An automatic calculation method for quantitative evaluation of local defect size of rolling bearing

Abstract

The application discloses an automatic calculation method for quantitative evaluation of local defect size of a rolling bearing, and comprises the following steps: firstly, the height sensitivity of signal mutation is utilized, and adaptive dynamic threshold determination is combined to realize automatic and accurate positioning and batch extraction of fault pulses in a long-time time domain signal; then, a frequency domain mask is introduced to act on a smooth pseudo-SPWVD distribution to enhance low-frequency characteristic energy in entering and leaving defect stages, and the automatic and accurate identification of entering and leaving damage time is realized through a Laplace operator; finally, kernel density estimation is utilized to statistically infer batch samples to realize accurate estimation of damage size. The application shows good applicability and stability under various rotating speeds and different damage sizes of inner and outer rings, and provides a new method for precise quantitative detection in actual engineering.

Description

Technical Field

[0001] This invention relates to the field of rotating machinery fault diagnosis technology, and in particular to an automated calculation method for quantitative assessment of the size of local defects in rolling bearings. Background Technology

[0002] As a core component of rotating machinery, rolling bearings directly impact the safety and stability of the entire industrial system. During long-term, high-load service, localized defects such as fatigue pitting or spalling inevitably occur on the bearing surface. Currently, various successful methods have been widely implemented for assessing the degree of bearing defects. With the rapid development of AI, many researchers are dedicated to using machine (deep) learning intelligent models to extract high-dimensional features from bearing fault data to describe the bearing's degradation state or perform quantitative damage assessment. For example, networks such as CNN, RNN, and LSTM are used to automatically extract damage-sensitive features from bearing signals to classify fault states or predict damage levels. These methods perform well in identifying whether a bearing is faulty and its "fault evolution trend," but these models require extensive manual annotation for training. Another popular approach is to directly calculate the true physical dimensions of local bearing defects using damage impact features or physical models. However, existing assessment methods are difficult for non-professional users to operate and computationally expensive due to cumbersome algorithm combinations, high dependence on physical model parameters, and computational complexity. Furthermore, several key challenges remain in practical applications, such as: 1) Reliance on manual judgment and selection: Most existing studies rely on manual observation or supervision to identify entry and exit feature points. This non-automatic extraction method is not only inefficient, but the estimation results of a single pulse (or a few pulses) also have a certain degree of randomness and bias, and are prone to introducing subjective measurement errors; 2) Extremely uneven feature distribution: The structural resonance generated at the trailing edge of a rolling body impact defect usually manifests as a main energy peak with extremely high amplitude, while the low-frequency entry and exit features characterizing the geometric dimensions of the loss manifest as weaker secondary peaks. This energy level difference makes it easy for weak features to be submerged by noise or the main peak, posing a great challenge to specific automatic identification. Summary of the Invention

[0003] Purpose of the invention: The purpose of this invention is to provide an automated calculation method for the quantitative assessment of local defect dimensions in rolling bearings. By utilizing pulse signal edge features and statistical inference mechanisms, this method eliminates the reliance on manual experience in the quantitative calculation of local defects, enabling rapid, effective, and convenient quantitative calculation of local damage dimensions on the inner and outer raceways of rolling bearings.

[0004] Technical solution: An automated calculation method for quantitative evaluation of local defect dimensions in rolling bearings, comprising the following steps:

[0005] S1, The vibration time-domain pulse signal x(n) of the rolling bearing under working conditions is obtained through a vibration sensor;

[0006] S2, using the Laplace operator combined with a dynamic adaptive threshold determination method, processes the vibration time-domain pulse signal x(n) to achieve automatic location and batch extraction of fault pulses;

[0007] S3. Based on the extracted fault pulse, a frequency domain mask is introduced, and the instantaneous energy at the entry and exit times is corrected using the SPWVD distribution. After obtaining the corrected instantaneous energy curve, the Laplace operator is used to detect abrupt boundary changes in instantaneous energy, automatically locating potential entry and exit times of damage, and obtaining the effective duration of the rolling element passing through the damage region. ;

[0008] S4, calculate the effective duration by batch pulse signals. Obtain a sample set, and use kernel density estimation to perform statistical inference on the sample set to obtain an accurate estimate of the damage size.

