A processing method and device for improving contact fatigue strength of a bearing raceway

Abstract

The application discloses a processing method and device for improving the contact fatigue strength of a bearing raceway, and relates to the technical field of bearing processing. The method comprises the following steps: a broadband acoustic emission sensor is installed on a sound transmission path to collect acoustic emission signals in a shot blasting process in real time, characteristic analysis is performed, and acoustic signal characteristic parameters are obtained; the acoustic signal characteristic parameters are adaptively corrected by dynamically estimating the interference strength of the acoustic signals; the curvature radius of the raceway and the corrected acoustic signal characteristic parameters are taken as inputs, a predicted residual stress field characteristic is obtained through model reasoning, and a predicted contact fatigue resistance is obtained by mapping; and the shot blasting impact processing is controlled by taking the approximation of the predicted contact fatigue resistance to a preset contact fatigue resistance index as a target. The technical problem of the prior art that the shot blasting processing method cannot realize real-time sensing of the residual stress and the contact fatigue resistance state of the raceway, is easily disturbed by the processing environment, and results in low signal-to-noise ratio of the detection signal and poor processing stability is solved.

Description

Technical Field

[0001] This invention relates to the field of bearing processing technology, and specifically to a processing method and apparatus for improving the contact fatigue strength of bearing raceways. Background Technology

[0002] With the increasing demands on bearing service life in fields such as wind power and high-end equipment, the bearing raceway, as the core part that bears alternating contact loads, determines the overall service life of the bearing through its resistance to contact fatigue strength. Shot peening is currently the mainstream process for improving the resistance to contact fatigue strength of bearing raceways. By introducing beneficial residual compressive stress on the raceway surface, it can effectively inhibit the initiation and propagation of fatigue cracks.

[0003] However, in existing shot peening processes, most rely on preset process parameters, making it impossible to perceive the formation state of residual stress field and contact fatigue strength on the raceway surface in real time during processing. This results in poor processing accuracy and stability. Some solutions that can detect the process may be affected by the complex environment of shot peening, leading to low signal-to-noise ratios and large errors in characteristic parameters of the collected acoustic emission and other detection signals. These solutions cannot accurately reflect the actual processing state, or they may not consider the influence of raceway curvature on acoustic signal propagation and residual stress field distribution, resulting in insufficient accuracy in the inferred contact fatigue strength. This makes it difficult to achieve adaptive control of the processing process and can easily lead to under-strength or over-strength, failing to reliably meet the contact fatigue strength requirements of high-end equipment for bearing raceways. Summary of the Invention

[0004] This invention provides a processing method and apparatus for improving the contact fatigue strength of bearing raceways, solving the technical problems of existing shot peening methods that cannot detect the residual stress and contact fatigue strength of raceways in real time, are easily affected by the processing environment, and result in low signal-to-noise ratio and poor processing stability.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a processing method for improving the contact fatigue strength of bearing raceways, the method comprising: When shot peening the bearing raceway, a broadband acoustic emission sensor is installed on the sound transmission path that is rigidly connected to the raceway to collect the acoustic emission signals in real time during the shot peening process. The acoustic emission signal is analyzed to obtain acoustic signal feature parameters; By dynamically estimating the intensity of acoustic signal interference, the acoustic signal characteristic parameters are adaptively corrected to obtain the corrected acoustic signal characteristic parameters. Using the raceway curvature radius and the modified acoustic signal characteristic parameters as input, the predicted residual stress field characteristics are obtained through model reasoning, and the predicted contact fatigue strength is obtained by mapping. The shot peening impact processing is controlled with the goal of making the predicted contact fatigue strength approach the preset contact fatigue strength index.

[0006] Secondly, the present invention provides a processing apparatus for improving the contact fatigue strength of bearing raceways, comprising: The acoustic emission signal acquisition module is used to install a broadband acoustic emission sensor on the sound transmission path that is rigidly connected to the raceway during shot peening of the bearing raceway, and to acquire the acoustic emission signal in real time during the shot peening process. The acoustic signal feature parameter acquisition module is used to perform feature analysis on the acoustic emission signal and acquire acoustic signal feature parameters; The acoustic signal feature parameter correction module is used to adaptively correct the acoustic signal feature parameters by dynamically estimating the acoustic signal interference intensity, and to obtain the corrected acoustic signal feature parameters. The prediction model training module is used to obtain the predicted residual stress field characteristics through model inference by taking the raceway curvature radius and the modified acoustic signal characteristic parameters as input, and to map and obtain the predicted contact fatigue strength. The shot peening impact execution module is used to control the shot peening impact process with the goal of making the predicted contact fatigue strength approach the preset contact fatigue strength index.

[0007] This invention provides one or more technical solutions, which have at least the following technical effects or advantages: This invention provides a processing method and apparatus for improving the contact fatigue strength of bearing raceways. First, broadband acoustic emission signals are acquired in real time during shot peening of the bearing raceways. The acoustic signal characteristic parameters are adaptively corrected by dynamically estimating the interference intensity of the processing environment, effectively reducing the interference of the complex shot peening environment on detection accuracy and obtaining more accurate characteristic signals. Second, the raceway curvature radius and the corrected acoustic signal characteristics are used together as model inputs. The influence of different raceway curvatures on acoustic signal propagation attenuation and residual stress field distribution is considered, improving the prediction accuracy of residual stress field characteristics and contact fatigue strength. Finally, the shot peening process is controlled through a closed-loop control of the prediction results, ensuring that the contact fatigue strength of the processed raceway consistently reaches the preset target, avoiding problems of insufficient or excessive strengthening.

[0008] Through the above technical solution, this invention solves the technical problems in the existing shot peening process, such as the inability to perceive the formation state of residual stress and contact fatigue strength of the raceway in real time, the low signal-to-noise ratio of the detection signal due to interference from complex processing environment, large errors in characteristic parameters, and the failure to consider the influence of raceway curvature on detection accuracy and residual stress field, resulting in insufficient accuracy in predicting contact fatigue strength and poor processing stability. It can realize adaptive closed-loop control of the shot peening process, effectively ensuring that the contact fatigue strength of the bearing raceway after processing is stable and meets the preset index requirements, and is suitable for the processing needs of long-life bearings in the high-end equipment field. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic flowchart of a processing method for improving the contact fatigue strength of bearing raceways according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the calculation of real-time interference intensity values ​​in a processing method for improving the contact fatigue resistance of bearing raceways, as provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of a processing device for improving the contact fatigue strength of bearing raceways, provided in an embodiment of the present invention.

[0011] The components represented by each number in the attached diagram are explained below: Acoustic emission signal acquisition module 11, acoustic signal characteristic parameter acquisition module 12, acoustic signal characteristic parameter correction module 13, prediction model training module 14, and shot peening impact execution module 15. Detailed Implementation

[0012] This invention provides a processing method and apparatus for improving the contact fatigue strength of bearing raceways, addressing the technical problems of existing shot peening methods that cannot detect the residual stress and contact fatigue strength of raceways in real time, are easily affected by the processing environment, resulting in low signal-to-noise ratio of detection signals and poor processing stability.

