

[0006] To achieve the above-mentioned technical effects, the technical solution adopted by the present invention is as follows:

[0007] A rolling bearing fault diagnosis method based on mechanism and data fusion includes:

[0008] Step 1: Establish model equations based on Hertz contact theory and construct a bearing dynamics model;

[0009] Step 2: Use the bearing dynamics model to simulate and generate healthy bearing state data;

[0010] Step 3: Introduce a geometrically regular rectangular spalling as a single-point damage fault of the raceway into the bearing dynamics model, and construct a contact deformation analysis model after introducing the damage fault.

[0011] Step 4: Use the contact deformation analysis model to simulate and obtain bearing failure status data. The bearing failure status data includes inner ring failure status data when rectangular spalling is located on the inner ring and outer ring failure status data when rectangular spalling is located on the outer ring.

[0012] Step 5: Using the difference between bearing fault state data and bearing health state data as the optimization function, the optimization objective is to minimize the output value of the optimization function. The damping coefficient and stiffness of the bearing inner ring and the bearing outer ring are used as optimization variables. The HFOA algorithm is used to correct the bearing dynamics model, and the optimized damping coefficient and stiffness of the bearing inner ring and the bearing outer ring are output, thus obtaining the corrected bearing dynamics model.

[0013] Step 6: Use the modified bearing dynamics model to generate bearing condition simulation data including fault size, spindle speed, load, and fault type;

[0014] Step 7: Establish a TCN-BiGRU model with embedded channel attention mechanism. Use the generated bearing state simulation data as the input of the TCN-BiGRU model, and the corresponding fault size label and fault type label as the output of the TCN-BiGRU model. Train the TCN-BiGRU model to obtain the trained TCN-BiGRU-CA model.

[0015] Step 8: Analyze the measured condition data of the bearing to be analyzed using the TCN-BiGRU-CA model, and output the fault diagnosis results.

[0016] Furthermore, in the TCN-BiGRU-CA model, the bearing's measured condition data first enters the TCN for feature extraction, and then is processed by the BiGRU to output the fault diagnosis results.

[0017] Furthermore, all translational motions within the bearing are constrained within a two-dimensional Cartesian coordinate system (xy plane), and all rotational motions of the bearing occur around the z-axis, which is perpendicular to the xy plane; the constructed bearing dynamics model is as follows:

[0018]

[0019] in For the mass of the bearing inner ring, This represents the damping coefficient of the bearing inner ring. This refers to the stiffness of the bearing inner ring; For the mass of the bearing outer ring, For the outer ring stiffness of the bearing, This refers to the damping coefficient of the bearing outer ring; Indicates the eccentricity; Let be the instantaneous displacement of the inner circle center in the x-direction. Let be the instantaneous displacement of the outer ring center in the x-direction. Let be the instantaneous displacement of the inner circle center in the y-direction. This represents the instantaneous displacement of the outer ring center in the y-direction; Let x be the component of the total supporting reaction force of the bearing on the inner ring in the x-direction. The rotational speed of the spindle that mates with the bearing. The rotation time relative to the initial position of the spindle. It is the acceleration due to gravity. This represents the component of the total supporting reaction force of the bearing on the inner ring in the y direction.

[0020] Furthermore, the constructed contact deformation analysis model is as follows: ,in To introduce the aforementioned damage fault, the first Contact deformation of each rolling element For the first A rolling element in The instantaneous angular position at a given moment. , It is the total number of rolling elements. It is the initial phase angle of the bearing. Represents the orbital angular velocity of the cage; This refers to the radial internal clearance of the bearing. This represents the equivalent deformation release caused by a local fault. , The fixed center corner position of the rectangular peeling area when the rectangular peeling is located on the outer ring. This represents the circumferential width of the rectangular peeling area when the rectangular peeling is located on the outer ring. The outer radius is... The fixed center corner position of the rectangular peeling area when the rectangular peeling is located in the inner circle. This represents the circumferential width of the rectangular peeling area when the rectangular peeling is located in the inner circle. The inner radius; ,in Where is the radius of the rolling element. The radial depth of the rectangular peeling.