[0009] Furthermore, the steps for automatically locating and batch extracting fault pulses in long-time time-domain signals include:

[0010] S21, construct the characteristic envelope function based on the Laplace operator to determine the position of each impact peak; define the calculation result L(n) and envelope function Elap(n) of the pulse signal x(n) as follows:

[0011] ,

[0012] ,

[0013] Where w(n) is a normalized rectangular window function; when written in discrete form, 2k+1 is the width of the window, and i is the summation variable; the window length is selected based on the actual data and is greater than a single period of the high-frequency resonance waveform induced by the damage.

[0014] S22, Calculate the adaptive threshold Γ based on the global statistical characteristics of the signal;

[0015] S23, when the envelope feature E lap When >Γ, the potential time of damage impact is marked as t. trig As the trigger time, t trig Starting from W, within the local search interval seg Searching for E inside lap The maximum point of (n):

[0016] ,

[0017] Where, n centerThis is the physical center of the damage pulse;

[0018] With a defined n center Using the geometric reference, N / 2 sampling points are extended to both the left and right ends to accurately extract discrete pulse samples S from the original signal. k , as a damage impact sample set;

[0019] ,

[0020] Where N represents the estimated number of pulse length points.

[0021] Furthermore, the adaptive threshold Γ is obtained by solving the feature envelope function E. lap The cumulative distribution function of (n) is obtained, and its expression is as follows:

[0022] ,

[0023] in, Let E be the probability density function of the envelope signal; Q is the envelope feature E. lap The percentile is used to identify the energy concentration interval in the envelope sequence, ensuring that effective pulses are extracted from the original signal.

[0024] Furthermore, the specific implementation steps of step S3 are as follows:

[0025] S31, defining the frequency domain mask function H(f) as a low-pass filter, then the corrected time-frequency distribution... Represented as:

[0026] ,

[0027] in, Let f represent the cutoff frequency f of the SPWVD distribution function H(f). mask Determined by the characteristic response range of the experimental data;

[0028] S32, based on the time-frequency distribution, calculate the corrected instantaneous energy IE, expressed as:

[0029] ,

[0030] in, This represents the lowest frequency at which non-zero energy begins to appear in the time-frequency graph;

[0031] S33, after obtaining the corrected instantaneous energy curve, uses the Laplace operator. To detect abrupt changes in instantaneous energy boundaries; by designing a double significance threshold, the entry and exit points from damage are determined, expressed as:

[0032] ,

[0033] Where, σ n α is the median absolute deviation, used to estimate the noise scale of the instantaneous energy curve; 1、 α² represents the peak significance coefficients at the entry and exit times, respectively; P entry P represents the significance threshold at the time of injury. exit The threshold for the time of departure from the injury is indicated; median() indicates taking the median.

[0034] S34, for each detected entry or exit point of damage, search for the nearest peak in its neighborhood and update the position of the point to the nearest peak index to ensure that the final location falls on the moment of energy concentration in a physical sense.

[0035] S35, calculate all possible time differences using a global matching method, and take the largest time difference as the effective duration for the rolling element to pass through the damaged area, as shown in the following expression:

[0036] ,

[0037] in, This indicates the time when the damage peak range is reached after correction. This indicates the time after correction to leave the peak damage range. .

[0038] Furthermore, a peak correction strategy is adopted to correct the time of entering / leaving the damage peak range, as shown in the following expression:

[0039] ,

[0040] Among them, U entry U exit These are the sets of entry and exit damage peaks, respectively; This indicates the time when the damage peak range is reached after correction. Indicates the time after correction to leave the peak damage range; Represents set U entry At a certain peak moment of damage, Represents set U exit A certain peak moment of damage; This indicates the initial detected moment of entry into the damage point. This indicates the initial detected moment of departure from the damage point.