[0013] Example 1, as Figure 1 As shown, this embodiment of the invention provides a processing method for improving the contact fatigue strength of bearing raceways, comprising: S10: When shot peening the bearing raceway, a broadband acoustic emission sensor is installed on the sound transmission path that is rigidly connected to the raceway to collect the acoustic emission signal during the shot peening process in real time. In this embodiment of the invention, shot peening refers to a mechanical strengthening process that uses a high-speed shot stream to impact the surface of the bearing raceway, causing plastic deformation of the raceway surface, introducing a residual compressive stress field, and thereby improving the contact fatigue strength of the raceway. During the process, the acoustic emission signal generated by the shot impacting the raceway surface carries relevant information such as impact energy and the degree of surface deformation, which can be used to characterize the formation of residual stress on the raceway surface under the current processing state.

[0014] Among them, the installation position of the broadband acoustic emission sensor can be selected from the mandrel or fixture of the inner ring of the bearing and the outer end face of the outer ring of the bearing. This can ensure that the sensor and the raceway are rigidly connected and the acoustic emission signal is stably transmitted, without interfering with the shot peening process.

[0015] Specifically, a broadband acoustic emission sensor is installed on the sound transmission path that is rigidly connected to the raceway, including: The wideband acoustic emission sensor operates in the frequency range of 100kHz to 1MHz and is mounted on a component that forms a rigid acoustic transmission path with the bearing raceway. For the outer raceway, the broadband acoustic emission sensor is fixedly installed on the outer cylindrical surface of the bearing outer ring, with the sensitive surface facing the bottom area of ​​the raceway. For the inner raceway, the broadband acoustic emission sensor is mounted on the mandrel or fixture, and signal transmission is achieved through rotational coupling with the inner raceway.

[0016] In this embodiment of the invention, firstly, a wideband acoustic emission sensor is installed on a component that forms a rigid acoustic transmission path with the bearing raceway. The operating frequency range is 100kHz to 1MHz, which can cover the main frequency range of acoustic emission signals generated by shot peening impact, thus avoiding the loss of effective feature information by narrowband sensors.

[0017] Secondly, appropriate installation positions are selected for different types of bearing raceways. Specifically, for the outer ring raceway, the broadband acoustic emission sensor is fixedly installed on the outer cylindrical surface of the outer ring, with the sensitive surface of the sensor facing the bottom area of ​​the raceway. This ensures that the elastic waves excited by the projectile impacting the raceway surface propagate to the sensor via the shortest acoustic path. For the inner ring raceway, since the inner ring rotates with the spindle, the broadband acoustic emission sensor is fixedly installed on the mandrel or the fixture holding the inner ring. The sensor and the inner ring are connected by rotational coupling to achieve signal transmission. Rotational coupling includes, but is not limited to, conductive slip ring contact transmission or wireless sensing non-contact transmission. The rotational coupling structure ensures stable transmission of acoustic signals from the rotating inner ring to the fixed sensor, which does not affect the rotational machining of the inner ring and ensures continuous and stable acquisition of acoustic emission signals.

[0018] Furthermore, the acoustic emission signals during the shot peening process are acquired in real time, including: During the first preset time period after the shot peening impact process is started, acoustic emission signals are collected at the first sampling frequency, or the process is in a state of waiting to be triggered for collection. When the processing time exceeds the first preset time period, the acoustic emission signal is continuously collected at a second sampling frequency higher than the first sampling frequency; During the continuous acquisition at the second sampling frequency, the acoustic emission signal is monitored in real time, and only the acoustic signal characteristic parameters are recorded. When the root mean square value of the acoustic emission signal or the ring count exceeds the preset trigger threshold, the acoustic emission waveform data within the preset time period before and after the current moment is fully recorded.

[0019] In this embodiment of the invention, firstly, during the first preset time period after the shot peening impact processing is started, the processing state is not yet stable and the impact of the shot flow has not yet entered a steady state. At this time, sampling is performed at a lower first sampling frequency or directly in a state of waiting to be triggered, which can reduce unnecessary storage and computing resource consumption. The first preset time period is preset according to the bearing specifications and shot peening process, for example, it can be set to 5 seconds; the first sampling frequency is preset according to the sensor performance and resource configuration, for example, it can be set to 10kHz.

[0020] When the processing time exceeds the first preset time period, the processing enters the stable shot peening stage, and the process switches to a higher second sampling frequency for formal acquisition to ensure the complete acquisition of effective signals. For example, the second sampling frequency can be set to 100kHz, which is higher than the first sampling frequency and can cover the broadband signal characteristics generated by shot peening impact, ensuring that the acquired signals are complete and effective.

[0021] During continuous acquisition at the second sampling frequency, the acoustic emission signal is monitored in real time. Only the calculated acoustic signal characteristic parameters are recorded, without the need to store complete waveform data, which can significantly reduce data storage. When the root mean square value or ring count of the acoustic emission signal exceeds the preset trigger threshold, it indicates that there is an abnormal impact energy or a sudden change in deformation state. At this time, the acoustic emission waveform data within the preset time before and after the current moment is recorded completely. The preset trigger threshold is pre-calibrated based on the maximum abnormal impact energy allowed on the raceway surface. The preset time can be set from 10ms before the abnormal trigger to 30ms after the trigger, which is convenient for subsequent tracking of abnormal processing state and also preserves complete original data for subsequent model iteration and optimization.

[0022] S20: Perform feature analysis on the acoustic emission signal to obtain acoustic signal feature parameters; The acoustic signal characteristic parameters include the centroid frequency of the spectrum, the peak-to-peak frequency of the spectrum, the ringing count, and the root mean square value.

[0023] In this embodiment of the invention, the acquired raw acoustic emission signal is first pre-filtered to remove low-frequency background noise caused by the mechanical vibration of the shot peening equipment, and then the above-mentioned characteristic parameters are extracted: the centroid frequency of the spectrum refers to the centroid frequency of the power spectrum of the acoustic emission signal, which is used to characterize the distribution centroid of the main frequency components of the signal; the peak-to-peak frequency of the spectrum refers to the frequency corresponding to the largest amplitude in the power spectrum of the acoustic emission signal, which is used to characterize the frequency point where the signal energy is most concentrated; the ringing count refers to the number of oscillations of the acoustic emission signal waveform exceeding the preset threshold voltage, which is used to characterize the event activity of the signal; the root mean square value is the effective voltage value of the acoustic emission signal per unit time, which is used to characterize the average energy level of the signal. The four types of characteristic parameters together characterize the state of plastic deformation and residual stress evolution of the raceway surface during shot peening impact.