[0021] Furthermore, in step five, the method for using the HFOA algorithm to correct the bearing dynamics model and output the optimized damping coefficient and stiffness of the bearing inner ring and the bearing outer ring includes:

[0022] S5.1 Within the search space formed by the given upper and lower limits of each optimization variable, a population generation strategy based on a fourth-order Chebyshev chaotic map is adopted to randomly generate an initial population formed by the decision values ​​of the initial positions of each optimization variable.

[0023] S5.2 Based on the initial preset step size vector and direction vector of each individual in the initial population, perform a preliminary update on the position of each individual in the corresponding search space to obtain the updated individual position. ,in It is the first The first individual The positions of the optimization variables after initial update It is the first The first individual Initial positions of the optimization variables For the first The first individual The initial preset step size vector of each optimization variable, For the first The first individual The initial preset direction vector of each optimization variable;

[0024] S5.3 After the initial update of individual positions, the entire population is divided into several subpopulations. Within each subpopulation, the optimal individual in the current subpopulation is identified as the subpopulation leader based on the fitness value of each individual. Within each subpopulation, individuals learn from the leader of that subpopulation and adjust their own positions to obtain the learned individual's position in the subpopulation. ,in, To control the subgroup learning coefficient of learning intensity, The leader of the current subgroup The position of each optimization variable The first outside the leader of the subgroup The first individual The position of each optimization variable;

[0025] S5.4 Iteratively learns and updates the position of the learned individual in the subpopulation. During the iterative learning and update process, the position of the corresponding individual in the subpopulation is generated by using the lens imaging back learning strategy based on the upper and lower limits of the search space for each individual in the population.

[0026] In each iteration of S5.4, S5.5 assigns a first visual range to the leader role within the subgroup, and assigns a second visual range to the other individuals within the subgroup as developers; the first visual range is larger than the second visual range; calculates the average fitness of the entire subgroup after learning and adjusting positions; if the average fitness is greater than a preset fitness threshold, a role switching mechanism for some developers is triggered; the developers who switch roles search for the neighborhood optimum with the lowest fitness within their corresponding visual range and move towards it; the positions of the developers who switch roles are updated until the average fitness of the subgroup is less than or equal to the preset fitness threshold; then, the individual with the lowest fitness within the subgroup is determined as the local optimum.

[0027] S5.6 The local optimum of the subpopulation with the lowest average fitness among all subpopulations in the current population is determined as the global optimum. The step size vector and direction vector of each individual are dynamically adjusted based on the global optimum of the current population and the local optimum of each subpopulation. Steps S5.3 to S5.6 are repeated until the preset maximum number of iterations is reached or the preset termination condition is met. The global optimum that records the minimum output value of the optimization function during the entire iteration process is output.

[0028] Furthermore, in step S5.1, the population generation strategy using a fourth-order Chebyshev chaotic map, and the method for randomly generating the initial population formed by the initial position decision values ​​of each optimization variable, includes:

[0029] Based on the search space formed by the given upper and lower limits of each optimization variable, the following approach is adopted: Randomly generate decision values ​​based on the initial positions of each optimization variable. ,in For the first The first individual The upper limit of each optimization variable within the search space. For the first The first individual The lower bound of each optimization variable within the search space. The initial chaotic variable is a random decimal between 0 and 1, representing the population generation strategy of the fourth-order Chebyshev chaotic map.

[0030] by The decision values ​​for the positions of each optimization variable are updated, and after reaching a preset number of iterations or convergence, information on the position of each optimization variable for each individual is generated. The initial population, It is the first The first individual Initial positions of the optimization variables The population generation strategy for the fourth-order Chebyshev chaotic map is the first... The chaotic variable is updated after the next iteration.