[0041] Furthermore, in step S4, the steps to obtain an accurate estimate of the damage size are as follows:

[0042] S41, Substituting the samples into the kernel density estimation formula yields the probability density curve, where the kernel estimate of the probability density function is defined as:

[0043] ,

[0044] In the formula, K is the Gaussian kernel function, and h is the adaptive bandwidth; The time interval of the i-th fault pulse is represented by M; M is the number of fault pulses.

[0045] After obtaining the probability density curve, the distribution center is used as the final quantitative calculation basis to obtain the most reliable time interval for the rolling element to pass through the damage area. ;

[0046] S42, calculate the defect size according to the following formula. :

[0047] ,

[0048] Among them, f r D is the rotational speed frequency. p d is the bearing pitch diameter, and d is the rolling element diameter.

[0049] Compared with the prior art, the significant advantages of this invention are as follows:

[0050] 1. This invention provides an automated calculation method for local defect dimensions of rolling bearings based on the Laplace operator and instantaneous energy. It utilizes the Laplace operator to detect abrupt boundary changes in instantaneous energy, automatically locating potential entry and exit points of damage. This achieves automated and high-precision identification of the entry / exit points of bearing rolling elements into / out of the damage region. It can quickly, effectively, and conveniently calculate the local damage dimensions on the inner and outer raceways of rolling bearings, providing an important technical approach for achieving rapid, automated, precise, and quantitative detection of rolling bearings in engineering.

[0051] 2. This invention achieves automated and precise batch segmentation and extraction of bearing fault pulse signals, eliminating the need for manual screening and efficiently obtaining a batch of valid analysis samples. Furthermore, it utilizes kernel density estimation (KDE) to perform statistical inference on the batch samples to complete defect size assessment. This eliminates random errors and noise interference from single extractions and avoids random errors in single-sample calculations, thereby effectively improving the accuracy and stability of quantitative results for local defect sizes in bearings. Attached Figure Description

[0052] Figure 1 This is a flowchart of the present invention;

[0053] Figure 2 This is a schematic diagram of a faulty bearing with different sizes of local damage on its inner and outer raceways;

[0054] Figure 3 This is a schematic diagram of the automatic segmentation of fault pulses;

[0055] Figure 4 This is a schematic diagram of the time domain of the first 8 pulses automatically extracted;

[0056] Figure 5 These are schematic diagrams showing the frequency domain mask identification and calculation results of instantaneous energy of 1.0mm and 1.4mm local defects in the inner circle. Among them, (a) is the time domain waveform of a single pulse of the 1.0mm local defect in the inner circle, (b) is the schematic diagram of the automatic identification and calculation results of the instantaneous energy of the 1.0mm local defect in the inner circle across the full frequency domain, (c) is the schematic diagram of the automatic identification and calculation results of the frequency domain mask of the instantaneous energy of the 1.0mm local defect in the inner circle, (d) is the time domain waveform of a single pulse of the 1.4mm local defect in the inner circle, (e) is the schematic diagram of the automatic identification and calculation results of the instantaneous energy of the 1.4mm local defect in the inner circle across the full frequency domain, and (f) is the schematic diagram of the automatic identification and calculation results of the frequency domain mask of the instantaneous energy of the 1.4mm local defect in the inner circle.