[0024] Among them, the pre-filtering process refers to the use of adaptive Wiener filtering to process the original acoustic emission signal. Taking the statistical characteristics of low-frequency background noise in the shot peening environment as a reference, the filtering parameters are dynamically adjusted to filter out mechanical vibration background interference with frequencies below 10kHz while retaining the effective impact sound signal characteristics. This avoids the problem of filtering out useful low-frequency feature components or failing to completely remove background noise by using fixed filtering parameters, and provides a processed signal with a higher signal-to-noise ratio for subsequent feature extraction.

[0025] Furthermore, the feature analysis of the acoustic emission signal specifically involves extracting the time-domain and frequency-domain statistical features of the filtered acoustic emission signal. The time-domain feature extraction specifically involves: counting the number of pulse oscillations exceeding a preset threshold to obtain the ring count; for example, counting the number of oscillation pulses exceeding twice the root mean square value of the background noise. Then, the time average of the squared signal is calculated to obtain the root mean square value of the signal. The frequency-domain feature extraction involves performing a fast Fourier transform on the processed signal to obtain the signal power spectrum. The centroid frequency of the spectrum is calculated from the power spectrum. For example, the centroid frequency is equal to the sum of the power spectrum amplitudes of each frequency point multiplied by the sum of the power spectrum amplitudes of all frequency points, and then divided by the sum of the power spectrum amplitudes of all frequency points. This weighted average of the power distribution yields the centroid frequency reflecting the overall distribution of signal energy. Then, the peak values ​​of the power spectrum are traversed to determine the frequency corresponding to the position of the maximum amplitude, which is the peak-to-peak frequency. Finally, the above four types of parameters are integrated to obtain the acoustic signal feature parameters that characterize the current processing state.

[0026] The Fast Fourier Transform (FFT) refers to the block transformation of a discrete-time acoustic emission signal of a certain length, calculating the corresponding power spectrum of each block, and then averaging the power spectra of multiple blocks to obtain the final stable power spectrum result, thereby reducing the impact of random fluctuations of a single block signal on feature calculation.

[0027] S30: By dynamically estimating the acoustic signal interference intensity, the acoustic signal characteristic parameters are adaptively corrected to obtain the corrected acoustic signal characteristic parameters; In this embodiment of the invention, there are multiple sources of interference at the shot peening site, such as equipment vibration and shot collision with equipment. The intensity of the interference changes dynamically with the processing progress. The directly extracted acoustic signal feature parameters will be deviated due to the interference. Therefore, it is necessary to first estimate the interference intensity of the current processing period and then adaptively correct the feature parameters to obtain the corrected acoustic signal feature parameters.

[0028] Specifically, step S30 in the method includes: The ratio of low-frequency energy to high-frequency energy in the acoustic emission signal is calculated in real time and denoted as the frequency band energy ratio. The dynamic baseline of the frequency band energy ratio is updated using an exponentially weighted sliding window, wherein when a sudden change in the frequency band energy ratio is detected, the update of the dynamic baseline is temporarily frozen; Calculate the relative deviation between the current frequency band energy ratio and the dynamic baseline; Calculate the first time derivative of the energy ratio of the frequency band as the rate of change of the energy ratio of the frequency band; The relative deviation and the rate of change are weighted and summed, and the result of the weighted sum is used as the real-time interference intensity value. The acoustic signal characteristic parameters are adaptively corrected based on the real-time interference intensity value to obtain the corrected acoustic signal characteristic parameters.

[0029] In embodiments of the present invention, such as Figure 2 As shown, firstly, the interference energy generated by equipment vibration and shot collision with the tooling during shot peening is mainly concentrated in the low-frequency band, while the effective acoustic emission energy generated by shot impact on the raceway surface is concentrated in the mid-to-high frequency band. Therefore, the ratio of low-frequency energy to high-frequency energy can directly reflect the relative intensity of the current environmental interference. For example, in this embodiment, the region below 10kHz is divided into the low-frequency band, and the region above 30kHz is divided into the high-frequency band. The power spectral amplitude values ​​in the two frequency bands are accumulated to obtain the corresponding energy, and then the ratio of the two is calculated to obtain the frequency band energy ratio.

[0030] Secondly, an exponentially weighted sliding window is used to update the dynamic baseline for frequency band energy ratios. The dynamic baseline corresponds to the reference frequency band energy ratio level under interference-free or stable weak interference conditions. When a sudden change in the frequency band energy ratio is detected, it indicates that a strong instantaneous interference has entered the system. At this point, the dynamic baseline update is frozen to prevent the instantaneous interference from raising the baseline and causing subsequent interference intensity estimation errors. The exponentially weighted sliding window assigns different weights to historical frequency band energy ratios within the sliding window, with new data having a higher weight and older data having an exponentially decreasing weight. This approach can track the slow changes in environmental interference while preserving the stability of the baseline and preventing large fluctuations in the baseline due to instantaneous interference.

[0031] Then, by combining the relative deviation between the frequency band energy ratio and the dynamic baseline with the rate of change of the frequency band energy ratio, the real-time interference intensity value is obtained by weighting. The larger the deviation and the faster the change, the stronger the current interference. The weighting coefficient can be calibrated according to the on-site environment. For example, the relative deviation weight is 0.7 and the rate of change weight is 0.3. The combined real-time interference intensity value can accurately reflect the degree of influence of the interference on the acoustic signal characteristics during the current processing period.

[0032] Finally, the original acoustic signal characteristic parameters are corrected based on the real-time interference intensity value. The greater the interference intensity, the more interference components are mixed into the original characteristic parameters. During the correction, the interference intensity is used as the weight to offset the original characteristic parameters and obtain the corrected acoustic signal characteristic parameters after eliminating the interference effect.

[0033] The process of adaptively correcting the acoustic signal feature parameters based on the real-time interference intensity value to obtain the corrected acoustic signal feature parameters includes: A feature parameter correction plugin is established, and the model parameters of the feature parameter correction plugin are updated online using a recursive filtering algorithm; Based on the updated model parameters and the real-time disturbance intensity value at the current moment, calculate the prediction correction amount of each feature parameter at the current moment; The acoustic signal feature parameters are corrected based on the predicted correction amount of each feature parameter, and the corrected acoustic signal feature parameters are output.

[0034] In this embodiment of the invention, firstly, a feature parameter correction plugin is established, with real-time interference intensity value as input and predicted correction amount corresponding to each feature parameter as output. The initial parameters of the model are preset according to the correspondence between interference intensity and feature deviation calibrated offline.

[0035] Specifically, the feature parameter correction plugin is a linear mapping model based on the interference intensity feature deviation. The model input is a one-dimensional real-time interference intensity value, and the output is the correction amount corresponding to four types of feature parameters: spectral centroid frequency, spectral peak-to-peak frequency, ringing count, and root mean square value. The model structure is simple and the computational load is small, which can meet the real-time requirements of online calculation in the processing process without occupying additional computing resources and delaying the processing process.