[0031] Furthermore, a lens imaging back-learning strategy is used to generate the position of the corresponding individual within the subgroup. ,in For the first The first individual The upper limit of each optimization variable within the search space. For the first The first individual The lower bound of each optimization variable within the search space. For adaptive scaling factor, , This is the initial value for the adaptive scaling factor. This represents the current iteration number. This represents the preset total number of iterations.

[0032] Furthermore, in step S5.5, the developer of the role-switching project updates the location. ,in, The developer of role transformation In the updated location, This is the developer's position before the update, where the role was changed. It is the step size factor that controls the movement distance. It is the location of the neighborhood optimal solution within the developer's visual range during the role transformation.

[0033] Furthermore, in step S5.6, the dynamically adjusted step size vector The dynamically adjusted direction vector ,in For dynamic adjustment of the first The first individual The step size vector of each optimization variable. The global learning coefficient. Let be a function that returns a random number in the interval [0, 1]. The globally optimal solution found in the current iteration is at the [number]th iteration. The position of each optimization variable For dynamic adjustment of the first The first individual The direction vector of each optimization variable, For local learning coefficients, The local optimum found in the current iteration number is the one at the [number]th iteration. The position of each optimization variable.

[0034] Compared with the prior art, the beneficial effects of this invention are as follows: While correcting the bearing dynamics model and ensuring the rapid convergence and strong robustness of the Eaglefish optimization algorithm, this invention improves the search efficiency of the subsequent TCN-BiGRU-CA model. The algorithm's search mechanism is naturally suitable for parallel computing, which can further improve the solution efficiency of large-scale problems. Moreover, it can effectively solve the problem of excessive differences between simulation data and measured data caused by parameter simplification in traditional dynamic models, and significantly improve the accuracy and reliability of fault diagnosis of aero-engine main bearings. Attached Figure Description

[0035] Figure 1 This is a flowchart of the rolling bearing fault diagnosis method in the embodiment. Detailed Implementation

[0036] The present invention will now be described in further detail with reference to the embodiments and accompanying drawings. However, this should not be construed as limiting the scope of the above-described subject matter of the present invention to the following embodiments; all technologies implemented based on the content of the present invention fall within the scope of the present invention.

[0037] Example

[0038] See Figure 1 A rolling bearing fault diagnosis method based on mechanism and data fusion includes:

[0039] Step 1: Establish model equations based on Hertz contact theory and construct a bearing dynamics model;

[0040] In this embodiment, to ensure the effectiveness of the bearing dynamics model and simplify the calculation and analysis process, the following four key idealization assumptions are made for the bearing system:

[0041] ① The nonlinearity of the model is mainly defined as the nonlinear contact force between the components, the time-varying stiffness, and the internal clearance of the bearing. The elastohydrodynamic lubrication effect is not considered for the time being.

[0042] ②The elastic contact deformation process between the rolling element and the inner and outer raceways is considered to completely follow Hertz contact theory.

[0043] ③ The mass and moment of inertia of the rolling elements and the cage are ignored, and all rolling elements are set to maintain a constant angular distance in the circumferential direction.

[0044] ④ All translational motions within the system are constrained within a two-dimensional Cartesian coordinate system (xy plane), and all rotational motions are performed around the z-axis, which is perpendicular to this plane.

[0045] Based on the above assumptions, the bearing dynamics model consists of four core components: the inner ring, the outer ring, the rolling elements, and the cage. It primarily captures the dynamic behavior of the inner and outer rings in the two orthogonal x and y directions. Taking into account the periodic excitation force generated by the eccentric mass due to manufacturing and assembly errors, a four-degree-of-freedom dynamic differential equation system describing the motion of the inner and outer rings in the xy plane can be constructed, forming the bearing dynamics model:

[0046]

