Rolling bearing defect detection method and system based on encoder feedback

Abstract

The invention relates to the technical field of bearing detection, in particular to a rolling bearing defect detection method and system based on encoder feedback, and the method comprises the steps: collecting a vibration signal of a to-be-detected rolling bearing, and obtaining a rotating speed signal through a coaxial encoder; selecting a defect impact dominant frequency band for the vibration signal according to the spectral kurtosis and carrying out band-pass filtering, and then carrying out envelope demodulation on a filtered signal to obtain an envelope signal; generating a speed measurement pulse according to the rotating speed signal, determining an equal-angle sampling mark and establishing time-angle mapping; resampling the envelope signal angular domain according to the mark to form an angular domain envelope sequence; order spectrum analysis is carried out on the angular domain envelope sequence to obtain an envelope order spectrum, feature orders of an outer ring, an inner ring and a rolling body are determined according to bearing structure parameters, and harmonic peak values of the feature orders of the outer ring, the inner ring and the rolling body are extracted; and calculating three-class order peak ratio characteristics by using the full-spectrum amplitude and normalization, inputting the three-class order peak ratio characteristics into a multi-core support vector machine model, and outputting defect classes. According to the invention, the problems of difficult identification caused by feature drift of the bearing under a non-stationary condition and strong dependence of the diagnosis process on expert experience can be solved.

Description

Technical Field

[0001] This invention relates to the field of bearing inspection technology, and in particular to a method and system for detecting defects in rolling bearings based on encoder feedback. Background Technology

[0002] Rolling bearings, as critical components in rotating machinery, are typically assessed for their operational status through vibration monitoring. However, under non-stationary conditions such as variable speed, start-stop, acceleration, and deceleration, the statistical characteristics of the signal change over time. The characteristic components corresponding to defect excitations drift and exhibit broadening and crosstalk in traditional frequency domain representations, leading to a decrease in the accuracy of diagnostic methods based on the fixed-frequency assumption. While existing technologies employ time-frequency analysis, demodulation methods, and speed-related order tracking / synchronization processing to improve diagnosability under non-stationary conditions, many solutions rely on additional measurements or complex parameter settings, and the interpretation of results often requires experienced diagnostic personnel. Meanwhile, although machine learning can be used for automated identification, achieving stable differentiation of the same defect type under variable speed conditions still faces challenges such as insufficient feature transferability and poor robustness. Summary of the Invention

[0003] In view of the above technical problems, the present invention provides a rolling bearing defect detection method and system based on encoder feedback, which solves the problems of difficulty in identification caused by feature drift under non-stationary conditions and the strong dependence of the diagnostic process on expert experience.

[0004] Other features and advantages of this disclosure will become apparent from the following detailed description, or may be learned in part from practice of this disclosure.

[0005] According to one aspect of the present invention, a method for detecting defects in rolling bearings based on encoder feedback is proposed, the method comprising: The vibration signal of the rolling bearing under test is acquired, and the rotational speed signal coaxial with the rolling bearing under test is acquired based on the encoder; The vibration signal is preprocessed, including: determining the target frequency band where the defect impact is dominant based on spectral kurtosis; performing bandpass filtering on the vibration signal in the target frequency band to obtain a filtered vibration signal; and performing envelope demodulation on the filtered vibration signal to obtain an envelope signal. A speed measurement pulse sequence for angular domain resampling reference is generated from the speed signal, an equal angle sampling mark is determined based on the speed measurement pulse sequence, and a mapping relationship between the time axis and the rotation axis is established. The envelope signal is resampled in the angular domain according to the sampling marker to obtain an angular domain envelope sequence with equal angular intervals; The envelope sequence is subjected to order spectrum analysis to obtain the envelope order spectrum; the outer ring rolling element passing order, inner ring rolling element passing order, and rolling element spin order are calculated according to the structural parameters of the rolling bearing under test, and the peak values ​​corresponding to the outer ring rolling element passing order, inner ring rolling element passing order, and rolling element spin order and their harmonics are extracted from the envelope order spectrum. Calculate the peak value ratio characteristics of the outer ring order, the inner ring order, and the rolling element order, wherein: the total peak value sum of the envelope order spectrum is the sum of the amplitude values ​​of each order line in the envelope order spectrum; the peak value ratio characteristic of the outer ring order is the ratio of the peak value sum of the outer ring rolling element passing order and its harmonics to the total peak value sum of the envelope order spectrum; the peak value ratio characteristic of the inner ring order is the ratio of the peak value sum of the inner ring rolling element passing order and its harmonics to the total peak value sum of the envelope order spectrum; and the peak value ratio characteristic of the rolling element order is the ratio of the peak value sum of the rolling element spin order and its harmonics to the total peak value sum of the envelope order spectrum. The outer ring order peak ratio feature, the inner ring order peak ratio feature, and the rolling element order peak ratio feature are input into a trained multi-core support vector machine classification model to output the defect category of the rolling bearing under test.

[0006] Furthermore, determining the target frequency band based on spectral kurtosis includes: The vibration signal is decomposed to form a set of candidate frequency bands covering different center frequencies and different bandwidths; Spectral kurtosis indices for characterizing the significance of impact components are calculated on each set of candidate frequency bands, and kurtosis distributions are formed. Candidate frequency bands that satisfy threshold conditions or ranking conditions of spectral kurtosis indices are selected as target frequency bands according to the extreme value criteria of the spectral kurtosis indices. The passband parameters of the bandpass filter are constructed based on the center position and bandwidth range of the target frequency bands, thereby suppressing background noise and highlighting the impact components of defects.