[0036] Secondly, a recursive filtering algorithm is used to update the model parameters online. After each processing and inspection cycle, the model parameters are corrected in reverse based on the residual stress prediction error corresponding to the currently corrected feature parameters. As the processing progresses, the accuracy of the corrected model gradually improves, better adapting to the interference characteristics of the current processing environment. Specifically, the recursive filtering algorithm can use the recursive least squares method. After each round of signal acquisition and feature extraction, the predicted residual stress obtained by inputting the current corrected feature parameters into the residual stress prediction model is subtracted from the offline calibrated standard residual stress value under the processing parameters to obtain the prediction error. Then, the recursive least squares method is used to update the model's mapping coefficients based on the prediction error, realizing online adaptive optimization of the corrected model and adapting to the changes in interference characteristics of different processing sites.

[0037] Finally, based on the updated model, the predicted corrections for each acoustic signal characteristic parameter at the current moment are calculated, namely, the predicted correction for centroid frequency, the predicted correction for peak-to-peak frequency, the predicted correction for root mean square value, and the predicted correction for ring count. Subtracting the corresponding predicted corrections from the original acoustic signal characteristic parameters yields the corrected acoustic signal characteristic parameters after eliminating interference bias, providing more accurate input data for subsequent residual stress prediction.

[0038] Furthermore, a feature parameter correction plugin is established, and a recursive filtering algorithm is used to update the model parameters of the feature parameter correction plugin online, including: Initial values ​​of the model parameter vector were obtained through offline calibration experiments; During the shot peening process, the on / off control signals of the shot peening valve of the shot peening equipment are acquired in real time; When the shot peening valve is detected to switch from open to closed, acoustic emission signals are collected within a preset collection time after the valve is closed. The signals collected during this period are used as pure interference signals, the corresponding real-time interference intensity values ​​are recorded, and the actual changes of each characteristic parameter relative to the reference interference-free state are calculated. Multiply the model parameter vector stored in the previous time step with the real-time disturbance intensity value at the current time step to calculate the prediction correction amount of each feature parameter at the current time step. The prediction deviation is calculated by subtracting the prediction correction from the actual change. Set a recursive forgetting factor, substitute the recursive forgetting factor into the gain calculation formula of the recursive filtering algorithm, and calculate the gain matrix at the current time. Multiply the gain matrix by the prediction bias to obtain the model parameter adjustment amount; The model parameter vector updated at the current time step is obtained by adding the model parameter adjustment amount to the model parameter vector at the previous time step.

[0039] In this embodiment of the invention, firstly, through offline calibration experiments, under the condition of no projectile impact on the raceway and only equipment interference, the deviations of characteristic parameters corresponding to different interference intensities are collected, and an initial model parameter vector is obtained by fitting, ensuring that the model has basic interference correction capabilities in its initial state. Specifically, the offline calibration experiment refers to performing air jetting under conditions of no workpiece or obstructed raceway, collecting acoustic emission signals under pure interference conditions, and statistically analyzing the proportional relationship between the pure interference intensity value and the changes in each characteristic parameter. This proportional relationship is used as the initial values ​​for the centroid frequency correction coefficient, spectral peak-to-peak frequency correction coefficient, root mean square value correction coefficient, and ringing count correction coefficient. The model parameter vector is composed of these four correction coefficients.

[0040] Secondly, during the shot peening process, the start-up and shutdown of the equipment and the opening and closing of the shot peening valve are clear and obtainable status signals. When the shot peening valve is closed, there are no more effective signals generated by the impact of the shot on the raceway. At this time, the acoustic emission signals collected are all generated by interference sources such as the vibration of the equipment itself and the impact of the shot on the tooling. They are natural pure interference samples. There is no need to set up a separate calibration process to obtain samples online, which will not take up extra processing time.

[0041] Furthermore, after identifying the moment of pure interference, the current real-time interference intensity value is recorded. At the same time, the actual changes of the currently extracted feature parameters relative to the feature parameters of the reference interference-free state are calculated, including the actual decrease in centroid frequency, the actual decrease in peak frequency, the actual increase in root mean square value, and the actual increase in ring count. The prediction correction of the current feature parameters is then calculated by combining the model parameters stored in the previous moment. The prediction deviation is obtained by subtracting the prediction correction from the actual change.

[0042] Then, a recursive forgetting factor is introduced. An appropriate forgetting factor, such as 0.95, is set in the recursive filtering algorithm to give new data higher weight, reduce the impact of old historical data on the model, and allow the model to adapt to the slow changes in environmental interference characteristics. The gain matrix at the current time is then calculated by substituting it into the gain formula. The gain matrix is ​​multiplied by the prediction bias to obtain the model parameter adjustment amount. Finally, the adjustment amount is superimposed on the model parameter vector at the previous time to complete the update of the model parameters at the current time. This allows the corrected model to continuously optimize the correction accuracy as the processing progresses and continuously and accurately eliminate the influence of interference on the feature parameters.

[0043] S40: Using the raceway curvature radius and the modified acoustic signal characteristic parameters as input, the predicted residual stress field characteristics are obtained through model reasoning, and the predicted contact fatigue strength is obtained by mapping. In this embodiment of the invention, the curvature radius of the raceway itself affects the incident angle and contact area of ​​the projectile during shot peening, thereby affecting the distribution of residual stress on the surface. Therefore, by inputting the curvature radius of the raceway and the corrected four types of acoustic signal characteristic parameters into the pre-trained prediction model, the predicted residual stress field characteristics of the current raceway surface can be inferred. Specifically, these include two core features: the maximum residual compressive stress value and the depth of the residual compressive stress layer. Then, through the pre-calibrated mapping relationship between the residual stress field characteristics and the contact fatigue strength, the predicted contact fatigue strength corresponding to the current processing state can be directly obtained.

[0044] The prediction model employs a neural network, with five-dimensional input variables: raceway curvature radius, corrected spectral centroid frequency, peak-to-peak frequency, ring count, and root mean square value. The output is a two-dimensional residual stress field characteristic. The model was trained using extensive offline shot peening experimental data, accurately establishing a nonlinear mapping relationship between the input parameters and the residual stress field characteristic. The mapping relationship between the residual stress field characteristic and contact fatigue strength was calibrated through material fatigue tests. Different combinations of maximum residual compressive stress and compressive stress layer depth correspond to different contact fatigue lives, and the corresponding predicted contact fatigue strength can be directly obtained from the calibrated mapping table.