[0047] in For the mass of the bearing inner ring, This represents the damping coefficient of the bearing inner ring. This refers to the stiffness of the bearing inner ring; For the mass of the bearing outer ring, For the outer ring stiffness of the bearing, This refers to the damping coefficient of the bearing outer ring; Indicates the eccentricity; Let be the instantaneous displacement of the inner circle center in the x-direction. Let be the instantaneous displacement of the outer ring center in the x-direction. Let be the instantaneous displacement of the inner circle center in the y-direction. This represents the instantaneous displacement of the outer ring center in the y-direction; Let x be the component of the total supporting reaction force of the bearing on the inner ring in the x-direction. The rotational speed of the spindle that mates with the bearing. The rotation time relative to the initial position of the spindle. It is the acceleration due to gravity. This represents the component of the total supporting reaction force of the bearing on the inner ring in the y direction.

[0048] Step 2: Use the bearing dynamics model to simulate and generate healthy bearing state data;

[0049] This embodiment uses a fourth-order Runge-Kutta numerical algorithm to solve the bearing dynamics model and generate the theoretical vibration signal of the bearing in a healthy state, which is the healthy bearing state data.

[0050] Step 3: Introduce a geometrically regular rectangular spalling as a single-point damage fault of the raceway into the bearing dynamics model, and construct a contact deformation analysis model after introducing the damage fault.

[0051] In this embodiment, to simulate local bearing faults, a single-point damage to the raceway is introduced into the bearing dynamics model. This damage is simplified as a geometrically regular, tiny rectangular spalling. The impact of the fault on the system's dynamic behavior is reflected by modifying the calculation method for contact deformation. The contact deformation analysis model after introducing the fault is as follows: ,in To introduce the aforementioned damage fault, the first Contact deformation of each rolling element For the first A rolling element in The instantaneous angular position at a given moment. , It is the total number of rolling elements. It is the initial phase angle of the bearing. Represents the orbital angular velocity of the cage; This refers to the radial internal clearance of the bearing. This represents the equivalent deformation release caused by a local fault. , The fixed center corner position of the rectangular peeling area when the rectangular peeling is located on the outer ring. This represents the circumferential width of the rectangular peeling area when the rectangular peeling is located on the outer ring. The outer radius is... The fixed center corner position of the rectangular peeling area when the rectangular peeling is located in the inner circle. This represents the circumferential width of the rectangular peeling area when the rectangular peeling is located in the inner circle. The inner radius; ,in Where is the radius of the rolling element. The radial depth of the rectangular peeling.

[0052] Step 4: Use the contact deformation analysis model to simulate and obtain bearing failure status data. The bearing failure status data includes inner ring failure status data when rectangular spalling is located on the inner ring and outer ring failure status data when rectangular spalling is located on the outer ring.

[0053] Step 5: Using the difference between bearing fault state data and bearing health state data as the optimization function, the optimization objective is to minimize the output value of the optimization function. The damping coefficient and stiffness of the bearing inner ring and the bearing outer ring are used as optimization variables. The HFOA algorithm is used to correct the bearing dynamics model, and the optimized damping coefficient and stiffness of the bearing inner ring and the bearing outer ring are output, thus obtaining the corrected bearing dynamics model.

[0054] In this embodiment, the method of using the improved HFOA algorithm to correct the bearing dynamics model and output the optimized damping coefficient and stiffness of the bearing inner ring and the bearing outer ring includes:

[0055] S5.1 Within the search space formed by the given upper and lower limits of each optimization variable, a population generation strategy based on a fourth-order Chebyshev chaotic map is used to randomly generate an initial population formed by the initial position decision values ​​of each optimization variable; the specific operation is as follows:

[0056] 5.1.1 Based on the search space range formed by the given upper and lower limits of each optimization variable, the following method is adopted: Randomly generate decision values ​​based on the initial positions of each optimization variable. ,in For the first The first individual The upper limit of each optimization variable within the search space. For the first The first individual The lower bound of each optimization variable within the search space. The initial chaotic variable is a random decimal between 0 and 1, representing the population generation strategy of the fourth-order Chebyshev chaotic map.