[0007] Furthermore, envelope demodulation of the filtered vibration signal includes: converting the filtered vibration signal into an analytical signal and calculating the amplitude of the analytical signal to obtain an instantaneous amplitude sequence, or rectifying and smoothing the filtered vibration signal to obtain the instantaneous amplitude sequence; and performing low-pass filtering on the instantaneous amplitude sequence to remove the carrier frequency component, thereby obtaining the envelope signal.

[0008] Furthermore, generating the speed measurement pulse sequence from the speed signal includes: When the speed signal is output by an incremental encoder, edge detection or phase decoding is performed on the speed signal to obtain a pulse time stamp corresponding to the rotation angle increment of the shaft; Instantaneous rotational speed is calculated based on adjacent pulse time stamps, and a rotational speed signal is generated; Using the pulse time marker as a reference for the sampling marker, the sampling marker is adaptively updated as the rotational speed signal changes.

[0009] Furthermore, when performing corner domain resampling, the following steps are included: Determine the target turning angle sequence with equal angular intervals based on the sampling marks; The time position corresponding to each target rotation angle is obtained by using the mapping relationship between the time axis and the rotation axis; The envelope signal is interpolated and sampled at each of the aforementioned time positions to form the angular domain envelope sequence; wherein the interpolation sampling adopts linear interpolation, polynomial interpolation, spline interpolation or a combination thereof, and the envelope signal is subjected to detrending, endpoint extension or boundary smoothing processing before interpolation to reduce interpolation error, so that the defect impact presents an approximately equally spaced distribution in the angular domain envelope sequence.

[0010] Furthermore, based on the structural parameters of the rolling bearing under test, the passing order of the outer ring rolling element, the passing order of the inner ring rolling element, and the spin order of the rolling element are calculated, including: Obtain the number of rolling elements, rolling element diameter, pitch circle diameter, and contact angle parameters of the rolling bearing under test; Based on the number of rolling elements, the geometric ratio of the rolling element diameter to the pitch circle diameter, and the geometric correction term corresponding to the contact angle parameter, the passing order of the outer rolling element, the passing order of the inner rolling element, and the spin order of the rolling element are determined respectively; and integer multiples of each order are defined as the corresponding harmonic orders for peak extraction in the envelope order spectrum.

[0011] Furthermore, extracting the peak values ​​corresponding to the passing order of the outer rolling element, the passing order of the inner rolling element, and the spin order and harmonics of the rolling element from the envelope order spectrum includes: An order search window is set for each of the feature orders in the envelope order spectrum; Peak detection is performed within each of the order search windows to obtain the peak amplitude and the order position of the peak; when multi-peak competition is detected, a representative peak is selected based on the difference in peak amplitude, peak width, peak sharpness, or peak relative to noise floor. The representative peak value is corrected to make the obtained peak value more accurately reflect the energy accumulation of the defect impact in the order domain.

[0012] Furthermore, the multi-kernel support vector machine classification model is constructed in the following manner: Training data with known health status labels is collected, and the training data is collected under different speed change modes to cover non-stationary operating conditions; For each training sample, extract the peak order ratio features of the outer ring, the peak order ratio features of the inner ring, and the peak order ratio features of the rolling element to form a feature vector; The feature vectors are normalized or scaled; a support vector machine is trained using a kernel mapping formed by a weighted combination of multiple kernel functions to obtain a classification decision boundary that distinguishes between healthy state, outer ring defects, inner ring defects, rolling element defects and composite defects, and the classification decision boundary is solidified into the multi-kernel support vector machine classification model.

[0013] Furthermore, the method also includes: Each vibration signal is divided into multiple segments along the acquisition time or along the rotation angle sequence, and each segment is used as an independent sample and inherits the health status label corresponding to the vibration signal. The independent samples are divided into training set, validation set and test set according to preset rules; the kernel function weights, penalty parameters or interval parameters of the multi-kernel support vector machine classification model are optimized based on the validation set, and the confusion matrix or accuracy index is output using the test set after optimization to achieve quantitative evaluation of the defect identification effect.

[0014] According to a second aspect of this disclosure, a rolling bearing defect detection system based on encoder feedback is provided, the system comprising: The signal acquisition module is used to acquire the vibration signal of the rolling bearing under test, and to acquire the rotational speed signal coaxial with the rolling bearing under test based on the encoder; The selective band demodulation module is used to preprocess the vibration signal, including: determining the target frequency band where the defect impact is dominant based on the spectral kurtosis; performing bandpass filtering on the vibration signal in the target frequency band to obtain a filtered vibration signal; and performing envelope demodulation on the filtered vibration signal to obtain an envelope signal. The time-angle mapping module is used to generate a speed measurement pulse sequence for angular domain resampling reference from the speed signal, determine the equal angle sampling mark based on the speed measurement pulse sequence, and establish the mapping relationship between the time axis and the rotation angle axis. An angle domain resampling module is used to perform angle domain resampling on the envelope signal according to the sampling marker to obtain an angle domain envelope sequence with equal angular intervals. The order analysis module is used to perform order spectrum analysis on the angular domain envelope sequence to obtain the envelope order spectrum; calculate the outer ring rolling element passing order, inner ring rolling element passing order, and rolling element spin order based on the structural parameters of the rolling bearing under test, and extract the peak values ​​corresponding to the outer ring rolling element passing order, inner ring rolling element passing order, and rolling element spin order and their harmonics from the envelope order spectrum. The feature construction module is used to calculate the peak value ratio features of the outer ring order, the inner ring order, and the rolling body order, wherein: the total peak value of the envelope order spectrum is the sum of the amplitude values ​​of each order line in the envelope order spectrum; the peak value ratio feature of the outer ring order is the ratio of the peak value sum of the outer ring rolling body passing order and its harmonics to the total peak value sum of the envelope order spectrum; the peak value ratio feature of the inner ring order is the ratio of the peak value sum of the inner ring rolling body passing order and its harmonics to the total peak value sum of the envelope order spectrum; and the peak value ratio feature of the rolling body order is the ratio of the peak value sum of the rolling body spin order and its harmonics to the total peak value sum of the envelope order spectrum. The defect classification module is used to input the peak order ratio features of the outer ring, the peak order ratio features of the inner ring, and the peak order ratio features of the rolling element into a trained multi-core support vector machine classification model, and output the defect category of the rolling bearing under test.

