A method and system for monitoring stress during press-fitting of a generator rotor bearing

Abstract

The application provides a kind of based on generator rotor bearing press mounting process stress monitoring method and system, by obtaining mechanical force data, mating surface normal pressure data, tangent direction friction data, three-dimensional space strain data data and stress influence data set, and by friction coefficient hybrid identification model, output real-time friction coefficient value and obtain total friction, by stress influence data set generation friction loss correction term and combine mechanical force data, total friction obtains net pressure force data, obtains temperature data and carries out dynamic calibration, improves the accuracy of real-time friction coefficient, obtains more accurate net pressure force data, and obtains plastic deformation factor and force distribution uniformity index, the stress of generator rotor bearing press mounting process is monitored in real time, through partial load early warning and dynamic control, obtains control data, and through control data inversion real-time data is used in real-time production process, improve the efficiency and accuracy in press mounting process.

Description

Technical Field

[0001] This invention relates to the field of force detection technology, and more specifically, to a method and system for monitoring the force during the press-fitting process of generator rotor bearings. Background Technology

[0002] As industrial equipment develops towards higher precision and higher reliability, the quality control of the generator rotor bearing press-fitting process has an increasingly significant impact on the overall performance of the machine. Accurate monitoring of the stress state during the press-fitting process has become a core technical challenge to ensure the safe operation of the equipment.

[0003] Traditional press-fit monitoring methods have significant limitations. For example, traditional methods often estimate the friction coefficient using empirical formulas, neglecting the dynamic effects of temperature and speed. During press-fitting at high temperatures, the actual friction coefficient increases, leading to excessively high deviations in the total friction force calculation and failing to reflect the true friction state. Other methods rely solely on a few strain gauges to monitor local strain, often failing to detect uneven stress distribution along the bearing's outer ring. For instance, if a region exhibits a plastic deformation factor exceeding a threshold during press-fitting, insufficient monitoring points may prevent timely detection, resulting in localized plastic deformation of the bearing. These limitations delay the identification of potential problems such as uneven loading and localized stress concentration, easily leading to premature bearing failure, abnormal rotor vibration, and other malfunctions.

[0004] Therefore, it is urgent to develop a multi-dimensional sensor fusion mechanism and stress-strain analysis model, and to construct a press-fitting stress monitoring framework with force-thermal data analysis and controllability. By integrating sensors into the press-fitting fixture and combining sampling calibration and model inversion for real-time production, a high-efficiency, safe, and high-quality generator rotor bearing press-fitting process can be achieved. Summary of the Invention

[0005] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for monitoring the force during the press-fitting process of a generator rotor bearing, the method comprising:

[0006] The sensor array is used to acquire real-time data on mechanically applied forces, mating surface normal pressure, tangential friction, three-dimensional spatial strain, and force influence.

[0007] Real-time friction coefficient data is output based on the friction coefficient hybrid identification model, and total friction force data is obtained through the friction coefficient data.

[0008] A friction loss correction term is generated based on the force influence dataset, and net pressure force data is obtained based on mechanical applied force data, total friction force data, and friction loss correction term.

[0009] Based on the continuum mechanics framework, three-dimensional spatial strain data is input into the deformation analysis model, and the plastic deformation factor and force distribution uniformity index are output.

[0010] During the press-fitting process, the stress on the generator rotor bearing is monitored based on net press-fitting force data, plastic deformation factor, and force distribution uniformity index.

[0011] The off-center load is determined based on the force distribution uniformity index and the plastic deformation factor. The off-center load result is then used to adjust the press-fitting machinery in real time.

[0012] As a further aspect of the present invention, the mechanically applied force data, mating surface normal pressure data, tangential friction force data, three-dimensional spatial strain data, and force influence dataset are acquired in real time through a monitoring sensor group, including:

[0013] The monitoring sensor group includes multiple strain gauges, multiple pressure sensors, multiple strain-type sensors, multiple acceleration sensors, displacement sensors, and a temperature compensation module. The force influence dataset includes relative motion velocity data of the contact area, temperature data, contact area data, and contact deformation data.

[0014] Multiple sets of strain gauges are deployed on the pressure head, bearing outer ring, rotor journal and press-fit base respectively. The multiple sets of strain gauges are strain rose structures. Three-dimensional spatial strain data are obtained based on the multiple sets of strain gauges.

[0015] Multiple pressure sensors are symmetrically arranged on the outer ring of the bearing. Based on the multiple pressure sensors, the normal pressure data of the mating surface is obtained, and the contact area data and relative motion speed data of the contact area are obtained simultaneously.

[0016] Multiple strain gauge sensors are installed at key parts of the rotor journal, and tangential friction force data and temperature data are obtained based on the strain gauge sensors and temperature compensation module, respectively.

[0017] A set of acceleration sensors is deployed at each of the four corners of the press-fit base to obtain mechanical force data based on the acceleration sensors;

[0018] A displacement sensor is installed at the contact point between the pressure head and the bearing to obtain contact deformation data.

[0019] As a further aspect of the present invention, real-time friction coefficient data is output based on a friction coefficient hybrid identification model, and total friction force data is obtained through the friction coefficient data, including:

[0020] Preprocessing operations are performed on the normal pressure data of the mating surface, the frictional force data in the tangential direction, the relative motion velocity data of the contact area in the force influence data, the temperature data, and the contact deformation data.

[0021] The normal pressure data of the mating surface, the friction force data in the tangential direction, the relative motion velocity data of the contact area, the temperature data, and the contact deformation data are input into the friction coefficient hybrid identification model;

[0022] The friction feature vector is output by the front-end neural network module and input to the back-end physical constraint module to obtain the initial friction coefficient data.

[0023] The initial friction coefficient data is dynamically calibrated using temperature data to obtain real-time friction coefficient data;

[0024] The friction force data of each contact point is obtained by multiplying the real-time friction coefficient data with the corresponding normal pressure data of the mating surface, and the total friction force data is obtained based on the friction force data of each contact point.

[0025] As a further aspect of the present invention, a friction coefficient hybrid identification model is constructed, including:

[0026] Collect historical pressing data under different working conditions, use the historical pressing data as a sample dataset, and standardize the sample dataset.

[0027] The historical press-fitting data includes historical mating surface normal pressure data, historical tangential friction force data, historical contact area relative motion speed data, historical temperature data, historical contact deformation data, and corresponding historical friction coefficient data.

[0028] The input layer is constructed based on historical mating surface normal pressure data, historical tangential friction force data, historical contact area relative motion velocity data, and historical temperature data. The hidden layer is constructed based on the modified linear unit function. The output layer is constructed based on historical friction feature vectors. The front-end neural network module is constructed based on the input layer, hidden layer, and output layer.

[0029] A corrected equation for the frictional heat effect is constructed, and a backend physical constraint module is built based on the corrected equation for the frictional heat effect.

[0030] An initial friction coefficient hybrid identification model is constructed based on a front-end neural network module and a back-end physical constraint module. The initial friction coefficient hybrid identification model is trained based on a sample dataset, with mean square error as the loss function and updated based on the Adam optimizer until the model converges, and the converged friction coefficient hybrid identification model is output.

[0031] The performance of the convergent friction coefficient hybrid identification model is evaluated, and the friction coefficient hybrid identification model is output.

[0032] As a further aspect of the present invention, a friction loss correction term is generated based on the force influence dataset, and net indentation force data is obtained based on the mechanically applied force data, total friction force data, and the friction loss correction term, including:

[0033] Based on wavelet denoising, the relative motion velocity data of the contact area in the force influence dataset is processed to obtain the uniform velocity component and the instantaneous fluctuation component.

[0034] The cumulative value of frictional work is obtained based on the uniform velocity component, and a thermo-mechanical coupling correction coefficient is established by combining the temperature data in the force influence dataset.

[0035] Perform Fourier transform on the instantaneous fluctuation components to extract their frequency characteristics and amplitude distribution, and generate a dynamic impact correction factor.

[0036] The comprehensive correction coefficient for friction loss is obtained by fusing the thermo-mechanical coupling correction coefficient and the dynamic impact correction factor.

[0037] The friction loss correction term is obtained based on the comprehensive correction coefficient for friction loss and the contact area data in the stress influence dataset.

[0038] The net indentation force equation is constructed based on the mechanically applied force data, total friction force data, and friction loss correction term, and the net indentation force data is obtained by solving it.