[0057] 5.1.2 The decision values ​​for the positions of each optimization variable are updated, and after reaching a preset number of iterations or convergence, information on the position of each optimization variable for each individual is generated. The initial population, It is the first The first individual Initial positions of the optimization variables The population generation strategy for the fourth-order Chebyshev chaotic map is the first... The chaotic variable is updated after the next iteration.

[0058] In this embodiment, the initial dynamic characteristics of chaotic mapping, the range of the search space, and the influence of Chebyshev order are comprehensively considered to make the population coverage more uniform, thereby improving the global search capability and convergence speed.

[0059] S5.2 Based on the initial preset step size vector and direction vector of each individual in the initial population, perform a preliminary update on the position of each individual in the corresponding search space to obtain the updated individual position. ,in It is the first The first individual The positions of the optimization variables after initial update It is the first The first individual Initial positions of the optimization variables For the first The first individual The initial preset step size vector of each optimization variable, For the first The first individual The initial preset direction vector of each optimization variable.

[0060] S5.3 After the initial update of individual positions, the entire population is divided into several subpopulations. Within each subpopulation, the optimal individual in the current subpopulation is identified as the subpopulation leader based on the fitness value of each individual. Within each subpopulation, individuals learn from the leader of that subpopulation and adjust their own positions to obtain the learned individual's position in the subpopulation. ,in, To control the subgroup learning coefficient of learning intensity, The leader of the current subgroup The position of each optimization variable The first outside the leader of the subgroup The first individual The position of each optimization variable.

[0061] S5.4 Iteratively learns and updates the position of the learned individual within the subpopulation. During the iterative learning and update process, based on the upper and lower bounds of the search space for each individual in the population, a lens imaging back-learning strategy is used to generate the corresponding individual's position within the subpopulation. ,in For the first The first individual The upper limit of each optimization variable within the search space. For the first The first individual The lower bound of each optimization variable within the search space. For adaptive scaling factor, , This is the initial value for the adaptive scaling factor. This represents the current iteration number. This is the preset total number of iterations; it allows you to calculate the position range of all individuals within the subgroup during the learning process. As a preset termination condition for stopping iteration, a lens imaging inverse learning strategy is introduced to inversely perturb and select individual positions, thereby enhancing the diversity of the local development stage and improving the ability to escape local extrema. This allows the inverse solution to exhibit elastic contraction in the search space, thus adaptively adjusting with iteration and continuously improving convergence accuracy.

[0062] In each iteration of S5.4, S5.5 assigns a first visual range to the leader role within the subgroup, and assigns a second visual range to the other individuals within the subgroup as developers; the first visual range is larger than the second visual range; the average fitness of the entire subgroup after learning and adjusting positions is calculated. If the average fitness is greater than a preset fitness threshold, a role-switching mechanism for some developers is triggered. Developers who have switched roles search for the neighborhood optimum with the lowest fitness within their corresponding visual range and move towards it, updating the positions of the developers who have switched roles. This process continues until the average fitness of the subgroup is less than or equal to the preset fitness threshold, at which point the individual with the lowest fitness within the subgroup is determined as the local optimum; the updated positions of the developers who have switched roles are then updated. ,in, The developer of role transformation In the updated location, This is the developer's position before the update, where the role was changed. It is the step size factor that controls the movement distance. It is the location of the neighborhood optimal solution within the developer's visual range during the role transformation.

[0063] S5.6 determines the local optimum of the subpopulation with the lowest average fitness among all subpopulations in the current population as the global optimum. Based on the global optimum of the current population and the local optima of each subpopulation, the step size vector and direction vector of each individual are dynamically adjusted; the dynamically adjusted step size vector... The dynamically adjusted direction vector ,in For dynamic adjustment of the first The first individual The step size vector of each optimization variable. The global learning coefficient. Let be a function that returns a random number in the interval [0, 1]. The globally optimal solution found in the current iteration is at the [number]th iteration. The position of each optimization variable For dynamic adjustment of the first The first individual The direction vector of each optimization variable, For local learning coefficients, The local optimum found in the current iteration number is the one at the [number]th iteration. The position of each optimization variable; repeat steps S5.3 to S5.6 until the preset maximum number of iterations is reached or the preset termination condition is met, and output the global optimal solution that is the smallest in the entire iteration process and satisfies the output value of the optimization function.