[0015] The technical solution disclosed herein has the following beneficial effects: Compared with existing technologies, this disclosure synchronously processes vibration information strongly correlated with rotational speed changes, making it easier for defect-related components to aggregate and be expressed within a unified characterization domain, thereby reducing the influence of feature drift caused by variable rotational speed. Based on this, discriminative features for different defect locations are constructed, and combined with a classification model, the defect category is automatically output. This enables the identification of outer ring, inner ring, rolling element, and composite defects without manual comparison of spectra or reliance on expert experience. Due to its compact feature form and clear computational flow, the overall solution has good engineering feasibility and adaptability to complex working conditions, lowering the threshold for on-site diagnosis and improving the automation level in online monitoring scenarios. Attached Figure Description

[0016] Figure 1 This is a flowchart of a rolling bearing defect detection method based on encoder feedback, as described in the embodiments of this specification. Figure 2 This is a structural block diagram of a rolling bearing defect detection system based on encoder feedback, as described in an embodiment of this specification. Detailed Implementation

[0017] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, apparatus, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0018] Furthermore, the accompanying drawings are merely illustrative of this disclosure. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0019] This invention provides a method for detecting defects in rolling bearings based on encoder feedback. (Refer to...) Figure 1 The diagram shown is a flowchart illustrating a rolling bearing defect detection method based on encoder feedback according to an embodiment of the present invention. This method can be applied to electronic devices such as personal computers, servers, controllers, and display control boards. The method can be executed by a device, which can be implemented by software and / or hardware. Specifically, the method may include the following steps S101-S107: In step S101, the vibration signal of the rolling bearing under test is acquired, and the rotational speed signal coaxial with the rolling bearing under test is acquired based on the encoder.

[0020] In step S102, the vibration signal is preprocessed, including: determining the target frequency band where the defect impact is dominant based on spectral kurtosis; performing bandpass filtering on the vibration signal in the target frequency band to obtain a filtered vibration signal; and performing envelope demodulation on the filtered vibration signal to obtain an envelope signal.

[0021] Before feature extraction, the vibration signal undergoes preprocessing to obtain more useful information for subsequent analysis and reduce the masking effect of background noise and measurement errors on the impact component. The vibration signal typically exhibits amplitude modulation characteristics under varying health conditions or operating conditions. Envelope analysis can separate the modulation information caused by the impact from the carrier frequency component, thus facilitating the observation of modulation patterns related to defects.

[0022] Specifically, determining the target frequency band based on spectral kurtosis includes: decomposing the vibration signal to form a set of candidate frequency bands covering different center frequencies and different bandwidths; calculating a spectral kurtosis index to characterize the significance of the impact component on each set of candidate frequency bands and forming a kurtosis distribution; selecting candidate frequency bands whose spectral kurtosis index satisfies a threshold condition or a ranking condition as the target frequency band according to the extreme value criterion of the spectral kurtosis index; and constructing the passband parameters of the bandpass filter based on the center position and bandwidth range of the target frequency band, thereby suppressing background noise and highlighting the impact component of the defect.

[0023] In this process, spectral kurtosis is used as a criterion for preprocessing frequency band selection. Its role is to determine the frequency band dominated by defect excitation and, based on this, to provide the optimal frequency range for passband filtering, further revealing the defect impact in the vibration signal. Specifically, the vibration signal can be divided into frequency bands at different center frequencies and bandwidth scales to form a candidate frequency band set. The spectral kurtosis is calculated for each candidate frequency band, and a kurtosis distribution map (i.e., a kurtosis map) is constructed to characterize the significance of the impact component in each frequency band. Although the original vibration signal shows impact, it is mostly masked by noise. Therefore, spectral kurtosis is used to determine the frequency band dominated by defect excitation to guide passband setting. The target frequency band is selected based on the extreme values ​​or ranking criteria of the kurtosis distribution, and the passband parameters of the bandpass filter are determined using this target frequency band. The bandpass filter is applied to the vibration signal to obtain the filtered vibration signal. After bandpass filtering of the target frequency band, the defect impact is clearer, noise masking is reduced, and the signal is more suitable for subsequent processing.

[0024] Furthermore, the envelope demodulation of the filtered vibration signal includes: converting the filtered vibration signal into an analytical signal and calculating the amplitude of the analytical signal to obtain an instantaneous amplitude sequence, or rectifying and smoothing the filtered vibration signal to obtain the instantaneous amplitude sequence; and performing low-pass filtering on the instantaneous amplitude sequence to remove the carrier frequency component, thereby obtaining the envelope signal.