[0057] Figure 6 These are schematic diagrams showing the frequency domain mask identification and calculation results of instantaneous energy of 1.8mm and 2.2mm local defects in the inner circle. Among them, (a) is the time domain waveform of a single pulse of the 1.8mm local defect in the inner circle, (b) is the schematic diagram of the automatic identification and calculation results of the instantaneous energy of the 1.8mm local defect in the inner circle across the entire frequency domain, (c) is the schematic diagram of the automatic identification and calculation results of the frequency domain mask of the instantaneous energy of the 1.8mm local defect in the inner circle, (d) is the time domain waveform of a single pulse of the 2.2mm local defect in the inner circle, (e) is the schematic diagram of the automatic identification and calculation results of the instantaneous energy of the 2.2mm local defect in the inner circle across the entire frequency domain, and (f) is the schematic diagram of the automatic identification and calculation results of the frequency domain mask of the instantaneous energy of the 2.2mm local defect in the inner circle.

[0058] Figure 7 This is a schematic diagram of the quantitative calculation results of four types of local defects in the inner circle based on kernel density. (a) is a local defect of 1.0 mm in the inner circle, (b) is a local defect of 1.4 mm in the inner circle, (c) is a local defect of 1.8 mm in the inner circle, and (d) is a local defect of 2.2 mm in the inner circle.

[0059] Figure 8 This is a schematic diagram of the quantitative calculation error of the inner and outer rings under four dimensions at different speeds, where (a) is the quantitative calculation result of the inner ring damage and (b) is the quantitative calculation result of the outer ring damage. Detailed Implementation

[0060] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0061] like Figure 1 As shown, this invention provides an automated calculation method for quantitative evaluation of local defect dimensions in rolling bearings, comprising the following steps:

[0062] Step 1: Obtain the vibration time-domain pulse signal x(n) of the rolling bearing under working conditions using a vibration sensor.

[0063] Step 2: Utilizing the high sensitivity of the Laplace operator to signal abrupt changes, combined with adaptive dynamic threshold determination, automatic and accurate location and batch extraction of fault pulses in long-term time-domain signals are achieved.

[0064] Step 21: Construct a characteristic envelope function based on the Laplace operator to determine the position of each impact peak; define the calculation result L(n) of the pulse signal x(n) Laplace operator and the envelope function Elap(n) as follows:

[0065] (1)

[0066] (2)

[0067] Where w(n) is a normalized rectangular window function; when written in discrete form, 2k+1 is the width of the window, and i is the summation variable (index variable), representing the relative position within the current calculation window.

[0068] The window length should be selected based on actual data and should be greater than a single period of the high-frequency resonance waveform induced by the damage in order to eliminate energy fluctuations caused by high-frequency oscillations. It should also not exceed the expected impact pulse width to prevent excessive energy smoothing from causing blurring of feature edges.

[0069] Step 22, calculate the dynamic adaptive threshold;

[0070] To achieve effective locking and batch extraction of impact features, a dynamic adaptive threshold determination method based on global statistical features of the signal is established. Let Γ be the adaptive threshold, and the method is developed by solving the feature envelope function E. lap The cumulative distribution function of (n) is obtained, that is:

[0071] (3)

[0072] in, Let E be the probability density function of the envelope signal; Q is the envelope feature E. lap The percentile of Q is used to identify the energy concentration range in the envelope sequence, ensuring that effective pulses are extracted from the original signal.

[0073] Step 23, Pulse Triggering and Alignment;

[0074] When the envelope feature E is satisfied lap When >Γ, the potential time of damage impact is marked as t. trig As the trigger time, t trig Starting from W, within the local search interval seg Searching for E insidelap The maximum point of (n):

[0075] (4)

[0076] Where, n center This is the physical center of the damage pulse.

[0077] With a defined n center Using the geometric reference, N / 2 sampling points are extended to both the left and right ends (the values ​​are related to the interval length between the damaged pulses to prevent impact aliasing), and discrete pulse samples S are accurately extracted from the original signal. k , as a damage impact sample set;

[0078] (5)

[0079] Where N represents the estimated number of pulse length points.

[0080] By performing steps 21-23 above on the vibration time-domain pulse signal, multiple damage impact sample sets can be extracted automatically.