[0039] As a further aspect of the present invention, it is characterized by constructing a net indentation force equation based on mechanically applied force data, total friction force data, and friction loss correction terms, and solving the equation to obtain the net indentation force data, including:

[0040] The net pressure equation is specifically expressed as follows:

[0041] ;

[0042] in, Represented as net pressure data, Represented as mechanically applied force data, This is expressed as total frictional force data. Represented as real-time friction coefficient data, This is represented as the normal force data for the mating surfaces. This is represented as a correction term for friction loss.

[0043] As a further aspect of the present invention, based on the framework of continuum mechanics, three-dimensional spatial strain data is input into a deformation analysis model, and the plastic deformation factor and force distribution uniformity index are output, including:

[0044] A deformation analysis model was constructed, and a three-dimensional mesh was generated based on 8-node hexahedral elements, while material property data was obtained.

[0045] The three-dimensional spatial strain data is mapped to the corresponding nodes of the deformation analysis model, and the three-dimensional spatial strain data is converted into strain components.

[0046] Convert strain components into stress components;

[0047] The residual data of stress and strain components are obtained by minimizing the least squares method, and the stress tensor is obtained based on the residual data.

[0048] The stress tensor is used to output the plastic deformation factor and the force distribution uniformity index.

[0049] As a further aspect of the present invention, the plastic deformation factor and force distribution uniformity index are output based on the stress tensor, including:

[0050] Extract the three principal stresses from the stress tensor and obtain the maximum and minimum principal stresses;

[0051] The maximum shear stress difference is obtained by processing the maximum principal stress and minimum principal stress based on the Tresca yield criterion.

[0052] The plastic deformation factor is obtained based on the maximum shear stress difference and the yield strength in the material property data;

[0053] The equivalent stress of each node is obtained based on the three principal stresses, and the equivalent stress of discrete nodes is obtained based on the spatial interpolation algorithm. The maximum equivalent stress value and the average equivalent stress value of the contact area are extracted simultaneously. The average equivalent stress value is expressed as the average value of the equivalent stress of all nodes in the contact area.

[0054] The uniformity index of force distribution is obtained based on the maximum equivalent stress value and the average equivalent stress value.

[0055] As a further aspect of the present invention, off-center loading is determined based on the force distribution uniformity index and the plastic deformation factor to obtain the off-center loading result. Real-time control of the press-fitting machinery is then performed based on the off-center loading result, including:

[0056] Determining the eccentric load level based on a combination of force distribution uniformity index and plastic deformation factor:

[0057] When the force distribution uniformity index is greater than or equal to 0.6 and less than 0.7, and the plastic deformation factor is less than 0.9, it is judged as a slight off-center load and an audible and visual alarm is triggered.

[0058] When the force distribution uniformity index is greater than or equal to 0.5 and less than 0.6, or the plastic deformation factor is greater than or equal to 0.9 and less than 1.0, it is judged as moderate off-center load, and the pressing speed is reduced.

[0059] When the force distribution uniformity index is less than 0.5 or the plastic deformation factor is greater than or equal to 1.0, it is judged as severe off-center loading, and the press-fitting machinery should be stopped immediately.

[0060] Based on the stress gradient direction of the deformation analysis model, the off-center load vector is obtained, and compensation commands are generated to control the press-fitting machinery in real time.

[0061] Real-time monitoring of stress distribution changes after compensation; when the force distribution uniformity index is greater than or equal to 0.7 and the plastic deformation factor is less than 0.9, the warning is lifted and normal press-fitting parameters are restored.

[0062] Furthermore, embodiments of the present invention also provide a stress monitoring system based on the press-fitting process of generator rotor bearings, comprising:

[0063] The acquisition module acquires mechanically applied force data, mating surface normal pressure data, tangential friction force data, three-dimensional spatial strain data, and force influence dataset in real time through a monitoring sensor group, and acquires total friction force data through friction coefficient data;

[0064] The input module, based on the continuum mechanics framework, inputs three-dimensional spatial strain data into the deformation analysis model;

[0065] The output module is used to output the plastic deformation factor and the force distribution uniformity index, and to output real-time friction coefficient data based on the friction coefficient hybrid identification model.

[0066] The processing module generates a friction loss correction term based on the force influence dataset, and obtains net pressure data based on mechanically applied force data, total friction force data, and friction loss correction term.

[0067] The detection module monitors the stress on the generator rotor bearing during the press-fitting process based on net press-in force data, plastic deformation factor, and force distribution uniformity index.

[0068] The control module determines the off-center load based on the force distribution uniformity index and the plastic deformation factor, obtains the off-center load result, and performs real-time control of the press-fitting machinery based on the off-center load result.

[0069] Based on the above, the embodiments of this application achieve real-time acquisition of mechanically applied force data, mating surface normal pressure data, tangential friction force data, three-dimensional spatial strain data, and force influence datasets. This overcomes the limitations of traditional single sensors or a few monitoring points that cannot reflect the overall force state, ensuring data integrity. The mating surface normal pressure data and tangential friction force data are input into a pre-constructed friction coefficient hybrid identification model, outputting real-time friction coefficient values. The total friction force is obtained based on the real-time friction coefficient values. A friction loss correction term is generated based on the force influence dataset. Finally, the net friction force is obtained based on the mechanically applied force data, the total friction force, and the friction loss correction term. By using indentation force data, a hybrid identification model of friction coefficient is constructed to extract friction characteristics and physical constraints are applied using Hertzian contact theory. Simultaneously, temperature data is acquired for dynamic calibration, improving the accuracy of the real-time friction coefficient and thus obtaining more precise net indentation force data. Through the continuum mechanics framework and deformation analysis model, plastic deformation factor and force distribution uniformity index are obtained. Combined with the net indentation force, the stress on the generator rotor bearing during the press-fitting process is monitored in real time. Through off-center load early warning and dynamic control, control data is acquired, and real-time data is inverted from the control data and applied to the real-time production process, improving the efficiency and accuracy of the generator rotor bearing press-fitting process. Attached Figure Description

[0070] Figure 1 This is a schematic diagram of the execution flow of a method for monitoring the force during the press-fitting process of a generator rotor bearing, provided by an embodiment of the present invention.

[0071] Figure 2 This is a schematic diagram of a force monitoring system based on the press-fitting process of generator rotor bearings provided in an embodiment of the present invention. Detailed Implementation

[0072] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a schematic diagram of the execution flow of a method for monitoring the force during the press-fitting of generator rotor bearings, provided by an embodiment of the present invention. The following is a detailed description of this method for monitoring the force during the press-fitting of generator rotor bearings.

[0073] Specifically, a method for monitoring the force during the press-fitting process of generator rotor bearings includes:

[0074] Step S1 involves acquiring real-time data on mechanically applied force, mating surface normal pressure, tangential friction force, three-dimensional spatial strain, and force influence through a monitoring sensor array.

[0075] Specifically, the monitoring sensor group includes multiple strain gauges, multiple pressure sensors, multiple strain gauge sensors, multiple accelerometers, displacement sensors, and a temperature compensation module. The force influence dataset includes relative motion velocity data of the contact area, temperature data, contact area data, and contact deformation data.

[0076] In this embodiment, step S1 includes:

[0077] Step S11: Deploy multiple sets of strain gauges onto the pressure head, bearing outer ring, rotor journal, and press-fit base respectively, and obtain three-dimensional spatial strain data based on the multiple sets of strain gauges.

[0078] In some possible embodiments, 18 sets of strain gauges can be installed on the contact surfaces of the press head and bearing of the press fitting equipment (4 sets), the outer circumference of the bearing (6 sets), the contact area of ​​the rotor journal (6 sets), and the contact area of ​​the press fitting base (2 sets). Each set includes 3 strain gauges. Each set of strain gauges is connected by a strain rosette structure and a full-bridge circuit. The strain rosette structure is represented by 3 strain gauges at 45° to each other. The full-bridge circuit is used to control the direction of DC or to achieve reversible voltage output. The original strain signal is acquired by sampling at a frequency greater than 2kHz. The influence of temperature on the resistance strain gauges is eliminated by temperature drift compensation. High-frequency noise is filtered out by a low-pass filter with a cutoff frequency of 100Hz to obtain the normal strain and shear strain in three-dimensional space.

[0079] Step S12: Multiple pressure sensors are symmetrically arranged on the outer ring of the bearing. Based on the multiple pressure sensors, the positive pressure data of the mating surface is obtained, and the contact area data of the mating surface and the relative motion speed data of the contact area are obtained simultaneously.