[0025] In the envelope demodulation stage, the instantaneous amplitude changes of the filtered vibration signal are extracted to form an envelope signal representing the modulation information. This can be achieved using an analytic signal method, where the filtered vibration signal is constructed into a corresponding analytic signal and its amplitude is taken as the instantaneous amplitude sequence; alternatively, rectification and smoothing can be used to obtain an equivalent instantaneous amplitude sequence. To suppress high-frequency components related to the carrier frequency, the instantaneous amplitude sequence can be low-pass processed to obtain the envelope signal, which is then used in subsequent angular domain resampling and order spectrum analysis processing.

[0026] In step S103, a speed measurement pulse sequence for angular domain resampling reference is generated from the speed signal, an equal angle sampling mark is determined based on the speed measurement pulse sequence, and a mapping relationship between the time axis and the rotation angle axis is established.

[0027] The process of generating the speed measurement pulse sequence from the speed signal includes: when the speed signal is the output of an incremental encoder, performing edge detection or phase decoding on the speed signal to obtain a pulse time stamp corresponding to the rotation angle increment of the shaft; calculating the instantaneous speed based on adjacent pulse time stamps and forming a speed signal; and using the pulse time stamp as the reference for the sampling mark, so that the sampling mark is adaptively updated as the speed signal changes.

[0028] As an explanation, the encoder speed measurement generates the reference signal required for angular domain resampling, enabling data acquired at uniform time intervals to be converted into data with uniform angular intervals. This conversion relies on a reference signal to define the uniform angular interval, typically taken from the speed measurement signal on the drive shaft. When the encoder output is an incremental pulse, edge detection or phase decoding is performed on the pulse to obtain the time stamp corresponding to each pulse. The interval between adjacent pulse time stamps is converted into a speed curve of the shaft rotation speed over time, forming a speed profile. Both the pulse sequence and the speed profile can be directly derived from the encoder signal. Using the pulse sequence as a phase increment reference, time stamps for equal-angle sampling are generated based on the pulse time stamps, i.e., each fixed angular increment corresponds to one sampling stamp, thus maintaining a constant angular increment even when the rotation speed changes. Based on this, a mapping relationship between the time axis and the rotation angle axis is established: the count of each pulse is associated with its time stamp to form a discrete correspondence table of rotation angle over time, and interpolation is performed on the discrete correspondence table to obtain the time position corresponding to any target rotation angle, providing a unified time positioning reference for subsequent resampling at equal angular intervals. To facilitate the subsequent characterization of stability features under varying rotational speeds in the order domain, a definition of order is introduced as a frequency normalization form: ; in Indicates order, The value represents the rotational frequency of the shaft, and n represents the rotational speed of the shaft.

[0029] In step S104, the envelope signal is resampled in the angular domain according to the sampling marker to obtain an angular domain envelope sequence with equal angular intervals.

[0030] The corner domain resampling process includes: determining a target corner sequence with equal angular intervals based on the sampling markers; obtaining the time position corresponding to each target corner using the mapping relationship between the time axis and the corner axis; and interpolating and sampling the envelope signal at each time position to form the corner domain envelope sequence. The interpolation and sampling employs linear interpolation, polynomial interpolation, spline interpolation, or a combination thereof. Before interpolation, the envelope signal undergoes detrending, endpoint extension, or boundary smoothing to reduce interpolation errors, ensuring that the defect impacts are approximately equally spaced in the corner domain envelope sequence.

[0031] In this process, based on the equal-angle sampling time markers obtained in step S103, the envelope signal is converted from a uniform time sampling form to a uniform angle sampling form, so that the data originally collected at uniform time intervals presents as a data sequence with uniform angle intervals in the angle domain. This conversion is accomplished using a digital adaptive resampling approach. The core of this approach is to use the uniform angle intervals defined by the time markers to map the envelope signal onto the angle axis, thereby obtaining the angle domain envelope sequence. During resampling, the target rotation angles corresponding to the sampling markers are used as a unified angle grid, and the time position corresponding to each target rotation angle is determined through the correspondence between the time axis and the rotation axis. At these time positions, the envelope signal is interpolated and sampled. Interpolation can be performed using linear, polynomial, spline, or other methods or combinations thereof to obtain the envelope amplitude at non-integer sampling point positions, thereby achieving reconstruction at equal angle intervals. Before interpolation, the envelope signal can be processed by detrending, endpoint extension, or boundary smoothing to reduce the impact of boundary effects and interpolation errors on the angular domain sequence, so that the impact after resampling is approximately equally spaced on the angular axis. The resampling result can present more equidistant defect impacts in the angular domain, which is more conducive to subsequent order domain analysis than the uneven interval caused by the rotation speed change in the time domain. Furthermore, the horizontal coordinate of the angular domain sequence changes from time to rotation angle (in revolutions), which makes it easier to represent non-stationary signals as more stationary shapes in the angular domain.

[0032] In step S105, the envelope sequence of the angular domain is subjected to order spectrum analysis to obtain the envelope order spectrum; the outer ring rolling element passing order, the inner ring rolling element passing order, and the rolling element spin order are calculated according to the structural parameters of the rolling bearing to be tested, and the peak values ​​corresponding to the outer ring rolling element passing order, the inner ring rolling element passing order, and the rolling element spin order and their harmonics are extracted from the envelope order spectrum.