[0081] Step 3: A frequency domain mask is introduced to act on the SPWVD distribution (smoothed pseudo Wigner-Ville distribution) to enhance the low-frequency feature energy of the entry and exit defect stages, and the Laplace operator is used to realize the automatic and accurate identification of the entry and exit feature moments.

[0082] Step 31: Define the frequency domain mask function H(f) as a low-pass filter and apply it to the time-frequency matrix generated by SPWVD;

[0083] (6)

[0084] Corrected time-frequency distribution Represented as:

[0085] (7)

[0086] Wherein, the cutoff frequency f of H(f) mask The range of characteristic response data is determined by the characteristic frequency of the low-frequency energy concentration region in the characteristic response band.

[0087] Step 32, Instantaneous Energy Analysis;

[0088] The corrected instantaneous energy IE is calculated and expressed as:

[0089] (8)

[0090] in, This represents the lowest frequency at which non-zero energy begins to appear in the time-frequency graph.

[0091] Step 33: Calculate the significance thresholds for entering the injury and leaving the injury.

[0092] After obtaining the corrected instantaneous energy curve, it is necessary to automatically extract the entry and exit points of damage. This is achieved using the Laplace operator. This method detects abrupt energy transitions and automatically locates potential entry and exit points from damage. A dual significance threshold P is designed. entry and P exit To determine the time of entry into damage and the time of exit from damage, we express them as follows:

[0093] (9)

[0094] Where, σ n The median absolute deviation is used to estimate the noise scale of the instantaneous energy curve; α1 and α2 are the peak significance coefficients at the entry and exit points of damage, respectively; P entry P represents the significance threshold at the time of injury. exit The threshold for significance at the time of injury is indicated; median() indicates taking the median.

[0095] Step 34: Locate the peak entry / exit time;

[0096] When detecting entry / exit from the damage peak interval ΔT (i.e., the effective duration ΔT for the rolling element to pass through the damage area), a peak correction strategy is defined to avoid positioning deviations caused by threshold fluctuations, expressed as:

[0097] (10)

[0098] Among them, U entry U exit These are the sets of entry and exit damage peaks, respectively; This indicates the time when the damage peak range is reached after correction. This indicates the time after correction to leave the peak damage range. , Representing set U respectively entry U exit At a certain peak moment of damage, This indicates the initial detected moment of entry into the damage point. This indicates the initial detected moment of departure from the damage point.

[0099] For each detected entry or exit point of damage, the nearest peak is searched within its neighborhood, and the location of the damage point is updated with the index of the nearest peak, ensuring that the final location falls on the moment of energy concentration in a physical sense.

[0100] Step 35: Global matching and optimization of feature interval time;

[0101] Calculate all possible time differences using a global matching method, and take the largest time difference as the effective duration for the rolling element to pass through the damaged area. ,Right now:

[0102] (11)

[0103] in, .

[0104] Step four: Statistical inference is performed on batch samples using kernel density estimation to eliminate random errors and noise interference from a single extraction, so as to achieve accurate estimation of damage size.

[0105] Introducing kernel density estimation, the calculated values ​​for batch pulses are... Perform statistical analysis, The most reliable time interval for the rolling element to pass through the damage region, obtained after kernel density estimation, is first calculated by analyzing the batch pulse signals. First, obtain the sample set, then substitute the samples into the kernel density estimation formula to obtain the probability density curve, and finally take the distribution center from the probability density curve to obtain the probability density curve.

[0106] This embodiment uses the Silverman rule, whose kernel estimate of the probability density function is defined as:

[0107] (12)

[0108] Where K is the Gaussian kernel function and h is the adaptive bandwidth; The time interval of the i-th fault pulse is represented by M; M is the number of fault pulses.

[0109] After obtaining the probability density curve, the distribution center is used as the final quantitative calculation basis to obtain the most reliable time interval for the rolling element to pass through the damage area. .