[0045] Specifically, step S40 in the method includes: The sample training dataset was collected through offline calibration experiments of bearing raceways of the same material. The sample training data includes the sample raceway curvature radius, sample acoustic signal characteristic parameters, and sample residual stress field characteristics. The sample residual stress field characteristics include residual stress peak value and residual stress depth. The bearing raceway of the same material refers to the bearing raceway that has the same grade of steel as the bearing raceway to be processed and has undergone the same pre-heat treatment. Using the sample raceway curvature radius and the sample acoustic signal characteristic parameters as inputs, and the sample residual stress field characteristics as outputs, a neural network model is trained to construct a residual stress field characteristic predictor; The raceway curvature radius and the corrected acoustic signal characteristic parameters are input into the residual stress field characteristic predictor, and the predicted residual stress field characteristics are output. The predicted contact fatigue strength is obtained based on the predicted residual stress field feature mapping.

[0046] In this embodiment of the invention, firstly, multiple sets of offline shot peening experiments are carried out using bearing raceways of the same material and specification range. Acoustic emission signals are collected for different shot peening parameters and different interference intensities, and the corresponding acoustic signal features are extracted. At the same time, the residual stress peak value and residual stress depth of the raceway after processing are measured using professional residual stress detection equipment. A training dataset containing samples of different working conditions is obtained to ensure that the dataset covers various interference situations and processing parameter ranges that may occur on the processing site.

[0047] Secondly, the neural network model is initialized and end-to-end training is performed using the training dataset to reduce the mean square error between the predicted residual stress field features and the measured values. Training is stopped when the error of the model on the validation set is lower than a preset threshold, thus obtaining a usable residual stress field feature predictor.

[0048] Finally, the roller curvature radius and corrected acoustic signal characteristic parameters obtained from the actual online acquisition are input into the trained residual stress field characteristic predictor, which can quickly output the predicted residual stress peak value and residual stress depth of the currently processed roller. Then, the corresponding predicted contact fatigue strength value can be directly obtained by querying through the pre-calibrated mapping relationship.

[0049] For example, a residual stress field feature predictor is constructed based on a neural network model, with the following specific steps: A three-layer feedforward neural network is used as the predictor model. The input layer has 5 neurons, corresponding to 5 input features; the hidden layer has 12 neurons, and the activation function is the ReLU function f(x)=max(0,x) to introduce nonlinear mapping capability; the output layer has 2 neurons, corresponding to 2 residual stress field feature outputs, and the output layer activation function is the linear function f(x)=x to allow the output value to vary within a continuous real number range. The Xavier uniform initialization method is used to initialize the weight matrix and bias vector of the network to keep the variance of the output of each layer consistent and avoid gradient vanishing or exploding. The initial learning rate is set to 0.001, the batch size is 32, the maximum training epochs are 500, and an early stopping mechanism is set: training stops when the validation set loss does not decrease for 20 consecutive epochs, and the result is saved as the final usable residual stress field feature predictor.

[0050] Further, obtaining the predicted contact fatigue strength based on the predicted residual stress field characteristic mapping includes: Under the premise of keeping the process parameters such as shot peening medium type and particle size, shot peening intensity, coverage and initial surface condition constant, a mapping relationship database between residual stress field characteristics and contact fatigue strength is established. The mapping relationship database is obtained through bench fatigue tests on bearing raceways of the same material. Specifically, under fixed process parameters, bench fatigue tests are conducted on multiple groups of bearing raceway samples with different residual stress peak values ​​and different residual stress depths. The number of cycles when each group of samples reaches fatigue spalling is recorded, and the number of cycles is used as the relative measure of contact fatigue strength under the fixed process parameters. The predicted residual stress peak value, the predicted residual stress depth, and the current shot peening process parameters are used together as joint query keys to perform matching and retrieval in the mapping relationship database. When a mapping record with the same key value as the joint query is found, the corresponding relative measure of contact fatigue strength is directly read as the predicted contact fatigue strength. The predicted contact fatigue strength is the potential contact fatigue strength that the raceway can achieve based on the residual stress field under the condition that the current shot peening process parameters remain unchanged. When no identical mapping record is found, an interpolation method is used to interpolate the data points adjacent to the joint query key value in the mapping relationship database to obtain the corresponding predicted contact fatigue strength value.

[0051] In this embodiment of the invention, firstly, while keeping the process parameters such as shot peening medium type and particle size, shot peening intensity, coverage, and initial surface condition constant, fatigue tests on bearing raceways under different residual stress states are conducted to obtain sufficient test samples to construct a mapping database. Specifically, bearing raceway samples of the same material and heat treatment state as the bearing to be processed are selected, and multiple sets of samples with different residual stress peak values ​​and residual stress depths are prepared using different shot peening process parameters.

[0052] Specifically, each group of samples consists of no fewer than three pieces. The raceway surface of each sample is measured at multiple points using an X-ray residual stress tester. The average value is taken as the peak residual stress σ of the sample group, in MPa. A negative value indicates compressive stress. The surface material is peeled off layer by layer using an electrolytic polishing peeling method combined with X-ray testing, and the residual stress of each layer is measured until the measured residual stress value is close to the stress level of the matrix. The total thickness of the residual compressive stress layer is taken as the residual stress depth D, in μm.

[0053] Then, each group of specimens was installed on a bench fatigue testing machine and subjected to rolling contact fatigue test under specified contact load conditions. The appearance of fatigue spalling pits with a diameter greater than 1 mm on the raceway surface of the specimen was used as the failure criterion. The number of cycles N tens of thousands from the start of the test to failure of each group of specimens was recorded. The number of cycles was used as the contact fatigue strength measure corresponding to the residual stress state. (σ, D, Nf) was stored in the database as a mapping record to form a database of the mapping relationship between residual stress field characteristics and contact fatigue strength.

[0054] For example, the residual stress peak is -600MPa, the residual stress depth is 50μm, and the contact fatigue strength is 12 million cycles; the residual stress peak is -580MPa, the residual stress depth is 48μm, and the contact fatigue strength is 11.5 million cycles.

[0055] Secondly, during online machining, after obtaining the predicted residual stress peak value and predicted residual stress depth of the current raceway through the residual stress field feature predictor, these values, combined with the current shot peening process parameters, are used as joint query keys for matching and retrieval in the mapping relationship database. The retrieval rules are as follows: First, an exact match is performed, i.e., it is searched to see if there is a mapping record whose residual stress peak value is equal to the predicted residual stress peak value and whose residual stress depth is equal to the predicted residual stress depth; if such a record exists, the contact fatigue strength value corresponding to that record is directly read as the predicted contact fatigue strength output. The predicted contact fatigue strength is the potential contact fatigue strength that the raceway can achieve based on the residual stress field prediction, under the condition that the current shot peening process parameters remain unchanged.

[0056] When no identical mapping record is found, the bilinear interpolation method is used to interpolate adjacent data points in the database. The specific interpolation process is as follows: the database is searched for the two residual stress peaks that are closest to the predicted residual stress peak and the two residual stress depths that are closest to the predicted residual stress depth. Four neighboring data points are located, and then interpolation is performed to obtain the predicted contact fatigue strength value corresponding to the current predicted residual stress field characteristics.