[0033] The calculation of the outer ring rolling element passage order, the inner ring rolling element passage order, and the rolling element spin order based on the structural parameters of the rolling bearing under test includes: obtaining the number of rolling elements, rolling element diameter, pitch circle diameter, and contact angle parameters of the rolling bearing under test; determining the outer ring rolling element passage order, the inner ring rolling element passage order, and the rolling element spin order based on the number of rolling elements, the geometric ratio of the rolling element diameter to the pitch circle diameter, and the geometric correction term corresponding to the contact angle parameters; and defining integer multiples of each order as corresponding harmonic orders for peak extraction in the envelope order spectrum.

[0034] Furthermore, extracting peak values ​​corresponding to the outer rolling element passage order, the inner rolling element passage order, and the rolling element spin order and its harmonics in the envelope order spectrum includes: setting an order search window for each characteristic order in the envelope order spectrum; performing peak detection within each order search window to obtain the peak amplitude and the order position of the peak; when multi-peak competition is detected, selecting a representative peak value based on the difference in peak amplitude, peak width, peak sharpness, or peak value relative to noise floor; and performing amplitude correction on the representative peak value to make the obtained peak value more accurately reflect the energy accumulation of defect impact in the order domain.

[0035] Specifically, in step S105, the angular domain envelope sequence has been converted into uniform angular interval data, allowing for order spectrum analysis to obtain the envelope order spectrum. This enables the defect components, which originally broadened in the frequency domain under varying rotational speed conditions, to exhibit stable spectral lines in the order domain. The order is used to normalize the frequency according to the shaft rotational speed, and its definition is: ; Order spectral analysis can be understood as performing a Fourier transform on the envelope sequence of the diagonal domain and forming a set of spectral lines with the order as the abscissa, thereby obtaining the envelope order spectrum. R is the number of spectral lines.

[0036] The defect-related characteristic order is determined by the bearing structural parameters, and the outer ring rolling elements pass through the order. The inner rolling elements pass through stages Spin order of rolling elements They are respectively: ; ; ; in, The number of rolling elements. The diameter of the rolling element, The diameter of the pitch circle. The contact angle. Since the order of the defect remains constant under varying rotational speeds, the envelope order spectrum can be centered around... , , The search target is established based on its integer multiples of harmonics, and the peak value is extracted accordingly.

[0037] In step S106, the peak value ratio characteristics of the outer ring order, the inner ring order, and the rolling element order are calculated, wherein: the total peak value sum of the envelope order spectrum is the sum of the amplitude values ​​of each order line in the envelope order spectrum; the peak value ratio characteristic of the outer ring order is the ratio of the peak value sum of the outer ring rolling element passing order and its harmonics to the total peak value sum of the envelope order spectrum; the peak value ratio characteristic of the inner ring order is the ratio of the peak value sum of the inner ring rolling element passing order and its harmonics to the total peak value sum of the envelope order spectrum; and the peak value ratio characteristic of the rolling element order is the ratio of the peak value sum of the rolling element spin order and its harmonics to the total peak value sum of the envelope order spectrum.

[0038] In step S106, the envelope order spectrum obtained in step S105 is used as a unified metric basis. The energy concentration of the spectral lines related to the three types of defects—outer ring, inner ring, and rolling element—is transformed into a dimensionless ratio characteristic to reduce the influence of overall amplitude level changes under different operating conditions on the discrimination. The total peak value of the envelope order spectrum is the sum of the amplitudes of all spectral lines. ; During peak extraction, an order search window is set for each target order and its harmonics on the envelope order spectrum. Peak detection is performed within the window to obtain the peak amplitude and peak position. If multiple candidate peaks appear within the window, a representative peak can be selected by combining criteria such as peak amplitude and noise floor difference. The extracted peaks can be used to form three types of peak ratio features: outer ring, inner ring, and rolling element. This type of feature is expressed as the ratio of the sum of defect peaks to the sum of peaks in the entire order spectrum. The total peak sum of the order spectrum is calculated by... Obtained by summing the amplitudes across all spectral lines. Outer peak-to-peak ratio characteristic. Inner circle peak ratio characteristics Rolling element peak ratio characteristics The expression is: ; ; ; in, , , These represent the peak values ​​of the u-th harmonics of the target order corresponding to defects in the outer ring, inner ring, and rolling element, respectively, where U is the harmonic number corresponding to the defect, and K is the proportionality coefficient. Under variable speed conditions, the envelope order spectrum can show that the defect order and its harmonics remain constant, thus ensuring that the above peak value extraction and peak value ratio characteristics remain comparable and stable under different speed variation modes.

[0039] In step S107, the peak order ratio features of the outer ring, the peak order ratio features of the inner ring, and the peak order ratio features of the rolling element are input into the trained multi-core support vector machine classification model, and the defect category of the rolling bearing under test is output.

[0040] The multi-kernel support vector machine classification model is constructed in the following way: Training data with known health status labels is collected, and the training data is collected under different speed change modes to cover non-stationary operating conditions; for each training sample, the peak order ratio features of the outer race, the peak order ratio features of the inner race, and the peak order ratio features of the rolling element are extracted to form a feature vector; the feature vector is normalized or scaled; a support vector machine is trained using a kernel mapping formed by a weighted combination of multiple kernel functions to obtain a classification decision boundary that distinguishes between health status, outer race defects, inner race defects, rolling element defects, and composite defects, and the classification decision boundary is solidified into the multi-kernel support vector machine classification model.

[0041] As an explanation, the peak order ratio features of the outer ring, the peak order ratio features of the inner ring, and the peak order ratio features of the rolling elements are combined into a feature vector of the same training sample. This feature vector is then input into a pre-trained multi-core support vector machine classification model, which outputs the defect category of the rolling bearing under test. This category is used to distinguish between healthy conditions, outer ring defects, inner ring defects, rolling element defects, and combined defects, so as to achieve automatic identification of rolling bearing defects under variable speed conditions.