[0110] Its defect size The calculation formula is:

[0111] (13)

[0112] Among them, f r D is the rotational speed frequency. p d is the bearing pitch diameter, and d is the rolling element diameter.

[0113] This invention utilizes experimental data from quantitative detection tests of local defects in the raceway of rolling bearings, covering two typical damage conditions: the inner and outer rings of the bearing. Figure 2 As shown, all raceway damage was prepared using precision electrical discharge wire cutting technology. The groove depth was uniformly set to 1.0 mm, while the width ranged from 1.0 mm to 2.2 mm (with a step size of 0.4 mm) to simulate local defects of different degrees.

[0114] The experiment was conducted using the ABLT-1A bearing simulation platform, which combines fault simulation and full-life fatigue analysis capabilities. The tested bearing was an HRB 6206 deep groove ball bearing (parameters detailed in Table 1), installed in a bearing housing integrated with an AI002 vibration sensor. A sampling frequency of 128 kHz was set to capture subtle damage characteristics in high-frequency impacts. Data was acquired at four speeds: 1200 r / min, 1500 r / min, 1800 r / min, and 2400 r / min. Specific parameters of the 6206 rolling bearing are shown in Table 1.

[0115] Table 1 Parameters of HRB 6206 Deep Groove Ball Bearing

[0116]

[0117] First, the automated fault pulse segmentation verification is illustrated using test results of different defect sizes on the inner ring at 1800 rpm as an example. This is achieved through... Feature envelope operations and spatial domain smoothing integrate discrete curvature abrupt changes into a continuous feature energy region. The adaptive dynamic threshold in Equation (2) is used to screen the effectiveness of the pulses. The value of P in Equation (2) is 90 to ensure that enough effective fault pulses can be automatically extracted in batches. Figure 3 This represents the automated segmentation and extraction result of the signal pulse from a bearing with a 1.0mm damage fault. The width of the red box represents the length of parameter N in the algorithm. Figure 4 for Figure 3 The visualization results of the first eight individual pulses are shown, with N selected as 500 sampling points to cover each individual fault pulse. This allows for precise location and automatic batch extraction of individual fault pulses. Pulses with weak impact amplitudes are discarded, laying the foundation for the next step of automated batch calculation of damage dimensions.

[0118] Secondly, the automated ΔT extraction method based on frequency domain masking was verified. Using full-frequency domain instantaneous energy analysis and the analysis results obtained through automated ΔT extraction, it was found that the positioning time interval ΔT showed a significant deviation, with the final calculated result being 1.25 mm, which differed considerably from the actual size. Therefore, in the automated ΔT extraction, appropriate peak significance coefficients α1 and α2 are needed to ensure the accuracy of the algorithm. This patent sets α1 and α2 to 3 and 5, respectively. To highlight the changes in low-frequency energy and weaken the influence of the high-frequency part, a frequency domain mask was used to improve the detection effect. The low-frequency band of the pulse's full-band instantaneous energy is concentrated around 10 kHz, and f is set... mask = 10 kHz frequency mask. The instantaneous energy obtained from this. The frequency mask improves the accuracy of entry / exit timing, with a final calculated result of 0.97 mm, an error of 3% compared to the actual value. Figure 5 and Figure 6 The automatic identification and calculation results of local damage dimensions of 1.0 mm, 1.4 mm, 1.8 mm, and 2.2 mm in the inner raceway at a speed of 1800 rpm are presented respectively. Figure 6 In (e), the automatic identification algorithm directly locates the entry time to the impact location. After applying the corresponding instantaneous energy frequency domain mask, the accuracy of automatic identification of ΔT is significantly improved, and the accuracy of fault calculation results for all four sizes is satisfactory. The above results fully demonstrate that the frequency domain masking strategy proposed in this patent has high accuracy in automatically extracting entry and exit features.