[0057] S50: With the goal of making the predicted contact fatigue strength approach the preset contact fatigue strength index, shot peening impact processing control is performed.

[0058] In this embodiment of the invention, the predicted contact fatigue strength obtained in real time during the current shot peening process is compared with the preset target index. The shot peening process parameters are adjusted according to the deviation between the two to achieve closed-loop adaptive control of the processing process. This ensures that the contact fatigue strength of the bearing raceway obtained in the final processing meets the design requirements, avoiding insufficient processing leading to substandard strength, or over-processing leading to cost waste and performance degradation.

[0059] Specifically, if the current predicted contact fatigue strength is lower than the preset target, the shot peening intensity or shot peening time will be increased according to the deviation, thereby increasing the residual compressive stress level on the raceway surface and increasing the depth of the residual compressive stress layer, thus further improving the contact fatigue strength; if the current predicted contact fatigue strength has reached or slightly exceeded the preset target, the current process parameters will be maintained to complete the processing, or the shot peening process will be ended in advance, controlling the processing cost while meeting the performance requirements.

[0060] For example, the preset contact fatigue strength is determined based on the bearing model, operating conditions and design life requirements. It is pre-entered into the processing control system before processing. During processing, the predicted contact fatigue strength is updated once after each detection cycle is completed, and the shot peening process parameters are adjusted accordingly. Processing is stopped when the predicted value reaches the preset index range. This maximizes processing efficiency and reduces processing costs while ensuring that the strength meets the standard.

[0061] In summary, compared to existing technologies, this invention extracts features by monitoring acoustic emission signals online and combines this with real-time prediction of residual stress field and contact fatigue strength based on raceway parameters. This solves the problem that traditional processing methods cannot obtain raceway contact fatigue strength online and can only rely on post-processing sampling. It enables closed-loop adaptive control of the processing, effectively ensuring the stability of contact fatigue strength after shot peening of bearing raceways, reducing the defect rate, and avoiding cost waste caused by over-processing. Furthermore, online recursive filtering achieves adaptive correction of interference features, adapting to changes in interference characteristics at different processing sites. This effectively reduces the impact of environmental interference such as equipment vibration on acoustic emission signal feature extraction, improves the accuracy of residual stress and fatigue strength prediction, and meets the actual needs of batch processing.

[0062] In summary, the embodiments of the present invention have at least the following technical effects: This invention provides a processing method to improve the contact fatigue strength of bearing raceways. First, broadband acoustic emission signals are acquired in real time during shot peening of the bearing raceways. The acoustic signal characteristic parameters are adaptively corrected by dynamically estimating the interference intensity of the processing environment, effectively reducing the interference of the complex shot peening environment on detection accuracy and obtaining more accurate characteristic signals. Second, the raceway curvature radius and the corrected acoustic signal characteristics are used together as model inputs. The influence of different raceway curvatures on acoustic signal propagation attenuation and residual stress field distribution is considered, improving the prediction accuracy of residual stress field characteristics and contact fatigue strength. Finally, the shot peening process is controlled through a closed-loop control of the prediction results, ensuring that the contact fatigue strength of the processed raceway consistently reaches the preset target, avoiding problems of insufficient or excessive strengthening.

[0063] Through the above technical solution, this invention solves the technical problems in the existing shot peening process, such as the inability to perceive the formation state of residual stress and contact fatigue strength of the raceway in real time, the low signal-to-noise ratio of the detection signal due to interference from complex processing environment, large errors in characteristic parameters, and the failure to consider the influence of raceway curvature on detection accuracy and residual stress field, resulting in insufficient accuracy in predicting contact fatigue strength and poor processing stability. It can realize adaptive closed-loop control of the shot peening process, effectively ensuring that the contact fatigue strength of the bearing raceway after processing is stable and meets the preset index requirements, and is suitable for the processing needs of long-life bearings in the high-end equipment field.

[0064] Example 2, as Figure 3 As shown, based on the same inventive concept as the processing method for improving the contact fatigue strength of bearing raceways provided in Embodiment 1, this embodiment of the invention also provides a processing apparatus for improving the contact fatigue strength of bearing raceways, comprising: The acoustic emission signal acquisition module 11 is used to install a broadband acoustic emission sensor on the sound transmission path that is rigidly connected to the raceway during shot peening impact processing of the bearing raceway, and to acquire the acoustic emission signal in real time during the shot peening process. The acoustic signal feature parameter acquisition module 12 is used to perform feature analysis on the acoustic emission signal and acquire acoustic signal feature parameters. The acoustic signal feature parameter correction module 13 is used to adaptively correct the acoustic signal feature parameters by dynamically estimating the acoustic signal interference intensity, and to obtain the corrected acoustic signal feature parameters. The prediction model training module 14 is used to obtain the predicted residual stress field characteristics through model inference by taking the raceway curvature radius and the modified acoustic signal characteristic parameters as input, and to map and obtain the predicted contact fatigue strength. The shot peening impact execution module 15 is used to control the shot peening impact process with the goal of making the predicted contact fatigue strength approach the preset contact fatigue strength index.

[0065] Furthermore, in one embodiment of the invention, a broadband acoustic emission sensor is installed on the sound transmission path rigidly connected to the raceway, comprising: The wideband acoustic emission sensor operates in the frequency range of 100kHz to 1MHz and is mounted on a component that forms a rigid acoustic transmission path with the bearing raceway. For the outer raceway, the broadband acoustic emission sensor is fixedly installed on the outer cylindrical surface of the bearing outer ring, with the sensitive surface facing the bottom area of ​​the raceway. For the inner raceway, the broadband acoustic emission sensor is mounted on the mandrel or fixture, and signal transmission is achieved through rotational coupling with the inner raceway.

[0066] Furthermore, the acoustic emission signals during the shot peening process are acquired in real time, including: During the first preset time period after the shot peening impact process is started, acoustic emission signals are collected at the first sampling frequency, or the process is in a state of waiting to be triggered for collection. When the processing time exceeds the first preset time period, the acoustic emission signal is continuously collected at a second sampling frequency higher than the first sampling frequency; During the continuous acquisition at the second sampling frequency, the acoustic emission signal is monitored in real time, and only the acoustic signal characteristic parameters are recorded. When the root mean square value of the acoustic emission signal or the ring count exceeds the preset trigger threshold, the acoustic emission waveform data within the preset time period before and after the current moment is fully recorded.

[0067] Furthermore, the acoustic signal characteristic parameters include the centroid frequency of the spectrum, the peak-to-peak frequency of the spectrum, the ringing count, and the root mean square value.