[0064] In this embodiment, the Hawk-Fish Optimization Algorithm (HFOA) is a novel metaheuristic optimization algorithm inspired by the hunting strategies of eagles and the group behavior of fish in nature. The core idea is to combine the "precise positioning-surprise attack" characteristics of eagles with the "group cooperation-information sharing" advantages of fish groups to balance the algorithm's global exploration capability and local development capability.

[0065] Step 6: Use the modified bearing dynamics model to generate bearing condition simulation data including fault size, spindle speed, load, and fault type;

[0066] Step 7: Establish a TCN-BiGRU model with embedded channel attention mechanism. Use the generated bearing state simulation data as the input of the TCN-BiGRU model, and the corresponding fault size label and fault type label as the output of the TCN-BiGRU model. Train the TCN-BiGRU model to obtain the trained TCN-BiGRU-CA model.

[0067] Step 8: Analyze the measured condition data of the bearing to be analyzed using the TCN-BiGRU-CA model, and output the fault diagnosis results.

[0068] In this embodiment, in the TCN-BiGRU-CA model, the output data of the modified bearing dynamics model first enters TCN for feature extraction, and then is processed by BiGRU to output the fault diagnosis result.

[0069] In this embodiment, a bearing dynamics model including the inner ring, rolling elements, outer ring, and cage is constructed. Key influencing factors such as bearing fault size, operating speed, and load are comprehensively considered to accurately capture the motion states of the inner and outer rings, generating bearing fault state data. The difference between the bearing fault state data and the bearing health state data is used as the optimization function, with the minimum output value of the optimization function as the optimization objective. An improved Hawkfish Optimization Algorithm (HFOA) is employed to optimize the damping coefficient and stiffness of the bearing inner and outer rings as optimization variables. Finally, the TCN-BiGRU-CA model is used to analyze the output data of the corrected bearing dynamics model, outputting fault diagnosis results. This invention, while correcting the bearing dynamics model and ensuring the rapid convergence and strong robustness of the Hawkfish Optimization Algorithm, improves the search efficiency of the subsequent TCN-BiGRU-CA model. The algorithm's search mechanism is naturally suitable for parallel computing, further improving the solution efficiency for large-scale problems. Furthermore, it effectively solves the problem of excessive differences between simulation and measured data caused by parameter simplification in traditional dynamic models, significantly improving the accuracy and reliability of fault diagnosis for aero-engine main bearings.

[0070] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for rolling bearing fault diagnosis based on mechanism and data fusion, characterized in that, The method comprises the following steps: Step 1, a model equation is established according to Hertz contact theory to construct a bearing dynamics model; Step 2, bearing dynamics model simulation is used to generate healthy bearing state data; Step 3, a geometrically regular rectangular spalling is introduced as a single-point damage fault of the raceway in the bearing dynamics model, and a contact deformation analysis model after the introduction of the damage fault is constructed; Step 4, the bearing fault state data is obtained by simulation using the contact deformation analysis model, which includes the inner ring fault state data when the rectangular spalling is located in the inner ring and the outer ring fault state data when the rectangular spalling is located in the outer ring; Step 5: the difference between the bearing fault state data and the bearing healthy state data is taken as an optimization function, the minimum output value of the optimization function is taken as the optimization target, the damping coefficient and the stiffness of the bearing inner ring and the damping coefficient and the stiffness of the bearing outer ring are taken as optimization variables, the HFOA algorithm is used to correct the bearing dynamics model, and the optimized damping coefficient and stiffness of the bearing inner ring and the optimized damping coefficient and stiffness of the bearing outer ring are output, and a corrected bearing dynamics model is obtained; Step 6: the bearing state simulation data containing the fault size, the spindle speed, the load and the fault type are generated by using the corrected bearing dynamics model; Step 7: a TCN-BiGRU model embedded with a channel attention mechanism is established, the generated bearing state simulation data is taken as the input of the TCN-BiGRU model, the corresponding fault size label and fault type label are taken as the output of the TCN-BiGRU model, the TCN-BiGRU model is trained, and a trained TCN-BiGRU-CA model is obtained; Step 8: the TCN-BiGRU-CA model is used to analyze the measured state data of the bearing to be analyzed, and a fault diagnosis result is output.