[0119] Finally, the probability density estimation of local defect sizes in bearings based on kernel density (KDE) is presented. The probability distributions of the four local defect sizes in the inner ring at 1800 rpm are shown below. Figure 7 As shown in the figure, 60 effective pulse calculation samples were used for each defect size to ensure that the estimated probability density function could reflect the true physical characteristics of bearing damage. The results show that the probability density curves for the four sizes exhibit significant main peak guidance, and for all sizes, the main peak of KDE is more accurately locked near the true defect size than the arithmetic mean. From the distribution pattern, although there are a few secondary peaks or fluctuations in the probability density curves, this reflects the random characteristics of bearing vibration signals under real-world conditions. In contrast, the KDE strategy can automatically identify and extract local features with the highest probability density in the sample space, giving the algorithm the ability to automatically mask discrete failure samples. This statistically demonstrates the effectiveness of the frequency domain masking enhancement strategy and achieves maximum likelihood estimation of local defect sizes. The calculation error for the four sizes is controlled within 3%, significantly improving the reliability and stability of automated quantitative diagnosis.

[0120] Figure 8This paper presents the quantitatively estimated errors under the method proposed in this patent for all operating conditions in this test (four speeds and four local defect sizes for the inner and outer rings, totaling 32 types). The calculation sample for each operating condition is also 60 automatically extracted valid fault pulses. Figure 8 The results in (a) of the inner circle indicate that the quantitative assessment error using KDE is smaller, stable at around 5%, while the error calculated using the arithmetic mean has a larger dispersion, especially at higher speeds of 1800 rpm and 2400 rpm. Figure 8 In the results of outer circle (b), the proposed method still demonstrates significantly better accuracy and robustness than the arithmetic mean. These results fully demonstrate the good adaptability of the proposed method under different fault locations and a wide range of speed fluctuations, and it is expected to meet the needs of rapid, automated, precise, and quantitative diagnosis in practical engineering.

[0121] This invention provides an automated calculation method for the quantitative assessment of local defect dimensions in rolling bearings. It should be noted that there are various technical paths to achieve this solution, and the above description is only one preferred embodiment of the invention. For those skilled in the art, various improvements, modifications, or substitutions can be made to the above embodiments without departing from the basic principles of the invention. These improvements and modifications should also fall within the protection scope of this invention. Furthermore, all components not explicitly described in this embodiment can be implemented with reference to existing technologies.

Claims

1. An automated calculation method for quantitative evaluation of the size of local defects in rolling bearings, characterized in that, The steps include the following: S1, The vibration time-domain pulse signal x(n) of the rolling bearing under working conditions is obtained through a vibration sensor; S2, using the Laplace operator combined with a dynamic adaptive threshold determination method, processes the vibration time-domain pulse signal x(n) to achieve automatic location and batch extraction of fault pulses; S3. Based on the extracted fault pulse, a frequency domain mask is introduced, and the instantaneous energy at the entry and exit times is corrected using the SPWVD distribution. After obtaining the corrected instantaneous energy curve, the Laplace operator is used to detect abrupt boundary changes in instantaneous energy, automatically locating potential entry and exit times of damage, and obtaining the effective duration of the rolling element passing through the damage region. ; S4, calculate the effective duration by batch pulse signals. Obtain a sample set, and use kernel density estimation to perform statistical inference on the sample set to obtain an accurate estimate of the damage size.

2. The automated calculation method for quantitative evaluation of local defect dimensions in rolling bearings according to claim 1, characterized in that, The steps for automatically locating and batch extracting fault pulses in long-time time-domain signals include: S21, construct the characteristic envelope function based on the Laplace operator to determine the position of each impact peak; define the calculation result L(n) and envelope function Elap(n) of the pulse signal x(n) as follows: ， ， Where w(n) is a normalized rectangular window function; when written in discrete form, 2k+1 is the width of the window, and i is the summation variable; the window length is selected based on the actual data and is greater than a single period of the high-frequency resonance waveform induced by the damage. S22, Calculate the adaptive threshold Γ based on the global statistical characteristics of the signal; S23, when the envelope feature E lap When >Γ, the potential time of damage impact is marked as t. trig As the trigger time, t trig Starting from W, within the local search interval seg Searching for E inside lap The maximum point of (n): ， Where, n center This is the physical center of the damage pulse; With a defined n center Using the geometric reference, N / 2 sampling points are extended to both the left and right ends to accurately extract discrete pulse samples S from the original signal. k , as a damage impact sample set; ， Where N represents the estimated number of pulse length points.