[0068] In one embodiment, the acoustic signal feature parameter correction module 13 is specifically used for: The ratio of low-frequency energy to high-frequency energy in the acoustic emission signal is calculated in real time and denoted as the frequency band energy ratio. The dynamic baseline of the frequency band energy ratio is updated using an exponentially weighted sliding window, wherein when a sudden change in the frequency band energy ratio is detected, the update of the dynamic baseline is temporarily frozen; Calculate the relative deviation between the current frequency band energy ratio and the dynamic baseline; Calculate the first time derivative of the energy ratio of the frequency band as the rate of change of the energy ratio of the frequency band; The relative deviation and the rate of change are weighted and summed, and the result of the weighted sum is used as the real-time interference intensity value. The acoustic signal characteristic parameters are adaptively corrected based on the real-time interference intensity value to obtain the corrected acoustic signal characteristic parameters.

[0069] Further, the acoustic signal feature parameters are adaptively corrected based on the real-time interference intensity value to obtain the corrected acoustic signal feature parameters, including: A feature parameter correction plugin is established, and the model parameters of the feature parameter correction plugin are updated online using a recursive filtering algorithm; Based on the updated model parameters and the real-time disturbance intensity value at the current moment, calculate the prediction correction amount of each feature parameter at the current moment; The acoustic signal feature parameters are corrected based on the predicted correction amount of each feature parameter, and the corrected acoustic signal feature parameters are output.

[0070] Furthermore, a feature parameter correction plugin is established, and a recursive filtering algorithm is used to update the model parameters of the feature parameter correction plugin online, including: Initial values ​​of the model parameter vector were obtained through offline calibration experiments; During the shot peening process, the on / off control signals of the shot peening valve of the shot peening equipment are acquired in real time; When the shot peening valve is detected to switch from open to closed, acoustic emission signals are collected within a preset collection time after the valve is closed. The signals collected during this period are used as pure interference signals, the corresponding real-time interference intensity values ​​are recorded, and the actual changes of each characteristic parameter relative to the reference interference-free state are calculated. Multiply the model parameter vector stored in the previous time step with the real-time disturbance intensity value at the current time step to calculate the prediction correction amount of each feature parameter at the current time step. The prediction deviation is calculated by subtracting the prediction correction from the actual change. Set a recursive forgetting factor, substitute the recursive forgetting factor into the gain calculation formula of the recursive filtering algorithm, and calculate the gain matrix at the current time. Multiply the gain matrix by the prediction bias to obtain the model parameter adjustment amount; The model parameter vector updated at the current time step is obtained by adding the model parameter adjustment amount to the model parameter vector at the previous time step.

[0071] In one embodiment, the prediction model training module 14 is specifically used for: The sample training dataset was collected through offline calibration experiments of bearing raceways of the same material. The sample training data includes the sample raceway curvature radius, sample acoustic signal characteristic parameters, and sample residual stress field characteristics. The sample residual stress field characteristics include residual stress peak value and residual stress depth. The bearing raceway of the same material refers to the bearing raceway that has the same grade of steel as the bearing raceway to be processed and has undergone the same pre-heat treatment. Using the sample raceway curvature radius and the sample acoustic signal characteristic parameters as inputs, and the sample residual stress field characteristics as outputs, a neural network model is trained to construct a residual stress field characteristic predictor; The raceway curvature radius and the corrected acoustic signal characteristic parameters are input into the residual stress field characteristic predictor, and the predicted residual stress field characteristics are output. The predicted contact fatigue strength is obtained based on the predicted residual stress field feature mapping.

[0072] Further, in one embodiment of the invention, obtaining the predicted contact fatigue strength based on the predicted residual stress field characteristic mapping includes: Under the premise of keeping the process parameters such as shot peening medium type and particle size, shot peening intensity, coverage and initial surface condition constant, a mapping relationship database between residual stress field characteristics and contact fatigue strength is established. The mapping relationship database is obtained through bench fatigue tests on bearing raceways of the same material. Specifically, under fixed process parameters, bench fatigue tests are conducted on multiple groups of bearing raceway samples with different residual stress peak values ​​and different residual stress depths. The number of cycles when each group of samples reaches fatigue spalling is recorded, and the number of cycles is used as the relative measure of contact fatigue strength under the fixed process parameters. The predicted residual stress peak value, the predicted residual stress depth, and the current shot peening process parameters are used together as joint query keys to perform matching and retrieval in the mapping relationship database. When a mapping record with the same key value as the joint query is found, the corresponding relative measure of contact fatigue strength is directly read as the predicted contact fatigue strength. The predicted contact fatigue strength is the potential contact fatigue strength that the raceway can achieve based on the residual stress field under the condition that the current shot peening process parameters remain unchanged. When no identical mapping record is found, an interpolation method is used to interpolate the data points adjacent to the joint query key value in the mapping relationship database to obtain the corresponding predicted contact fatigue strength value.

Claims

1. A machining method for improving the contact fatigue strength of bearing raceways, characterized in that the method... include: When shot peening the bearing raceway, a broadband acoustic emission sensor is installed on the sound transmission path that is rigidly connected to the raceway to collect the acoustic emission signals in real time during the shot peening process. The acoustic emission signal is analyzed to obtain acoustic signal feature parameters; By dynamically estimating the acoustic signal interference intensity, the acoustic signal characteristic parameters are adaptively corrected to obtain the corrected acoustic signal characteristic parameters, including: The ratio of low-frequency energy to high-frequency energy in the acoustic emission signal is calculated in real time and denoted as the frequency band energy ratio. The dynamic baseline of the frequency band energy ratio is updated using an exponentially weighted sliding window, wherein when a sudden change in the frequency band energy ratio is detected, the update of the dynamic baseline is temporarily frozen; Calculate the relative deviation between the current frequency band energy ratio and the dynamic baseline; Calculate the first time derivative of the energy ratio of the frequency band as the rate of change of the energy ratio of the frequency band; The relative deviation and the rate of change are weighted and summed, and the result of the weighted sum is used as the real-time interference intensity value. The acoustic signal feature parameters are adaptively corrected based on the real-time interference intensity value to obtain the corrected acoustic signal feature parameters. Using the raceway curvature radius and the modified acoustic signal characteristic parameters as input, the predicted residual stress field characteristics are obtained through model reasoning, and the predicted contact fatigue strength is obtained by mapping. The shot peening impact processing is controlled with the goal of making the predicted contact fatigue strength approach the preset contact fatigue strength index.

2. The processing method for improving the contact fatigue strength of bearing raceways according to claim 1, characterized in that, A broadband acoustic emission sensor is installed on the sound transmission path that is rigidly connected to the raceway, including: The wideband acoustic emission sensor operates in the frequency range of 100kHz to 1MHz and is mounted on a component that forms a rigid acoustic transmission path with the bearing raceway. For the outer raceway, the broadband acoustic emission sensor is fixedly installed on the outer cylindrical surface of the bearing outer ring, with the sensitive surface facing the bottom area of ​​the raceway. For the inner raceway, the broadband acoustic emission sensor is mounted on the mandrel or fixture, and signal transmission is achieved through rotational coupling with the inner raceway.