2. The rolling bearing fault diagnosis method according to claim 1, characterized in that, In the TCN-BiGRU-CA model, the bearing measured state data first enters the TCN for feature extraction, and then is calculated and processed by the BiGRU to output the fault diagnosis result.

3. The rolling bearing fault diagnosis method according to claim 1, characterized in that, All translational movements of the bearing are constrained in the x-y plane of the two-dimensional Cartesian coordinate system, and all rotational movements of the bearing are around the z-axis perpendicular to the x-y plane; the bearing dynamics model constructed is: wherein is a mass of the inner ring of the bearing, is a damping coefficient of the inner ring of the bearing, is a stiffness of the inner ring of the bearing; is a mass of the outer ring of the bearing, is a stiffness of the outer ring of the bearing, is a damping coefficient of the outer ring of the bearing; denotes an eccentricity; is an instantaneous displacement of the center of the inner ring in the x direction, is an instantaneous displacement of the center of the outer ring in the x direction, is an instantaneous displacement of the center of the inner ring in the y direction, is an instantaneous displacement of the center of the outer ring in the y direction; is a component of the total support reaction of the bearing on the inner ring in the x direction, is a rotational speed of a main shaft cooperating with the bearing, is a rotation time relative to an initial position of the main shaft, is a gravitational acceleration, is a component of the total support reaction of the bearing on the inner ring in the y direction.

4. The rolling bearing fault diagnosis method according to claim 3, characterized in that, The constructed contact deformation amount analysis model is wherein is the contact deformation amount of the th rolling element after the damage failure is introduced, is the instantaneous angular position of the th rolling element at the time , , is the total number of rolling elements, is the initial phase angle of the bearing, represents the orbital angular velocity of the cage; is the radial internal clearance of the bearing, is the equivalent deformation release amount caused by the local failure, , is the fixed central angular position of the rectangular flaking area when the rectangular flaking is located at the outer ring, is the circumferential width of the rectangular flaking area when the rectangular flaking is located at the outer ring, is the outer ring radius, is the fixed central angular position of the rectangular flaking area when the rectangular flaking is located at the inner ring, is the circumferential width of the rectangular flaking area when the rectangular flaking is located at the inner ring, is the inner ring radius; wherein is the rolling element radius, is the radial depth of the rectangular flaking.