[0042] In the construction of the multi-core support vector machine classification model, the training data consists of vibration signals with known health status labels, and is collected under various speed change modes to cover non-stationary working conditions. The speed change modes include speed increase, speed decrease, speed increase followed by decrease, and speed decrease followed by increase. Each health status corresponds to a set of samples, enabling the classification model to maintain the ability to distinguish defect categories under different operating speed changes.

[0043] For the training samples, the three-dimensional feature space input is constructed based on the outer ring order peak ratio feature, inner ring order peak ratio feature and rolling body order peak ratio feature extracted from each sample. In order to facilitate classifier training and comparability, all features are scaled. The feature can be normalized to the interval [0,1] to achieve a unified expression of features with different dimensions and amplitude ranges.

[0044] In one embodiment, to improve the generalization ability of the multi-kernel support vector machine classification model, each vibration signal is divided into multiple segments along the acquisition time or along the rotation sequence, and each segment is used as an independent sample and inherits the health status label corresponding to the vibration signal; the independent samples are divided into training set, validation set and test set according to preset rules; the kernel function weights, penalty parameters or interval parameters of the multi-kernel support vector machine classification model are optimized based on the validation set, and after optimization, the confusion matrix or accuracy index is output using the test set to achieve quantitative evaluation of the defect identification effect.

[0045] In this model, a single vibration signal can be divided into multiple segments according to the acquisition time axis or the rotation sequence. Each segment maintains the same health status label as the original signal, transforming long-term time-series data into a more abundant set of independent samples. This facilitates the coverage of fluctuations caused by different speed changes and reduces the accidental influence of a small number of samples on the classification boundary. The segment length and division method can be determined according to preset rules. Equal-length division can be used, or overlapping can be set between adjacent segments to improve sample representativeness. A unified division strategy avoids inconsistencies in sample granularity between different health states. When entering the modeling stage, independent samples are divided into training, validation, and test sets according to preset rules. The distribution of each health state in each subset can be kept consistent during the division to reduce the interference of class imbalance on the evaluation results. The training set is used to learn the classification decision boundary, the validation set is used to adjust the model hyperparameters without touching the test set, and the test set is used to provide the final objective evaluation, thereby avoiding the bias of parameter selection on the test results.

[0046] The optimization of multi-kernel support vector machines revolves around the weighted combination of kernel function parameters, penalty parameters, and margin parameters. The validation set is used to compare classification performance under different parameter combinations and select the configuration with stronger generalization ability. During this process, the inputs to both the training and validation sets are in the form of feature vectors of the same type, and a consistent scaling strategy is used to ensure the comparability of parameter comparisons. After optimization, the test set features are input into a classification model with fixed parameters, outputting the predicted category for each health state. A confusion matrix is ​​used to statistically analyze the correct or incorrect classification of each category. When the samples belong to a class-balanced scenario, accuracy can be used as the evaluation metric. Accuracy is obtained by counting true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN). ; The confusion matrix and accuracy are used together to quantify the defect identification effect, so that the identification stability and misjudgment distribution of the model under different speed change modes can be presented intuitively and can be reproduced and compared.

[0047] Based on the same line of thought, such as Figure 2The diagram shown is a structural block diagram of a rolling bearing defect detection system based on encoder feedback according to an embodiment of the present invention. The system includes: The signal acquisition module 201 is used to acquire the vibration signal of the rolling bearing under test, and to acquire the rotational speed signal coaxial with the rolling bearing under test based on the encoder; The selective band demodulation module 202 is used to preprocess the vibration signal, including: determining the target frequency band where the defect impact is dominant based on the spectral kurtosis; performing bandpass filtering on the vibration signal in the target frequency band to obtain a filtered vibration signal; and performing envelope demodulation on the filtered vibration signal to obtain an envelope signal. The time-angle mapping module 203 is used to generate a speed measurement pulse sequence for angular domain resampling reference from the speed signal, determine the equal angle sampling mark based on the speed measurement pulse sequence, and establish the mapping relationship between the time axis and the rotation angle axis. Angle domain resampling module 204 is used to perform angle domain resampling on the envelope signal according to the sampling marker to obtain an angle domain envelope sequence with equal angular intervals; The order analysis module 205 is used to perform order spectrum analysis on the angular domain envelope sequence to obtain the envelope order spectrum; calculate the outer ring rolling element passing order, inner ring rolling element passing order, and rolling element spin order based on the structural parameters of the rolling bearing under test, and extract the peak values ​​corresponding to the outer ring rolling element passing order, inner ring rolling element passing order, and rolling element spin order and their harmonics from the envelope order spectrum. The feature construction module 206 is used to calculate the peak value ratio features of the outer ring order, the inner ring order, and the rolling body order, wherein: the total peak value sum of the envelope order spectrum is the sum of the amplitude values ​​of each order line in the envelope order spectrum; the peak value ratio feature of the outer ring order is the ratio of the peak value sum of the outer ring rolling body passing order and its harmonics to the total peak value sum of the envelope order spectrum; the peak value ratio feature of the inner ring order is the ratio of the peak value sum of the inner ring rolling body passing order and its harmonics to the total peak value sum of the envelope order spectrum; and the peak value ratio feature of the rolling body order is the ratio of the peak value sum of the rolling body spin order and its harmonics to the total peak value sum of the envelope order spectrum. The defect classification module 207 is used to input the outer ring order peak ratio feature, the inner ring order peak ratio feature and the rolling element order peak ratio feature into the trained multi-core support vector machine classification model, and output the defect category of the rolling bearing under test.