[0080] In some possible embodiments, multiple pressure sensors employ thin-film pressure sensors. For example, six thin-film pressure sensors can be symmetrically arranged on the inner contact surface of the bearing outer ring. The thickness of the thin-film pressure sensors is less than 0.1 mm, so they will not affect the relationship between the bearing and the journal. The voltage signal output by the thin-film pressure sensor is proportional to the applied pressure and is converted into normal pressure data of the mating surface through a signal conditioning circuit. At the same time, the real-time contact area is obtained by measuring the effective area and pressure distribution range of the thin-film pressure sensor. Simultaneously, a laser displacement sensor is installed on the pressure head to obtain the relative motion speed of the contact area. The dynamic friction state on the contact surface is reflected by obtaining the relative motion speed of the contact area.

[0081] Step S13: Install multiple strain gauge sensors at key locations on the rotor journal, and acquire tangential friction force data and temperature data based on the strain gauge sensors and temperature compensation module.

[0082] In some possible embodiments, five sets of strain gauge sensors can be installed at the contact point between the rotor journal and the bearing. The resistance change generated by the strain gauge sensors is converted into a voltage signal, which is then amplified and filtered to obtain tangential friction force data. This tangential friction force data reflects the magnitude of the friction force on the contact surface that hinders relative motion. Simultaneously, a temperature compensation module acquires ambient temperature data. This temperature compensation module can be installed at the contact point between the rotor journal and the bearing and is composed of temperature sensors.

[0083] Step S14: Deploy a set of acceleration sensors at each of the four corners of the press-fit base, and obtain mechanical force data based on the acceleration sensors.

[0084] In some possible embodiments, four sets of acceleration sensors can be deployed at the four corners of the pressing base of the pressing equipment. The acceleration sensors are piezoelectric acceleration sensors, which can convert mechanical vibration into charge signals, and the charge amplifier converts the charge signals into voltage signals. The mechanically applied force will cause the base to vibrate. The vibration signal is acquired, and the vibration signal is subjected to spectrum analysis such as Fourier transform to obtain the characteristic frequency components related to the mechanically applied force. The mechanically applied force data is obtained according to the force-frequency relationship curve.

[0085] It should be noted that the force-frequency relationship curve is obtained by directly measuring the mechanical force applied to the pressing base using a standard force sensor under the same environment as the actual pressing conditions, while simultaneously collecting vibration signals through piezoelectric accelerometers deployed at the four corners of the base; controlling the pressing machine to output all constant force values ​​within the output range, and maintaining the pressing state stable for a certain period of time (e.g., 30s) at each force value, collecting vibration signals within 30s and performing spectrum analysis to extract the characteristic frequency and amplitude of the vibration signal under the stress value; constructing a rectangular coordinate system with the standard value of the mechanically applied force on the x-axis and the vibration amplitude of the corresponding characteristic frequency on the y-axis, and fitting multiple sets of data points into a smooth curve to obtain the force-frequency relationship curve; for example, when the mechanically applied force is 60kN, the amplitude of the vibration signal at 50Hz is 0.5V, while at 80kN the amplitude of the same frequency is 0.8V. By plotting the force-frequency relationship curve in the rectangular coordinate system using the above data points, the corresponding mechanically applied force can be obtained based on the amplitude of the 50Hz frequency in actual pressing.

[0086] Step S15: Install a displacement sensor at the contact point between the pressure head and the bearing, and obtain contact deformation data based on the displacement sensor.

[0087] In some possible embodiments, the displacement sensor can be a high-precision inductive displacement sensor, with four sets evenly arranged along the circumference of the pressure head, and the detection end of the sensor pointing perpendicularly to the end face of the outer ring of the bearing. Before the pressing begins, the reference value of the four sets of displacement sensors is 0.5mm. After pressing for 20 seconds, the real-time displacement values ​​measured by the sensors are 1.2mm, 1.18mm, 1.22mm, and 1.19mm, respectively. The calculated contact deformation amounts are 0.7mm, 0.68mm, 0.72mm, and 0.69mm, respectively. The average value of the four sets of contact deformation amounts is taken as the final contact deformation amount of 0.6975mm.

[0088] Step S16: The relative motion speed data, temperature data, contact area data, and contact deformation data of the contact area are used to form a force influence dataset.

[0089] In some possible embodiments, the stress influence dataset is dynamically updated over time. For example, in the early stage of pressing, the stress influence dataset is obtained as relative motion speed data of 4 mm / s, temperature data of 28°C, contact area data of 30 cm², and contact deformation of 0.35 mm; in the later stage of pressing, the stress influence dataset is obtained as relative motion speed data of 0.5 mm / s, temperature data of 40°C, contact area data of 80 cm², and contact deformation of 0.6975 mm.

[0090] Step S2: Output real-time friction coefficient data based on the friction coefficient hybrid identification model, and obtain total friction force data through the friction coefficient data.

[0091] In this embodiment, step S2 includes:

[0092] Step S21: Construct a hybrid identification model for friction coefficients.

[0093] In this embodiment, step S21 includes:

[0094] Step S211: Collect historical pressing data under different working conditions, use the historical pressing data as a sample dataset, and standardize the sample dataset.

[0095] Specifically, historical press-fitting data includes historical mating surface normal pressure data, historical tangential friction force data, historical relative motion velocity data of contact areas, historical temperature data, historical contact deformation data, and corresponding historical friction coefficient data.

[0096] Furthermore, press-fitting data under different working conditions are obtained through historical data, including normal pressure data of mating surfaces, tangential friction data, relative motion speed data of contact areas, temperature data, contact deformation and corresponding measured values ​​of friction coefficients. The above data are then standardized using methods such as Z-score, converting all parameters into normally distributed data with a mean of 0 and a standard deviation of 1, providing standardized data input for subsequent processing.

[0097] In some possible embodiments, the sample data collected include: normal pressure data of mating surfaces 50kN, tangential friction force data 8kN, velocity data 5mm / s, temperature data 40℃, contact deformation amount 0.2mm, and friction coefficient value 0.16; and the above data are standardized, with the standardized value corresponding to the normal pressure data of mating surfaces 50kN being 1.2, and the standardized value corresponding to the tangential friction force data of 8kN being 1.5, etc.

[0098] Step S212: Construct an input layer based on historical mating surface normal pressure data, historical tangential friction force data, historical contact area relative motion velocity data, and historical temperature data; construct a hidden layer based on a modified linear unit function; construct an output layer based on historical friction feature vectors; and construct a front-end neural network module based on the input layer, hidden layer, and output layer.

[0099] In some possible embodiments, after receiving inputs of normalized values ​​of 1.2 for the mating surface normal pressure, 1.5 for the tangential friction force, 0.8 for the relative motion velocity of the contact area, and 1.0 for the temperature, the front-end neural network outputs multiple intermediate values, such as 0.3, 0.5, and -0.2, through the hidden layer. It retains non-negative values ​​and outputs multiple high-level features by modifying the linear unit function. Finally, the output layer outputs a friction feature vector, which includes a stick-slip characteristic factor of 0.3, a velocity sensitivity coefficient of 0.2, and a temperature influence weight of 0.1.

[0100] It should be noted that the modified linear unit function is a commonly used activation function in neural networks. Specifically, it outputs values ​​greater than 0 from the input and 0 otherwise. This method can simply and efficiently speed up the training of the model.

[0101] Step S213: Construct the frictional heat effect correction equation, and construct the backend physical constraint module based on the frictional heat effect correction equation.

[0102] Specifically, a corrected equation for the frictional heat effect is constructed based on Hertzian contact theory. This corrected equation is specifically expressed as follows:

[0103] ,in, This is expressed as the real-time friction coefficient value. , This is represented by the coupling coefficient, which is obtained experimentally. The feature fusion function is represented as the friction feature vector. This feature fusion can be processed using methods such as weighted summation. Represented as contact deformation, Represented as relative velocity data within the contact area. The data is expressed in Fahrenheit.

[0104] In some possible embodiments, it is assumed that the coefficient of friction is 0.25 calculated through the feature fusion function of the friction feature vector, the coupling coefficients are 0.7 and 0.003, the contact deformation is 0.2 mm, the relative velocity of the contact area is 5 mm / s, and the temperature is 313℃. Substituting these data into the frictional heat effect correction equation yields a real-time friction coefficient value of 0.175.

[0105] It should be noted that Hertzian contact theory describes the stress distribution and deformation behavior of two elastic bodies in contact. This theory assumes that the contacting bodies are homogeneous and isotropic elastic materials, the contact area is much smaller than the size of the objects, and the influence of friction is ignored, thus deriving the calculation formulas for key parameters such as contact area, contact pressure, and maximum stress. The frictional heat effect correction equation constructed through Hertzian contact theory takes into account the influence of contact deformation, relative velocity, and temperature on the coefficient of friction.