3. The processing method for improving the contact fatigue strength of bearing raceways according to claim 1, characterized in that, Real-time acquisition of acoustic emission signals during shot peening, including: During the first preset time period after the shot peening impact process is started, acoustic emission signals are collected at the first sampling frequency, or the process is in a state of waiting to be triggered for collection. When the processing time exceeds the first preset time period, the acoustic emission signal is continuously collected at a second sampling frequency higher than the first sampling frequency; During the continuous acquisition at the second sampling frequency, the acoustic emission signal is monitored in real time, and only the acoustic signal characteristic parameters are recorded. When the root mean square value of the acoustic emission signal or the ring count exceeds the preset trigger threshold, the acoustic emission waveform data within the preset time period before and after the current moment is fully recorded.

4. The processing method for improving the contact fatigue strength of bearing raceways according to claim 1, characterized in that, The acoustic emission signal is subjected to feature analysis to obtain acoustic signal feature parameters, wherein the acoustic signal feature parameters include the centroid frequency of the spectrum, the peak-to-peak frequency of the spectrum, the ringing count, and the root mean square value.

5. The processing method for improving the contact fatigue strength of bearing raceways according to claim 1, characterized in that, The acoustic signal feature parameters are adaptively corrected based on the real-time interference intensity value to obtain the corrected acoustic signal feature parameters, including: A feature parameter correction plugin is established, and the model parameters of the feature parameter correction plugin are updated online using a recursive filtering algorithm; Based on the updated model parameters and the real-time disturbance intensity value at the current moment, calculate the prediction correction amount of each feature parameter at the current moment; The acoustic signal feature parameters are corrected based on the predicted correction amount of each feature parameter, and the corrected acoustic signal feature parameters are output.

6. The processing method for improving the contact fatigue strength of bearing raceways according to claim 5, characterized in that, A feature parameter correction plugin is established, and the model parameters of the feature parameter correction plugin are updated online using a recursive filtering algorithm, including: Initial values ​​of the model parameter vector were obtained through offline calibration experiments; During the shot peening process, the on / off control signals of the shot peening valve of the shot peening equipment are acquired in real time; When the shot peening valve is detected to switch from open to closed, acoustic emission signals are collected within a preset collection time after the valve is closed. The signals collected during this period are used as pure interference signals, the corresponding real-time interference intensity values ​​are recorded, and the actual changes of each characteristic parameter relative to the reference interference-free state are calculated. Multiply the model parameter vector stored in the previous time step with the real-time disturbance intensity value at the current time step to calculate the prediction correction amount of each feature parameter at the current time step. The prediction deviation is calculated by subtracting the prediction correction from the actual change. Set a recursive forgetting factor, substitute the recursive forgetting factor into the gain calculation formula of the recursive filtering algorithm, and calculate the gain matrix at the current time. Multiply the gain matrix by the prediction bias to obtain the model parameter adjustment amount; The model parameter vector updated at the current time step is obtained by adding the model parameter adjustment amount to the model parameter vector at the previous time step.

7. The processing method for improving the contact fatigue strength of bearing raceways according to claim 1, characterized in that, Using the raceway curvature radius and the modified acoustic signal characteristic parameters as input, the predicted residual stress field characteristics are obtained through model inference, and the predicted contact fatigue strength is obtained by mapping, including: The sample training dataset was collected through offline calibration experiments of bearing raceways of the same material. The sample training data includes the sample raceway curvature radius, sample acoustic signal characteristic parameters, and sample residual stress field characteristics. The sample residual stress field characteristics include residual stress peak value and residual stress depth. The bearing raceway of the same material refers to the bearing raceway that has the same grade of steel as the bearing raceway to be processed and has undergone the same pre-heat treatment. Using the sample raceway curvature radius and the sample acoustic signal characteristic parameters as inputs, and the sample residual stress field characteristics as outputs, a neural network model is trained to construct a residual stress field characteristic predictor; The raceway curvature radius and the corrected acoustic signal characteristic parameters are input into the residual stress field characteristic predictor, and the predicted residual stress field characteristics are output. The predicted contact fatigue strength is obtained based on the predicted residual stress field feature mapping.

8. The processing method for improving the contact fatigue strength of bearing raceways according to claim 7, characterized in that, The predicted contact fatigue strength is obtained based on the predicted residual stress field feature mapping, including: Under the premise of keeping the process parameters such as shot peening medium type and particle size, shot peening intensity, coverage and initial surface condition constant, a mapping relationship database between residual stress field characteristics and contact fatigue strength is established. The mapping relationship database is obtained through bench fatigue tests on bearing raceways of the same material. Specifically, under fixed process parameters, bench fatigue tests are conducted on multiple groups of bearing raceway samples with different residual stress peak values ​​and different residual stress depths. The number of cycles when each group of samples reaches fatigue spalling is recorded, and the number of cycles is used as the relative measure of contact fatigue strength under the fixed process parameters. The predicted residual stress peak value, the predicted residual stress depth, and the current shot peening process parameters are used together as joint query keys to perform matching and retrieval in the mapping relationship database. When a mapping record with the same key value as the joint query is found, the corresponding relative measure of contact fatigue strength is directly read as the predicted contact fatigue strength. The predicted contact fatigue strength is the potential contact fatigue strength that the raceway can achieve based on the residual stress field under the condition that the current shot peening process parameters remain unchanged. When no identical mapping record is found, an interpolation method is used to interpolate the data points adjacent to the joint query key value in the mapping relationship database to obtain the corresponding predicted contact fatigue strength value.

9. A processing device for improving the contact fatigue strength of bearing raceways, characterized in that, A processing method for improving the contact fatigue strength of a bearing raceway as described in any one of claims 1-8, comprising: The acoustic emission signal acquisition module is used to install a broadband acoustic emission sensor on the sound transmission path that is rigidly connected to the raceway during shot peening of the bearing raceway, and to acquire the acoustic emission signal in real time during the shot peening process. The acoustic signal feature parameter acquisition module is used to perform feature analysis on the acoustic emission signal and acquire acoustic signal feature parameters; The acoustic signal feature parameter correction module is used to adaptively correct the acoustic signal feature parameters by dynamically estimating the acoustic signal interference intensity, and to obtain the corrected acoustic signal feature parameters. The prediction model training module is used to obtain the predicted residual stress field characteristics through model inference by taking the raceway curvature radius and the modified acoustic signal characteristic parameters as input, and to map and obtain the predicted contact fatigue strength. The shot peening impact execution module is used to control the shot peening impact process with the goal of making the predicted contact fatigue strength approach the preset contact fatigue strength index.

Images