5. The rolling bearing fault diagnosis method according to claim 4, characterized in that, In step 5, the HFOA algorithm is used to correct the bearing dynamics model, and the optimized damping coefficient and stiffness of the bearing inner ring and the optimized damping coefficient and stiffness of the bearing outer ring are output, which comprises the following steps: S5.1, a four-order Chebyshev chaos mapping population generation strategy is adopted to randomly generate an initial population formed by initial position decision values of each optimization variable in a search space range formed by given upper and lower limit values of each optimization variable; S5.2 Based on the initial preset step size vector and direction vector of each individual in the initial population, perform a preliminary update on the position of each individual in the corresponding search space to obtain the updated individual position. ,in It is the first The first individual The positions of the optimization variables after initial update It is the first The first individual Initial positions of the optimization variables For the first The first individual The initial preset step size vector of each optimization variable, For the first The first individual The initial preset direction vector of each optimization variable; S5.3 After the initial update of individual positions, the entire population is divided into several subpopulations. Within each subpopulation, the optimal individual in the current subpopulation is identified as the subpopulation leader based on the fitness value of each individual. Within each subpopulation, individuals learn from the leader of that subpopulation and adjust their own positions to obtain the learned individual's position in the subpopulation. ,in, To control the subgroup learning coefficient of learning intensity, The leader of the current subgroup The position of each optimization variable The first outside the leader of the subgroup The first individual The position of each optimization variable; S5.4, the position of the learned individual in the subpopulation is iteratively learned and updated, and during the iterative learning and updating process, the position of the corresponding individual in the subpopulation is generated by using the lens imaging reverse learning strategy according to the upper and lower limit values of each individual in the population in the search space. S5.5 In each iteration of S5.4, assign a first visual range to the leader role within the sub-population, and assign a second visual range to the other individuals as developers within the sub-population; the first visual range is greater than the second visual range; calculate the average fitness of the entire sub-population after learning and adjusting the positions, if the average fitness is greater than a preset fitness threshold, trigger the role conversion mechanism of part of the developers, the role-converted developers search for the neighborhood optimal solution with the minimum fitness within the corresponding visual range and move to it, update the positions of the role-converted developers until the average fitness of the sub-population is less than or equal to the preset fitness threshold, and determine the individual with the minimum fitness within the sub-population as the local optimal solution; S5.6 Determine the local optimal solution of the sub-population with the minimum average fitness in the current entire population as the global optimal solution, and dynamically adjust the step vector and direction vector of each individual according to the global optimal solution of the current entire population and the local optimal solution of each sub-population; repeat steps S5.3 to S5.6 until a preset maximum number of iterations is reached or a preset termination condition is met, and output the global optimal solution recorded in the entire iteration process and satisfying the minimum output value of the optimization function.

6. The rolling bearing fault diagnosis method according to claim 5, characterized in that, The population generation strategy of the fourth-order Chebyshev chaotic mapping in step S5.1 includes the following steps: Based on the search space formed by the given upper and lower limits of each optimization variable, the following approach is adopted: Randomly generate decision values ​​based on the initial positions of each optimization variable. ,in For the first The first individual The upper limit of each optimization variable within the search space. For the first The first individual The lower bound of each optimization variable within the search space. The initial chaotic variable is a random decimal between 0 and 1, representing the population generation strategy of the fourth-order Chebyshev chaotic map. With updating each optimization variable position decision value, forming each individual's each optimization variable position information after reaching a preset iteration number or convergence initial population, is the initial position of the th optimization variable of the th individual, is the updated value of the chaotic variable after the th iteration of the population generation strategy of the fourth-order Chebyshev chaotic mapping.

7. The rolling bearing fault diagnosis method according to claim 5, characterized in that, The position of the corresponding individual within the subpopulation is generated using a lens imaging back-learning strategy. ,in For the first The first individual The upper limit of each optimization variable within the search space. For the first The first individual The lower bound of each optimization variable within the search space. For adaptive scaling factor, , This is the initial value for the adaptive scaling factor. This represents the current iteration number. This represents the preset total number of iterations.

8. The rolling bearing fault diagnosis method according to claim 5, characterized in that, In step S5.5, the updated position of the role transition developer where, is the role transition developer at the updated position, is the position of the role transition developer before the update, is a step factor that controls the movement distance, is the position of the neighborhood optimum solution within the visual range of the role transition developer.

9. The rolling bearing fault diagnosis method according to claim 7, characterized in that, In step S5.6, the dynamically adjusted step size vector The dynamically adjusted direction vector ,in For dynamic adjustment of the first The first individual The step size vector of each optimization variable. The global learning coefficient. Let be a function that returns a random number in the interval [0, 1]. The globally optimal solution found in the current iteration is at the [number]th iteration. The position of each optimization variable For dynamic adjustment of the first The first individual The direction vector of each optimization variable, For local learning coefficients, The local optimum found in the current iteration number is the one at the [number]th iteration. The position of each optimization variable.

Images