[0106] Step S214: Construct an initial friction coefficient hybrid identification model based on the front-end neural network module and the back-end physical constraint module. Train the initial friction coefficient hybrid identification model based on the sample dataset, using the mean square error as the loss function, and update it based on the Adam optimizer until the model converges, and output the converged friction coefficient hybrid identification model.

[0107] Specifically, the standardized sample data is input into the model, and the mean square error between the predicted friction coefficient and the measured value is calculated as the loss function. The connection weights, biases, and coupling coefficients of the physical constraint module of the neural network are updated through backpropagation using the Adam optimizer. , The convergence refers to the process of reducing the loss function value through gradient descent during training until the mean squared error is less than 0.001. Only then can the model fit the sample data and have the ability to predict new data.

[0108] It should be noted that mean squared error is often used to measure the difference between the model's predicted value and the true value. Specifically, it is expressed as the average of the squares of the prediction error. By amplifying the impact of larger errors, it enables the model to improve the accuracy of predictions by reducing bias during training.

[0109] Step S215: Evaluate the performance of the convergent friction coefficient hybrid identification model and output the friction coefficient hybrid identification model.

[0110] Understandably, the performance of the model can be evaluated using a method such as 5-fold cross-validation. Specifically, the sample dataset is divided into 5 parts, with 4 parts used as the training set and 1 part as the validation set each time, repeated 5 times, and the mean absolute error of the validation set is calculated. The mean absolute error reflects the average deviation between the predicted value and the measured value. When the mean absolute error is less than or equal to 0.01, it indicates that the model can still maintain high accuracy on unseen data and output a friction coefficient hybrid identification model that meets the conditions.

[0111] In some possible implementations, it is assumed that 5-fold cross-validation is used, with the absolute errors of each fold validation set being 0.009, 0.007, 0.008, 0.010, and 0.008, respectively, and the mean absolute error being 0.0084, which is less than 0.01, thus meeting the evaluation criteria. For example, if the measured friction coefficient of a validation sample is 0.18, and the model predicts a value of 0.172, with an error of 0.008, and this error is within the allowable range, the model is output.

[0112] Furthermore, if the convergent friction coefficient hybrid identification model passes performance evaluation, the model performance is iteratively improved through a multi-dimensional optimization strategy. The multi-dimensional optimization strategy includes checking the sample dataset to avoid data omissions that could lead to insufficient coverage; adjusting the front-end neural network structure and training parameters; correcting the back-end physical constraint module and recalibrating the coupling coefficients or optimizing the equation structure; retraining and evaluating after adjustment until the mean absolute error is less than or equal to 0.01. If the standard is still not met after more than 3 iterations, the input data needs to be acquired again.

[0113] Step S22 involves preprocessing the data on normal pressure on the mating surface, frictional force in the tangential direction, relative velocity of the contact area in the force-affected data set, temperature data, and contact deformation data.

[0114] Understandably, outliers in the acquired data can be removed using criteria such as the 3σ criterion. Specifically, if the deviation of a data point from the mean exceeds three times the standard deviation, the data is considered an outlier and is removed. The data after outlier removal can be smoothed by using a sliding window averaging method. The window size can be set to 5-10 sampling points, and the mean of the data within the window can be obtained to eliminate the influence of high-frequency noise, thereby improving the accuracy of the friction coefficient prediction.

[0115] Step S23: Input the normal pressure data of the mating surface, the friction force data in the tangential direction, the relative motion velocity data of the contact area, the temperature data, and the contact deformation data into the friction coefficient hybrid identification model.

[0116] Step S24: Based on the friction feature vector output by the front-end neural network module, input the friction feature vector to the back-end physical constraint module, and obtain the initial friction coefficient data through the back-end physical constraint module.

[0117] In some possible embodiments, it is assumed that the pre-processed mating surface normal pressure data of 48kN, tangential friction force data of 7.5kN, relative motion speed data of contact area of ​​1.5mm / s, and temperature data of 37℃ after pressing for 25s are input into the friction coefficient hybrid identification model, and the output stick-slip characteristic factor of 0.25, speed sensitivity coefficient of 0.18, temperature influence weight of 0.12 are formed to form a friction feature vector.

[0118] In some possible embodiments, the friction feature vector is calculated using a fusion function such as weighted summation to obtain a base friction coefficient of 0.14, a contact deformation of 0.15 mm, a relative motion velocity of 1.5 mm / s in the contact area, and a temperature of 37 °C. Substituting these values ​​into the frictional heat effect correction equation yields a correction term of 0.012, and the initial friction coefficient is obtained as 0.152 using the base friction coefficient and the correction term.

[0119] Step S25: Dynamically calibrate the initial friction coefficient data using temperature data to obtain real-time friction coefficient data.

[0120] Understandably, high temperatures often cause changes in the coefficient of friction. A calibration formula is established based on the relationship curve between temperature and the coefficient of friction. This calibration formula is specifically expressed as follows:

[0121] ,in This is the temperature influence coefficient, which is obtained experimentally. For real-time temperature, The reference temperature is typically 25°C. The calibrated real-time friction coefficient value is obtained using the formula described above.

[0122] It should be noted that before performing calculations using the calibration formula, the acquired data undergoes a normalization preprocessing operation, such as the min-max method, to ensure that the dimensions of the data are consistent.

[0123] In some possible embodiments, it is assumed that the initial friction coefficient is 0.152 when the pressing process is 25s, the real-time temperature is 37℃, the reference temperature is 25℃, the temperature influence coefficient k=0.002, and the real-time friction coefficient value is obtained by substituting into the calibration formula.

[0124] Step S26: Obtain the friction force data of each contact point by multiplying the real-time friction coefficient data with the corresponding normal pressure data of the mating surface, and obtain the total friction force data based on the friction force data of each contact point.

[0125] Specifically, the total friction force is expressed as the sum of the friction forces at each contact point, and the friction force at each contact point is obtained by multiplying the real-time friction coefficient value and the corresponding normal pressure data of the mating surface.

[0126] In some possible embodiments, assuming that the real-time friction coefficient is 0.155 at 25s during the pressing process, the normal pressure of the mating surface is distributed at different contact points as 10kN, 12kN, 8kN, 10kN, 5kN, and 3kN, and the friction forces at each contact point are 1.55kN, 1.86kN, 1.24kN, 1.55kN, 0.775kN, and 0.465kN, respectively, with a total friction force of 7.44kN.

[0127] Step S3: Generate a friction loss correction term based on the force influence dataset, and obtain net pressure data based on mechanically applied force data, total friction force data, and friction loss correction term.

[0128] It should be noted that in this scheme, step S2 achieves the first correction of friction force through a friction coefficient hybrid identification model. Specifically, it involves first extracting friction feature vectors using a front-end neural network, combining the initial friction coefficient with the output of the back-end physical constraint module, obtaining the real-time friction coefficient through dynamic temperature calibration, and then calculating the total friction force. The correction in step S2 mainly addresses the influence of basic parameters such as normal pressure, speed, and temperature on the friction coefficient itself, ensuring the basic accuracy of friction force calculation. Step S3, on the other hand, constructs a thermo-mechanical coupling correction coefficient and a dynamic impact correction factor, fuses them to obtain a comprehensive friction loss correction coefficient, and calculates the friction loss correction term based on the contact area, achieving the second correction of friction force. This correction mainly addresses the influence caused by the accumulation of friction work and speed fluctuations during the friction process, compensating for the dynamic friction effect not covered by the first correction. Through the methods of first and second corrections, the stability of the model is ensured, and the adaptability to complex situations is improved, providing accurate input for net pressing force and improving the accuracy and reliability of force monitoring during the pressing process.

[0129] In this embodiment, step S3 includes:

[0130] Step S31: Process the relative motion velocity data of the contact area in the force-affected dataset based on wavelet denoising to obtain the uniform velocity component and the instantaneous fluctuation component.

[0131] Specifically, wavelet denoising decomposes the velocity signal into three levels using the db4 wavelet basis, removes the wavelet coefficients corresponding to noise by using a pre-set threshold, reconstructs the processed coefficients to obtain the denoised velocity signal, extracts the uniform velocity component using a moving average method, the uniform velocity component reflects the overall trend of the velocity, and subtracts the uniform velocity component from the original signal to obtain the instantaneous fluctuation component, the instantaneous fluctuation component reflects the high-frequency changes of the velocity.