[0048] The specific details of the above system have been described in detail in the method section of the implementation plan. For any undisclosed details, please refer to the implementation plan of the method section, and therefore will not be repeated here.

[0049] This system synchronously processes vibration information strongly correlated with rotational speed changes, making it easier for defect-related components to aggregate and be expressed within a unified characterization domain, thereby reducing the influence of feature drift caused by variable rotational speed. Based on this, it constructs discriminative features for different defect locations and, combined with a classification model, automatically outputs defect categories. This enables the identification of outer ring, inner ring, rolling element, and composite defects without requiring manual comparison of spectra or reliance on expert experience. Due to its compact feature format and clear computational flow, the overall solution has good engineering feasibility and adaptability to complex working conditions, lowering the threshold for on-site diagnosis and improving the automation level in online monitoring scenarios.

[0050] The accompanying drawings are merely illustrative of the processes included in the methods according to exemplary embodiments of this disclosure and are not intended to be limiting. It is readily understood that the processes shown in the drawings do not indicate or limit the temporal order of these processes. Furthermore, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0051] It should be noted that although several modules or units of the system have been mentioned in the detailed description above, this division is not mandatory. In fact, according to exemplary embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0052] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0053] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for detecting defects in rolling bearings based on encoder feedback, characterized in that, The method includes: The vibration signal of the rolling bearing under test is acquired, and the rotational speed signal coaxial with the rolling bearing under test is acquired based on the encoder; The vibration signal is preprocessed, including: determining the target frequency band where the defect impact is dominant based on spectral kurtosis; performing bandpass filtering on the vibration signal in the target frequency band to obtain a filtered vibration signal; and performing envelope demodulation on the filtered vibration signal to obtain an envelope signal. A speed measurement pulse sequence for angular domain resampling reference is generated from the speed signal, an equal angle sampling mark is determined based on the speed measurement pulse sequence, and a mapping relationship between the time axis and the rotation axis is established. The envelope signal is resampled in the angular domain according to the sampling marker to obtain an angular domain envelope sequence with equal angular intervals; The envelope sequence is subjected to order spectrum analysis to obtain the envelope order spectrum; the outer ring rolling element passing order, inner ring rolling element passing order, and rolling element spin order are calculated according to the structural parameters of the rolling bearing under test, and the peak values ​​corresponding to the outer ring rolling element passing order, inner ring rolling element passing order, and rolling element spin order and their harmonics are extracted from the envelope order spectrum. Calculate the peak value ratio characteristics of the outer ring order, the inner ring order, and the rolling element order, wherein: the total peak value sum of the envelope order spectrum is the sum of the amplitude values ​​of each order line in the envelope order spectrum; the peak value ratio characteristic of the outer ring order is the ratio of the peak value sum of the outer ring rolling element passing order and its harmonics to the total peak value sum of the envelope order spectrum; the peak value ratio characteristic of the inner ring order is the ratio of the peak value sum of the inner ring rolling element passing order and its harmonics to the total peak value sum of the envelope order spectrum; and the peak value ratio characteristic of the rolling element order is the ratio of the peak value sum of the rolling element spin order and its harmonics to the total peak value sum of the envelope order spectrum. The outer ring order peak ratio feature, the inner ring order peak ratio feature, and the rolling element order peak ratio feature are input into a trained multi-core support vector machine classification model to output the defect category of the rolling bearing under test.

2. The method for detecting defects in rolling bearings based on encoder feedback according to claim 1, characterized in that, Determining the target frequency band based on spectral kurtosis includes: The vibration signal is decomposed to form a set of candidate frequency bands covering different center frequencies and different bandwidths; Spectral kurtosis indices for characterizing the significance of impact components are calculated on each set of candidate frequency bands, and kurtosis distributions are formed. Candidate frequency bands that satisfy threshold conditions or ranking conditions of spectral kurtosis indices are selected as target frequency bands according to the extreme value criteria of the spectral kurtosis indices. The passband parameters of the bandpass filter are constructed based on the center position and bandwidth range of the target frequency bands, thereby suppressing background noise and highlighting the impact components of defects.

3. The method for detecting defects in rolling bearings based on encoder feedback according to claim 1, characterized in that, Envelope demodulation of the filtered vibration signal includes: converting the filtered vibration signal into an analytical signal and calculating the amplitude of the analytical signal to obtain an instantaneous amplitude sequence, or rectifying and smoothing the filtered vibration signal to obtain the instantaneous amplitude sequence; and performing low-pass filtering on the instantaneous amplitude sequence to remove the carrier frequency component, thereby obtaining the envelope signal.

4. The method for detecting defects in rolling bearings based on encoder feedback according to claim 1, characterized in that, Generating the speed measurement pulse sequence from the speed signal includes: When the speed signal is output by an incremental encoder, edge detection or phase decoding is performed on the speed signal to obtain a pulse time stamp corresponding to the rotation angle increment of the shaft; Instantaneous rotational speed is calculated based on adjacent pulse time stamps, and a rotational speed signal is generated; Using the pulse time marker as a reference for the sampling marker, the sampling marker is adaptively updated as the rotational speed signal changes.