3. The automated calculation method for quantitative evaluation of local defect dimensions in rolling bearings according to claim 2, characterized in that, The adaptive threshold Γ is obtained by solving the feature envelope function E. lap The cumulative distribution function of (n) is obtained, and its expression is as follows: ， in, Let E be the probability density function of the envelope signal; Q is the envelope feature E. lap The percentile is used to identify the energy concentration interval in the envelope sequence, ensuring that effective pulses are extracted from the original signal.

4. The automated calculation method for quantitative evaluation of local defect dimensions in rolling bearings according to claim 1, characterized in that, The specific implementation steps of step S3 are as follows: S31, defining the frequency domain mask function H(f) as a low-pass filter, then the corrected time-frequency distribution... Represented as: ， in, Let f represent the cutoff frequency f of the SPWVD distribution function H(f). mask Determined by the characteristic response range of the experimental data; S32, based on the time-frequency distribution, calculate the corrected instantaneous energy IE, expressed as: ， in, This represents the lowest frequency at which non-zero energy begins to appear in the time-frequency graph; S33, after obtaining the corrected instantaneous energy curve, uses the Laplace operator. To detect abrupt changes in instantaneous energy boundaries; by designing a double significance threshold, the entry and exit points from damage are determined, expressed as: ， Where, σ n α is the median absolute deviation, used to estimate the noise scale of the instantaneous energy curve; 1、 α² represents the peak significance coefficients at the entry and exit times, respectively; P entry P represents the significance threshold at the time of injury. exit The threshold for the time of departure from the injury is indicated; median() indicates taking the median. S34, for each detected entry or exit point of damage, search for the nearest peak in its neighborhood and update the position of the point to the nearest peak index to ensure that the final location falls on the moment of energy concentration in a physical sense. S35. Calculate all possible time differences using a global matching method, and take the largest time difference as the effective duration for the rolling element to pass through the damaged area. The expression is as follows: ， in, This indicates the time when the damage peak range is reached after correction. This indicates the time after correction to leave the peak damage range. .

5. The automated calculation method for quantitative evaluation of local defect dimensions in rolling bearings according to claim 4, characterized in that, A peak correction strategy is adopted to correct the time of entering / leaving the damage peak range, as shown in the following expression: ， Among them, U entry U exit These are the sets of entry and exit damage peaks, respectively; This indicates the time when the damage peak range is reached after correction. Indicates the time after correction to leave the peak damage range; Represents set U entry At a certain peak moment of damage, Represents set U exit A certain peak moment of damage; This indicates the initial detected moment of entry into the damage point. This indicates the initial detected moment of departure from the damage point.

6. The automated calculation method for quantitative evaluation of local defect dimensions in rolling bearings according to claim 1, characterized in that, In step S4, the steps to obtain an accurate estimate of the damage size are as follows: S41, Substituting the samples into the kernel density estimation formula yields the probability density curve, where the kernel estimate of the probability density function is defined as: ， In the formula, K is the Gaussian kernel function, and h is the adaptive bandwidth; The time interval of the i-th fault pulse is represented by M; M is the number of fault pulses. After obtaining the probability density curve, the distribution center is used as the final quantitative calculation basis to obtain the most reliable time interval for the rolling element to pass through the damage area. ; S42, calculate the defect size according to the following formula. : ， Among them, f r D is the rotational speed frequency. p d is the bearing pitch diameter, and d is the rolling element diameter.

Images