[0132] In some possible embodiments, it is assumed that the original data of the relative motion speed of the contact area when pressing for 25 seconds fluctuates between 1.3-1.7 mm / s. After denoising by 3-layer decomposition using the db4 wavelet basis, the signal is smoothed to 1.4-1.6 mm / s. The window size is set to 10 seconds, and the uniform velocity component is extracted by moving average to 1.5 mm / s, while the instantaneous fluctuation component is -0.2 to 0.1 mm / s.

[0133] Step S32: Obtain the cumulative value of frictional work based on the uniform velocity component, and establish a thermo-mechanical coupling correction coefficient by combining the temperature data in the force influence dataset.

[0134] Specifically, the relative motion distance is obtained by integrating the uniform velocity component over time. The cumulative frictional work is obtained by multiplying the relative motion distance, frictional force, and contact area. The larger the cumulative frictional work value, the more heat is generated, the higher the temperature, and the more obvious the change in the frictional properties of the material. Therefore, the thermo-mechanical coupling correction coefficient is positively correlated with the cumulative frictional work value and temperature. The functional relationship between the two is calibrated experimentally to obtain calibration coefficients a and b. The thermo-mechanical coupling correction coefficient is obtained by multiplying a by the cumulative frictional work value, adding b by the temperature difference, and adding 1. The thermo-mechanical coupling correction coefficient is used to quantify the influence of frictional heat and temperature on frictional loss.

[0135] In some possible embodiments, it is assumed that when pressing for 25 seconds, the uniform velocity component is 1.5 mm / s, the pressing time is 25 seconds, the relative motion distance is 0.0375 m, the friction force is 7.44 kN, the contact area is 0.008 m², the cumulative friction work is 50 J, the temperature is 37 °C, the initial temperature is 25 °C, the value of a is 0.001, the value of b is 0.005, and the thermo-mechanical coupling correction coefficient is calculated to be 1.11.

[0136] Step S33: Perform Fourier transform on the instantaneous fluctuation component to extract its frequency characteristics and amplitude distribution, and generate a dynamic impact correction factor.

[0137] Specifically, the instantaneous fluctuation components are transformed from the time domain to the frequency domain using Fourier transform to obtain the amplitude at different frequencies. By analyzing the frequency spectrum, the main frequency components and their corresponding amplitudes are obtained. The dynamic impact correction factor is positively correlated with the amplitude of the main frequencies. The larger the amplitude, the more severe the speed fluctuation, and the greater the impact on friction loss. The calibration coefficient c is obtained through experimental calibration. The dynamic impact correction factor is obtained by multiplying the maximum amplitude by the calibration coefficient c and adding 1.

[0138] In some possible embodiments, it is assumed that after performing a Fourier transform on the instantaneous fluctuation component, the maximum amplitude is 0.2 mm / s at a frequency of 5 Hz, and the calibration value of c is set to 0.5. The dynamic impact correction factor is calculated to be 1.1.

[0139] Step S34: The comprehensive correction coefficient for friction loss is obtained by fusing the thermo-mechanical coupling correction coefficient and the dynamic impact correction factor.

[0140] Understandably, the fusion can be achieved using a weighted multiplication method, whereby the comprehensive correction coefficient is the product of the thermo-mechanical coupling correction coefficient and the dynamic impact correction factor.

[0141] It should be noted that the weighted multiplication method not only considers the effects of accumulated frictional heat and velocity fluctuations on frictional loss, but also the effects of long-term frictional heat reflected by the thermo-mechanical coupling correction coefficient, and the effects of instantaneous velocity fluctuations reflected by the dynamic impact correction factor. The product of the two can quantify the degree of correction of frictional loss.

[0142] In some possible embodiments, assuming the thermo-mechanical coupling correction factor is 1.11 and the dynamic impact correction factor is 1.1, the calculated comprehensive friction loss correction factor is 1.221.

[0143] Step S35: Obtain the friction loss correction term based on the comprehensive correction coefficient for friction loss and the centralized contact area data of the force influence dataset.

[0144] Specifically, the friction loss correction term is positively correlated with the contact area and the comprehensive correction coefficient. The larger the contact area, the greater the friction loss; the larger the comprehensive correction coefficient, the more significant the additional impact of friction loss. The friction loss correction term can be expressed as the product of the comprehensive correction coefficient, the contact area, and the calibration coefficient d, where the unit of the calibration coefficient is... .

[0145] In some possible embodiments, assuming the obtained comprehensive correction factor for friction loss is 1.221, the contact area is 0.008 m², the calibration factor d is 100, and the friction loss correction term is calculated to be 0.9768 kN.

[0146] Step S36: Based on the mechanically applied force data, total friction force data, and friction loss correction term, construct the net indentation force equation and solve it to obtain the net indentation force data.

[0147] The net pressure equation is specifically expressed as follows:

[0148] ;

[0149] in, Represented as net pressure data, Represented as mechanically applied force data, This is expressed as total frictional force data. Represented as real-time friction coefficient data, This is represented as the normal force data for the mating surfaces. This is represented as a correction term for friction loss.

[0150] In some possible embodiments, it is assumed that when the pressing reaches 25s, the mechanically applied force is 62kN, and the total friction force obtained through the above step S26 is 7.44kN, the friction loss correction term is 0.9768kN, and the net pressing force equation is substituted to obtain a net pressing force of 53.5832kN.

[0151] It should be noted that the net pressure equation is obtained through Newton's third law, and the reaction force of the mechanically applied force is composed of net pressure data, total friction data, and friction loss correction term.

[0152] Step S4: Based on the continuum mechanics framework, input the three-dimensional spatial strain data into the deformation analysis model and output the plastic deformation factor and force distribution uniformity index.

[0153] In this embodiment, step S4 includes:

[0154] Step S41: Construct a deformation analysis model, mesh the three-dimensional space based on 8-node hexahedral elements, and simultaneously acquire material property data.

[0155] Specifically, the three-dimensional space of the bearing-journal contact body is meshed using 8-node hexahedral elements with an element size of less than or equal to 0.5 mm. The meshing must cover the entire contact area and surrounding key parts. Simultaneously, material property data, including elastic modulus, Poisson's ratio, and yield strength, are acquired and input into the deformation analysis model.

[0156] It should be noted that the elastic modulus is a key data point for stress-strain conversion in the generalized Hooke's law, while Poisson's ratio is used to correct the strain-stress coupling relationship in different directions. The principal stress is accurately extracted through the elastic modulus and Poisson's ratio. The yield strength is used to determine the deformation risk by calculating the plastic deformation factor. The accuracy of the deformation analysis model is ensured by the above material property data.

[0157] In some possible embodiments, it is assumed that when constructing the deformation analysis model, the bearing outer ring, rotor journal and contact area are divided into 8000 8-node hexahedral elements with an element size of 0.4 mm; and the elastic modulus of 205 GPa, Poisson's ratio of 0.28 and yield strength of 720 MPa are input into the deformation analysis model. The coordinate system of the deformation analysis model can be defined as a cylindrical coordinate system to satisfy the cylindrical structure of the bearing.

[0158] Step S42: Map the three-dimensional spatial strain data to the corresponding nodes of the deformation analysis model, and convert the three-dimensional spatial strain data into strain components.

[0159] Specifically, the three-dimensional spatial strain is mapped to the corresponding node of the deformation analysis model according to the sensor spatial coordinates. The three-dimensional spatial strain collected by the sensor is obtained in the Cartesian coordinate system. Through the coordinate transformation formula, the strain components in the Cartesian coordinate system are converted into strain components in the cylindrical coordinate system.

[0160] In some possible embodiments, it is assumed that when the compression is completed for 25 seconds, the strain in the Cartesian coordinate system is 200με, -300με, 150με, 50με, 30με, and 20με, and the cylindrical coordinate angle corresponding to the sensor coordinate is θ=30°. The strain components in the cylindrical coordinate system are calculated as 150με, 180με, 150με, 45με, 25με, and 30με according to the coordinate transformation formula, and the strain components in the cylindrical coordinate system are mapped to the corresponding nodes of the deformation analysis model.

[0161] Step S43: Convert the strain components into stress components.

[0162] Specifically, based on the generalized Hooke's law, strain components are converted into stress components to obtain the elastic matrix and strain components. The elastic matrix is ​​constructed using the elastic modulus and Poisson's ratio of the material. The elastic matrix and strain components are then converted into stress components through matrix operations based on the generalized Hooke's law.

[0163] It should be noted that one of the provisions of the generalized Hooke's Law states that after a solid material is subjected to force, there is a linear relationship between the stress and strain in the material. Therefore, the strain component can be converted into a stress component through the generalized Hooke's Law.