5. The method for detecting defects in rolling bearings based on encoder feedback according to claim 1, characterized in that, When performing corner domain resampling, the following are included: Determine the target turning angle sequence with equal angular intervals based on the sampling marks; The time position corresponding to each target rotation angle is obtained by using the mapping relationship between the time axis and the rotation axis; The envelope signal is interpolated and sampled at each of the aforementioned time positions to form the angular domain envelope sequence; wherein the interpolation sampling adopts linear interpolation, polynomial interpolation, spline interpolation or a combination thereof, and the envelope signal is subjected to detrending, endpoint extension or boundary smoothing processing before interpolation to reduce interpolation error, so that the defect impact presents an approximately equally spaced distribution in the angular domain envelope sequence.

6. The method for detecting defects in rolling bearings based on encoder feedback according to claim 1, characterized in that, The calculation of the outer ring rolling element passage order, the inner ring rolling element passage order, and the rolling element spin order based on the structural parameters of the rolling bearing under test includes: Obtain the number of rolling elements, rolling element diameter, pitch circle diameter, and contact angle parameters of the rolling bearing under test; Based on the number of rolling elements, the geometric ratio of the rolling element diameter to the pitch circle diameter, and the geometric correction term corresponding to the contact angle parameter, the passing order of the outer rolling element, the passing order of the inner rolling element, and the spin order of the rolling element are determined respectively; and integer multiples of each order are defined as the corresponding harmonic orders for peak extraction in the envelope order spectrum.

7. The method for detecting defects in rolling bearings based on encoder feedback according to claim 1, characterized in that, The peak values ​​extracted from the envelope order spectrum corresponding to the passing order of the outer rolling element, the passing order of the inner rolling element, and the spin order and harmonics of the rolling element include: An order search window is set for each of the feature orders in the envelope order spectrum; Peak detection is performed within each of the order search windows to obtain the peak amplitude and the order position of the peak; when multi-peak competition is detected, a representative peak is selected based on the difference in peak amplitude, peak width, peak sharpness, or peak relative to noise floor. The representative peak value is corrected to make the obtained peak value more accurately reflect the energy accumulation of the defect impact in the order domain.

8. The method for detecting defects in rolling bearings based on encoder feedback according to claim 1, characterized in that, The multi-kernel support vector machine classification model is constructed in the following way: Training data with known health status labels is collected, and the training data is collected under different speed change modes to cover non-stationary operating conditions; For each training sample, extract the peak order ratio features of the outer ring, the peak order ratio features of the inner ring, and the peak order ratio features of the rolling element to form a feature vector; The feature vectors are normalized or scaled; a support vector machine is trained using a kernel mapping formed by a weighted combination of multiple kernel functions to obtain a classification decision boundary that distinguishes between healthy state, outer ring defects, inner ring defects, rolling element defects and composite defects, and the classification decision boundary is solidified into the multi-kernel support vector machine classification model.

9. The method for detecting defects in rolling bearings based on encoder feedback according to claim 8, characterized in that, The method further includes: Each vibration signal is divided into multiple segments along the acquisition time or along the rotation angle sequence, and each segment is used as an independent sample and inherits the health status label corresponding to the vibration signal. The independent samples are divided into training set, validation set and test set according to preset rules; the kernel function weights, penalty parameters or interval parameters of the multi-kernel support vector machine classification model are optimized based on the validation set, and the confusion matrix or accuracy index is output using the test set after optimization to achieve quantitative evaluation of the defect identification effect.

10. A rolling bearing defect detection system based on encoder feedback, the system comprising: The signal acquisition module is used to acquire the vibration signal of the rolling bearing under test, and to acquire the rotational speed signal coaxial with the rolling bearing under test based on the encoder; The selective band demodulation module is used to preprocess the vibration signal, including: determining the target frequency band where the defect impact is dominant based on the spectral kurtosis; performing bandpass filtering on the vibration signal in the target frequency band to obtain a filtered vibration signal; and performing envelope demodulation on the filtered vibration signal to obtain an envelope signal. The time-angle mapping module is used to generate a speed measurement pulse sequence for angular domain resampling reference from the speed signal, determine the equal angle sampling mark based on the speed measurement pulse sequence, and establish the mapping relationship between the time axis and the rotation angle axis. An angle domain resampling module is used to perform angle domain resampling on the envelope signal according to the sampling marker to obtain an angle domain envelope sequence with equal angular intervals. The order analysis module is used to perform order spectrum analysis on the angular domain envelope sequence to obtain the envelope order spectrum; calculate the outer ring rolling element passing order, inner ring rolling element passing order, and rolling element spin order based on the structural parameters of the rolling bearing under test, and extract the peak values ​​corresponding to the outer ring rolling element passing order, inner ring rolling element passing order, and rolling element spin order and their harmonics from the envelope order spectrum. The feature construction module is used to calculate the peak value ratio features of the outer ring order, the inner ring order, and the rolling body order, wherein: the total peak value of the envelope order spectrum is the sum of the amplitude values ​​of each order line in the envelope order spectrum; the peak value ratio feature of the outer ring order is the ratio of the peak value sum of the outer ring rolling body passing order and its harmonics to the total peak value sum of the envelope order spectrum; the peak value ratio feature of the inner ring order is the ratio of the peak value sum of the inner ring rolling body passing order and its harmonics to the total peak value sum of the envelope order spectrum; and the peak value ratio feature of the rolling body order is the ratio of the peak value sum of the rolling body spin order and its harmonics to the total peak value sum of the envelope order spectrum. The defect classification module is used to input the peak order ratio features of the outer ring, the peak order ratio features of the inner ring, and the peak order ratio features of the rolling element into a trained multi-core support vector machine classification model, and output the defect category of the rolling bearing under test.

Images