[0164] In some possible embodiments, it is assumed that the elastic matrix is ​​obtained through the elastic modulus and Poisson's ratio data in step S41, and the elastic matrix and strain components are substituted into the generalized Hooke's theorem formula through the strain components obtained in step S42 to obtain stress components of 50MPa, 60MPa, 40MPa, 5MPa, 3MPa, and 4MPa.

[0165] Step S44: Minimize the residual data of stress components and strain components using the least squares method, and obtain the stress tensor based on the residual data.

[0166] Specifically, the residual data between the calculated stress components and strain components is minimized using the least squares method, and the obtained stress components are corrected using the residual data; the corrected stress components are then used to construct a stress tensor, which includes radial stress, circumferential stress, axial stress, and shear stress components.

[0167] In some possible embodiments, it is assumed that the residual data calculated by the stress component obtained by the deformation analysis model and the strain component obtained by the sensor is 10 MPa. The residual data is reduced to below 2 MPa by the least squares method to obtain the corrected and optimized stress tensor. The corrected stress tensor is 52 MPa, 58 MPa, 41 MPa, 4.5 MPa, 3.2 MPa, and 3.8 MPa.

[0168] Step S45: Output the plastic deformation factor and force distribution uniformity index based on the stress tensor.

[0169] In this embodiment, step S45 includes:

[0170] Step S451: Extract the three principal stresses from the stress tensor and obtain the maximum principal stress and the minimum principal stress.

[0171] Specifically, the principal stresses are obtained by solving the characteristic equation of the stress tensor. The principal stress is expressed as the shear stress being zero in the direction. Three principal stresses are obtained in the three directions, and the three principal stresses are ordered in order of magnitude as the maximum principal stress, the intermediate principal stress, and the minimum principal stress.

[0172] In some possible embodiments, it is assumed that solving the characteristic equation for the stress tensor of a node yields three principal stresses of 90 MPa, 50 MPa, and 10 MPa.

[0173] It should be noted that there is no shear stress in the principal stress direction. By obtaining the three principal stresses, the complex stress state inside the material is simplified, and the maximum, intermediate and minimum normal stresses borne by a point in different directions can be directly reflected.

[0174] Furthermore, by obtaining the three principal stresses, the plastic deformation factor can be directly obtained according to the subsequent Tresca yield criterion. At the same time, the equivalent stress can also be calculated from the principal stresses, and then the stress distribution uniformity can be analyzed to generate a force distribution uniformity index.

[0175] Step S452: Process the maximum principal stress and minimum principal stress based on the Tresca yield criterion to obtain the maximum shear stress difference.

[0176] Specifically, the maximum shear stress difference is obtained through the Tresca yield criterion. The maximum shear stress difference is obtained by dividing the difference between the maximum principal stress and the minimum principal stress by 2. The maximum shear stress difference can directly reflect the plastic deformation trend of the material.

[0177] In some possible embodiments, it is assumed that the three principal stresses are 90 MPa, 50 MPa, and 10 MPa, respectively, and the maximum shear stress difference is 80 MPa obtained by the Tresca yield criterion.

[0178] Step S453: Obtain the plastic deformation factor based on the maximum shear stress difference and the yield strength in the material property data.

[0179] Specifically, the plastic deformation factor is obtained by dividing the maximum shear stress difference by the yield strength.

[0180] Understandably, the closer the plastic deformation factor is to 1, the closer the material is to the yield state, and the higher the risk of plastic deformation; when the plastic deformation factor is greater than or equal to 1, the material has already undergone plastic deformation.

[0181] In some possible embodiments, it is assumed that the material yield strength is 80 MPa, the maximum shear stress difference is 80 MPa, and the plastic deformation factor is 1.0, indicating that the material has just reached the yield state and has undergone plastic deformation.

[0182] Step S454: Obtain the equivalent stress of each node based on the three principal stresses, and simultaneously obtain the equivalent stress of discrete nodes based on the spatial interpolation algorithm. Simultaneously extract the maximum equivalent stress value and the average equivalent stress value of the contact area. The average equivalent stress value is represented as the average value of the equivalent stress of all nodes in the contact area.

[0183] Specifically, the three principal stresses are substituted into the equivalent stress formula to calculate the equivalent stress. At the same time, spatial interpolation is performed on the equivalent stress of discrete nodes to obtain a continuous stress distribution field, from which the maximum equivalent stress and the average equivalent stress of the contact area are extracted.

[0184] In some possible embodiments, it is assumed that the three principal stresses are 90MPa, 50MPa, and 10MPa, respectively, and the equivalent stress is obtained by substituting the three principal stresses into the equivalent stress formula. The equivalent stress is approximately 97.98MPa. After spatial interpolation, the maximum equivalent stress of the contact area is obtained as 100MPa, and the average equivalent stress is 60MPa.

[0185] It should be noted that spatial interpolation algorithms can obtain data from unknown regions using known discrete data points. In this embodiment, methods such as Kriging interpolation can be used. By analyzing the spatial correlation of the equivalent stress at known nodes, a semi-variogram function is constructed. The equivalent stress value of the interpolation point is obtained through the spatial positional relationship and correlation weight between the interpolation point and its surrounding points. Finally, the discrete node stress data is transformed into a continuous stress distribution field. The semi-variogram function describes the spatial variation characteristics and intensity of the regionalized variable and is the mathematical expectation of the square of the increment of the regionalized variable.

[0186] Step S455: Obtain the force distribution uniformity index based on the maximum equivalent stress value and the average equivalent stress value.

[0187] Specifically, the difference between the maximum equivalent stress value and the average equivalent stress value is divided by the average equivalent stress value to obtain the first value. The force distribution uniformity index is obtained by subtracting the first value from 1.

[0188] In some possible embodiments, assuming the maximum equivalent stress is 100 MPa and the average equivalent stress is 60 MPa, the calculated force distribution uniformity index is approximately 0.333. The closer the force distribution uniformity index is to 1, the more uniform the stress distribution; the smaller the force distribution uniformity index, the less uniform the local stress distribution.

[0189] Step S5: During the press-fitting process, the stress on the generator rotor bearing is monitored based on the net press-fitting force data, plastic deformation factor, and force distribution uniformity index.

[0190] In some possible embodiments, it is assumed that during the press-fitting process, the net press-fitting force is monitored in the range of 53-55 kN, the plastic deformation factor is from 0.6 to 0.85, and the force distribution uniformity index is from 0.7 to 0.5.

[0191] It should be noted that the net pressing force threshold range is set to 53-55k. This range is a reasonable range for the pressing machinery. If the real-time net pressing force data is detected to be outside the range and there are abnormal fluctuations such as a sudden increase or decrease, it indicates a sudden change in pressing resistance or a malfunction in the pressing machinery. The plastic deformation factor and force distribution uniformity index are also detected simultaneously. Changes in these two values ​​will directly affect the pressing quality. The specific impact can be reflected in step S6.

[0192] Understandably, abnormal net pressing force may cause the bearing to be installed too loosely or too tightly during the repressing process; excessive plastic deformation factor may cause permanent deformation of the bearing or journal; and low force distribution uniformity index may cause local wear and cracks. Only by monitoring and controlling the force during the press-fitting process by combining the changes of the above three data can the quality of generator rotor bearing press-fitting be ensured.

[0193] Step S6: Based on the force distribution uniformity index and plastic deformation factor, determine the off-center load, obtain the off-center load result, and adjust the press-fitting machinery in real time based on the off-center load result.

[0194] In this embodiment, step S6 includes:

[0195] Step S61: Determine the eccentric load level based on the combination of the force distribution uniformity index and the plastic deformation factor.

[0196] When the force distribution uniformity index is greater than or equal to 0.6 and less than 0.7, and the plastic deformation factor is less than 0.9, it is judged as a slight off-center load and an audible and visual alarm is triggered.

[0197] When the force distribution uniformity index is greater than or equal to 0.5 and less than 0.6, or the plastic deformation factor is greater than or equal to 0.9 and less than 1.0, it is judged as moderate off-center load, and the pressing speed is reduced.

[0198] When the force distribution uniformity index is less than 0.5 or the plastic deformation factor is greater than or equal to 1.0, it is judged as severe off-center loading, and the press-fitting machinery should be stopped immediately.

[0199] In some possible embodiments, assuming that at a certain moment the force distribution uniformity index is 0.55 and the plastic deformation factor is 0.88, it is determined to be a moderate off-center load, and the system automatically reduces the pressing speed of the pressing machine from 5 mm / s to 2.5 mm / s; at a certain moment the force distribution uniformity index is 0.45 and the plastic deformation factor is 0.98, it is determined to be a severe off-center load, and the pressing machine is immediately stopped and inspected.

[0200] Step S62: Obtain the off-center load vector based on the stress gradient direction of the deformation analysis model, and generate compensation commands to control the press-fitting machinery in real time.

[0201] Specifically, the direction of the off-center load vector is determined by analyzing the direction of the stress gradient in the deformation analysis model. The direction of the stress gradient is the direction of the fastest stress change. The off-center load amplitude is obtained based on the difference between the maximum stress and the average stress. Through the structure of the press-fitting machine, the off-center load vector is converted into the thrust adjustment of four sets of hydraulic cylinders. The thrust of the hydraulic cylinders corresponding to high stress areas is increased, and the thrust of the hydraulic cylinders corresponding to low stress areas is decreased, thereby achieving balance.

[0202] In some possible embodiments, assuming the obtained off-center load vector direction is 60° and the amplitude is 6kN, the four sets of hydraulic cylinders of the press machine correspond to the directions of 0°, 90°, 180° and 270° respectively. The generated compensation command specifically means that the hydraulic cylinder thrust is increased by 2kN in the 0° direction, the hydraulic cylinder thrust is increased by 4kN in the 90° direction, the hydraulic cylinder thrust is decreased by 2kN in the 180° direction and the hydraulic cylinder thrust is decreased by 4kN in the 270° direction. The off-center load generated in the 60° direction is offset by the above compensation command.

[0203] Step S63: Monitor the stress distribution changes after compensation in real time. When the force distribution uniformity index is greater than or equal to 0.7 and the plastic deformation factor is less than 0.9, the warning is lifted and the normal pressing parameters are restored.

[0204] In some possible embodiments, the compensation command in step S62 detects that the force distribution uniformity index has changed to 0.72 and the plastic deformation factor has changed to 0.85, which meets the conditions for lifting the warning. The system then stops the audible and visual alarm and restores the pressing speed from 2.5 mm / s to 5 mm / s.

[0205] It should be noted that components such as strain gauges and thin-film pressure sensors are mainly integrated into the tooling fixtures of the press-fitting equipment and can be reused in the press-fitting process of all rotors of the same model. For cases where it is necessary to monitor the local strain of the bearing outer ring or journal, mass production can be achieved through sampling calibration and model migration.

[0206] For example, sensors can be temporarily installed on a small number of rotors produced in the first batch. The stress and strain data obtained can be used to train and optimize the model until the model converges. The complete stress state of the rotor during the pressing process can be derived from the converged model and the data obtained by the sensors on the tooling fixture.

[0207] Figure 2 The diagram shows a schematic of a force monitoring system based on the press-fitting process of a generator rotor bearing, which can realize the ideas of this application, according to some embodiments of this application.

[0208] Specifically, a stress monitoring system based on the press-fitting process of generator rotor bearings includes:

[0209] The acquisition module acquires mechanically applied force data, mating surface normal pressure data, tangential friction force data, three-dimensional spatial strain data, and force influence dataset in real time through a monitoring sensor group, and acquires total friction force data through friction coefficient data;

[0210] The input module, based on the continuum mechanics framework, inputs three-dimensional spatial strain data into the deformation analysis model;

[0211] The output module is used to output the plastic deformation factor and the force distribution uniformity index, and to output real-time friction coefficient data based on the friction coefficient hybrid identification model.

[0212] The processing module generates a friction loss correction term based on the force influence dataset, and obtains net pressure data based on mechanically applied force data, total friction force data, and friction loss correction term.

[0213] The detection module monitors the stress on the generator rotor bearing during the press-fitting process based on net press-in force data, plastic deformation factor, and force distribution uniformity index.

[0214] The control module determines the off-center load based on the force distribution uniformity index and the plastic deformation factor, obtains the off-center load result, and performs real-time control of the press-fitting machinery based on the off-center load result.

[0215] The specific usage and function of this embodiment are explained below:

[0216] First, multi-dimensional information such as mechanical applied force, mating surface normal pressure, tangential friction force, three-dimensional spatial strain, and force influence data are collected in real time using multiple types of sensors. This overcomes the limitations of traditional technologies that rely on a single sensor or a small number of monitoring points, ensuring data integrity and providing data support for subsequent analysis. Then, a friction coefficient hybrid identification model outputs the real-time friction coefficient, while a friction loss correction term is generated based on the force influence dataset. Through physical constraints of Hertzian contact theory and dynamic temperature calibration, the accuracy of friction coefficient calculation is improved, resulting in more precise total friction force and net indentation force data. Next, using a continuum mechanics framework and deformation analysis model, plastic deformation factors and force distribution uniformity indices are obtained from the three-dimensional spatial strain data. Combined with the net indentation force, the press-fitting process is monitored in real time to obtain the bearing stress distribution and deformation risk. Finally, through off-center load early warning and dynamic control, control data is obtained, and real-time data is inverted from the control data and applied to the real-time production process, improving the efficiency and accuracy of the generator rotor bearing press-fitting process, thereby ensuring press-fitting quality.

[0217] Furthermore, embodiments of the present invention also provide an electronic device, comprising:

[0218] At least one processor; and at least one memory communicatively connected to the processor; wherein the memory stores instructions executable by at least one processor, the instructions being executed by at least one processor to enable at least one processor to perform the method proposed in Embodiment 1 of the present invention.

[0219] The following is a detailed introduction to the various components of the electronic device:

[0220] In this context, the processor is the control center of the electronic device. It can be a single processor or a collective term for multiple processing elements. For example, a processor can be one or more central processing units (CPUs), specific integrated circuits (SIs), or one or more integrated circuits configured to implement Embodiment 1 of this invention, such as one or more microprocessors (DSPs), or one or more field-programmable gate arrays (FPGs).

[0221] The processor can perform various functions of an electronic device by running or executing software programs stored in memory and by calling data stored in memory.

[0222] The memory is used to store the software program that executes the solution of the present invention, and the execution is controlled by the processor. For specific implementation methods, please refer to the above method embodiments, which will not be repeated here.

[0223] The memory can be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, a random access memory (RM) or other type of dynamic storage device capable of storing information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a read-only optical disc (D-ROM), or other optical disc storage, optical disk storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. The memory can be integrated with the processor or exist independently and coupled to the processor through an interface circuit of an electronic device; this embodiment of the invention does not specifically limit this.

[0224] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via limited means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0225] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, "and / or I-frame / P-frame interval offset rate" can represent three cases: existing alone, existing simultaneously with "and / or I-frame / P-frame interval offset rate", and existing alone. The I-frame / P-frame interval offset rate can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it may also indicate an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0226] It should be understood that, in the embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0227] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for monitoring the force during the press-fitting process of generator rotor bearings, characterized in that, The method includes: The sensor array is used to acquire real-time data on mechanically applied forces, mating surface normal pressure, tangential friction, three-dimensional spatial strain, and force influence. Real-time friction coefficient data is output based on the friction coefficient hybrid identification model, and total friction force data is obtained through the friction coefficient data. A friction loss correction term is generated based on the force influence dataset, and net pressure force data is obtained based on mechanical applied force data, total friction force data, and friction loss correction term. Based on the continuum mechanics framework, three-dimensional spatial strain data is input into the deformation analysis model, and the plastic deformation factor and force distribution uniformity index are output. During the press-fitting process, the stress on the generator rotor bearing is monitored based on net press-fitting force data, plastic deformation factor, and force distribution uniformity index. The off-center load is determined based on the force distribution uniformity index and the plastic deformation factor. The off-center load result is then used to adjust the press-fitting machinery in real time.

2. The method for monitoring the force during the press-fitting process of generator rotor bearings according to claim 1, characterized in that, The system acquires real-time data on mechanically applied forces, mating surface normal pressure, tangential friction, three-dimensional strain, and force influence by monitoring a sensor array. The monitoring sensor group includes multiple strain gauges, multiple pressure sensors, multiple strain-type sensors, multiple acceleration sensors, displacement sensors, and a temperature compensation module. The force influence dataset includes relative motion velocity data of the contact area, temperature data, contact area data, and contact deformation data. Multiple sets of strain gauges are deployed on the pressure head, bearing outer ring, rotor journal and press-fit base respectively. The multiple sets of strain gauges are strain rose structures. Three-dimensional spatial strain data are obtained based on the multiple sets of strain gauges. Multiple pressure sensors are symmetrically arranged on the outer ring of the bearing. Based on the multiple pressure sensors, the normal pressure data of the mating surface is obtained, and the contact area data and relative motion speed data of the contact area are obtained simultaneously. Multiple strain gauge sensors are installed at key parts of the rotor journal, and tangential friction force data and temperature data are obtained based on the strain gauge sensors and temperature compensation module, respectively. A set of acceleration sensors is deployed at each of the four corners of the press-fit base to obtain mechanical force data based on the acceleration sensors; A displacement sensor is installed at the contact point between the pressure head and the bearing to obtain contact deformation data.

3. The method for monitoring the force during the press-fitting process of generator rotor bearings according to claim 1, characterized in that, Real-time friction coefficient data is output based on the friction coefficient hybrid identification model. Total friction force data is obtained from the friction coefficient data, including: Preprocessing operations are performed on the normal pressure data of the mating surface, the frictional force data in the tangential direction, the relative motion velocity data of the contact area in the force influence data, the temperature data, and the contact deformation data. The normal pressure data of the mating surface, the friction force data in the tangential direction, the relative motion velocity data of the contact area, the temperature data, and the contact deformation data are input into the friction coefficient hybrid identification model; The friction feature vector is output by the front-end neural network module and input to the back-end physical constraint module to obtain the initial friction coefficient data. The initial friction coefficient data is dynamically calibrated using temperature data to obtain real-time friction coefficient data; The friction force data of each contact point is obtained by multiplying the real-time friction coefficient data with the corresponding normal pressure data of the mating surface, and the total friction force data is obtained based on the friction force data of each contact point.

4. The method for monitoring the force during the press-fitting process of generator rotor bearings according to claim 3, characterized in that, Constructing a hybrid identification model for friction coefficients, including: Collect historical pressing data under different working conditions, use the historical pressing data as a sample dataset, and standardize the sample dataset. The historical press-fitting data includes historical mating surface normal pressure data, historical tangential friction force data, historical contact area relative motion speed data, historical temperature data, historical contact deformation data, and corresponding historical friction coefficient data. The input layer is constructed based on historical mating surface normal pressure data, historical tangential friction force data, historical contact area relative motion velocity data, and historical temperature data. The hidden layer is constructed based on the modified linear unit function. The output layer is constructed based on historical friction feature vectors. The front-end neural network module is constructed based on the input layer, hidden layer, and output layer. A corrected equation for the frictional heat effect is constructed, and a backend physical constraint module is built based on the corrected equation for the frictional heat effect. An initial friction coefficient hybrid identification model is constructed based on a front-end neural network module and a back-end physical constraint module. The initial friction coefficient hybrid identification model is trained based on a sample dataset, with mean square error as the loss function and updated based on the Adam optimizer until the model converges, and the converged friction coefficient hybrid identification model is output. The performance of the convergent friction coefficient hybrid identification model is evaluated, and the friction coefficient hybrid identification model is output.

5. The method for monitoring the force during the press-fitting process of generator rotor bearings according to claim 1, characterized in that, Friction loss correction terms are generated based on the force influence dataset. Net indentation force data is obtained based on mechanically applied force data, total friction force data, and friction loss correction terms, including: Based on wavelet denoising, the relative motion velocity data of the contact area in the force influence dataset is processed to obtain the uniform velocity component and the instantaneous fluctuation component. The cumulative value of frictional work is obtained based on the uniform velocity component, and a thermo-mechanical coupling correction coefficient is established by combining the temperature data in the force influence dataset. Perform Fourier transform on the instantaneous fluctuation components to extract their frequency characteristics and amplitude distribution, and generate a dynamic impact correction factor. The comprehensive correction coefficient for friction loss is obtained by fusing the thermo-mechanical coupling correction coefficient and the dynamic impact correction factor. The friction loss correction term is obtained based on the comprehensive correction coefficient for friction loss and the contact area data in the stress influence dataset. The net indentation force equation is constructed based on the mechanically applied force data, total friction force data, and friction loss correction term, and the net indentation force data is obtained by solving it.

6. The method for monitoring the force during the press-fitting process of generator rotor bearings according to claim 5, characterized in that, Based on the mechanically applied force data, total friction force data, and friction loss correction term, a net indentation force equation is constructed, and the net indentation force data is obtained by solving it, including: The net pressure equation is specifically expressed as follows: ； in, Represented as net pressure data, Represented as mechanically applied force data, This is expressed as total frictional force data. Represented as real-time friction coefficient data, This is represented as the normal force data for the mating surfaces. This is represented as a correction term for friction loss.

7. The method for monitoring the force during the press-fitting process of generator rotor bearings according to claim 1, characterized in that, Based on the continuum mechanics framework, three-dimensional spatial strain data is input into the deformation analysis model, and the output plastic deformation factor and force distribution uniformity index are included: A deformation analysis model was constructed, and a three-dimensional mesh was generated based on 8-node hexahedral elements, while material property data was obtained. The three-dimensional spatial strain data is mapped to the corresponding nodes of the deformation analysis model, and the three-dimensional spatial strain data is converted into strain components. Convert strain components into stress components; The residual data of stress and strain components are obtained by minimizing the least squares method, and the stress tensor is obtained based on the residual data. The stress tensor is used to output the plastic deformation factor and the force distribution uniformity index.

8. The method for monitoring the force during the press-fitting process of generator rotor bearings according to claim 7, characterized in that, Based on the stress tensor output plastic deformation factor and force distribution uniformity index, including: Extract the three principal stresses from the stress tensor and obtain the maximum and minimum principal stresses; The maximum shear stress difference is obtained by processing the maximum principal stress and minimum principal stress based on the Tresca yield criterion. The plastic deformation factor is obtained based on the maximum shear stress difference and the yield strength in the material property data; The equivalent stress of each node is obtained based on the three principal stresses, and the equivalent stress of discrete nodes is obtained based on the spatial interpolation algorithm. The maximum equivalent stress value and the average equivalent stress value of the contact area are extracted simultaneously. The average equivalent stress value is expressed as the average value of the equivalent stress of all nodes in the contact area. The uniformity index of force distribution is obtained based on the maximum equivalent stress value and the average equivalent stress value.

9. The method for monitoring the force during the press-fitting process of generator rotor bearings according to claim 1, characterized in that, Off-center loading is determined based on force distribution uniformity index and plastic deformation factor, and the off-center loading results are obtained. Real-time control of the press-fitting machinery is then performed based on these results, including: Determining the eccentric load level based on a combination of force distribution uniformity index and plastic deformation factor: When the force distribution uniformity index is greater than or equal to 0.6 and less than 0.7, and the plastic deformation factor is less than 0.9, it is judged as a slight off-center load and an audible and visual alarm is triggered. When the force distribution uniformity index is greater than or equal to 0.5 and less than 0.6, or the plastic deformation factor is greater than or equal to 0.9 and less than 1.0, it is judged as moderate off-center load, and the pressing speed is reduced. When the force distribution uniformity index is less than 0.5 or the plastic deformation factor is greater than or equal to 1.0, it is judged as severe off-center loading, and the press-fitting machinery should be stopped immediately. Based on the stress gradient direction of the deformation analysis model, the off-center load vector is obtained, and compensation commands are generated to control the press-fitting machinery in real time. Real-time monitoring of stress distribution changes after compensation; when the force distribution uniformity index is greater than or equal to 0.7 and the plastic deformation factor is less than 0.9, the warning is lifted and normal press-fitting parameters are restored.

10. A stress monitoring system based on the press-fitting process of generator rotor bearings, characterized in that, include: The acquisition module acquires mechanically applied force data, mating surface normal pressure data, tangential friction force data, three-dimensional spatial strain data, and force influence dataset in real time through a monitoring sensor group, and acquires total friction force data through friction coefficient data; The input module, based on the continuum mechanics framework, inputs three-dimensional spatial strain data into the deformation analysis model; The output module is used to output the plastic deformation factor and the force distribution uniformity index, and to output real-time friction coefficient data based on the friction coefficient hybrid identification model. The processing module generates a friction loss correction term based on the force influence dataset, and obtains net pressure data based on mechanically applied force data, total friction force data, and friction loss correction term. The detection module monitors the stress on the generator rotor bearing during the press-fitting process based on net press-in force data, plastic deformation factor, and force distribution uniformity index. The control module determines the off-center load based on the force distribution uniformity index and the plastic deformation factor, obtains the off-center load result, and performs real-time control of the press-fitting machinery based on the off-center load result.

Images