Rolling bearing fault diagnosis method based on improved phase motion estimation and optimized IMCKD

Abstract

The invention discloses a rolling bearing fault diagnosis method based on improved phase motion estimation and optimized IMCKD, and the method comprises the steps: obtaining a video of the operation of a rolling bearing, and obtaining an image frame of the rolling bearing; performing complex number steerable pyramid decomposition on the image frame, and obtaining a multi-scale amplitude image and a multi-scale phase image through decomposition; constructing a scale evaluation index function; according to a scale evaluation index function, calculating a scale evaluation index for each scale amplitude graph; selecting the scale with the highest scale evaluation index value as the optimal scale; according to the phase diagram, performing phase-based motion estimation on the selected optimal scale to obtain a vibration signal of the rolling bearing; taking a rolling bearing vibration signal as an input for optimizing IMCKD, and outputting a fault period enhancement signal; wherein IMCKD optimization takes improved maximum correlation kurtosis deconvolution as a framework, and an improved period estimation method and a two-stage search method are introduced; and carrying out envelope analysis on the fault period enhancement signal so as to identify the characteristic frequency of the fault. According to the invention, fault diagnosis of the rolling bearing is realized in a non-contact-based visual measurement mode.

Description

Technical Field

[0001] This invention relates to a rolling bearing fault diagnosis method based on improved phase motion estimation and optimized IMCKD, belonging to the field of mechanical fault diagnosis technology. Background Technology

[0002] Rolling bearings are critical components of rotating machinery, and their failure is one of the most common types of rotating machinery failure. Accurate detection of rolling bearing faults is therefore crucial. Currently, the common fault diagnosis method involves analyzing the vibration signals generated during bearing operation; therefore, accurately acquiring vibration signals is the core of fault analysis.

[0003] Currently, vibration measurement is mainly divided into two categories based on installation method and measurement principle: contact and non-contact. Contact measurement requires the sensor to be directly mounted on the structure to obtain the dynamic response, but it often suffers from installation difficulties and susceptibility to load effects. For example, it is cumbersome to arrange on large structures, and may introduce load effects on lightweight structures. With the rapid development of computer vision technology, non-contact vibration measurement is widely used due to its advantages such as high accuracy, simple installation, and no additional load. Among them, the tracking method is a common non-contact method, which estimates motion by locking feature points and tracking their displacement changes over time. However, it requires the feature points to have obvious texture or sufficient amplitude; accuracy is low when the amplitude is small or the texture is not obvious. In contrast, optical flow methods in non-contact measurement are techniques that estimate subpixel-level motion by analyzing continuous image frames. Among them, Horn–Schunck (HS) and Lucas–Kanade (LK) are typical intensity-based methods. Intensity-based optical flow is susceptible to changes in illumination and noise interference, leading to inaccurate displacement estimation and affecting fault feature identification. To improve robustness, phase-based optical flow estimation has been proven to be a more reliable approach. Compared to intensity-based methods, this method is insensitive to changes in illumination and noise, and exhibits high robustness. Nevertheless, phase-based motion estimation still has limited systematic applications in the measurement and diagnosis of rolling bearing vibration.

[0004] Rolling bearing vibration signals are rich in information but difficult to analyze, with periodic pulses being a key indicator of faults. Blind deconvolution (MED) is an effective filtering approach, constructing an adaptive inverse filter by optimizing the objective function to highlight fault characteristics. Existing methods target maximum kurtosis but are limited to isolated pulses and unsuitable for periodic pulses. To balance periodicity and impulsivity, subsequent developments have included Maximum correlated kurtosis deconvolution (MCKD) and blind deconvolution based on cyclostationarity criterion (CYCBD). Their performance is highly dependent on prior knowledge. To reduce prior knowledge, improved maximum correlated kurtosis deconvolution (IMCKD) and adaptive maximum second-order cyclostationarity blind deconvolution (ACYCBD) have been introduced. On the one hand, existing period estimation methods are prone to failure under noise and random pulse interference, and remain prone to failure under strong noise conditions. On the other hand, the adaptive selection of filter length often relies on prior knowledge or general optimization algorithms, resulting in low efficiency and sensitivity to parameters, which in turn affects the overall performance of mechanical fault diagnosis.

[0005] In summary, existing blind deconvolution filtering frameworks still have two shortcomings: First, the period estimation is not robust to noise and random impulse interference; second, the selection of filter length often depends on prior knowledge or general optimization algorithms, resulting in low efficiency and sensitivity to parameters, thus affecting the overall performance of mechanical fault diagnosis. Therefore, this invention proposes a rolling bearing fault diagnosis method based on improved phase motion estimation and optimized IMCKD (Optimized Improved Maximum Correlated Kurtosis Deconvolution, OIMCKD). Summary of the Invention

[0006] This invention provides a rolling bearing fault diagnosis method based on improved phase motion estimation and optimized IMCKD, which realizes rolling bearing fault diagnosis in a non-contact visual measurement manner.

[0007] The technical solution of this invention is:

[0008] According to a first aspect of the present invention, a method for diagnosing rolling bearing faults based on improved phase motion estimation and optimized IMCKD is provided, comprising: acquiring video of rolling bearing operation to obtain image frames of the rolling bearing; performing complex-turnable pyramid decomposition on the image frames to obtain multi-scale amplitude maps and phase maps; constructing a scale evaluation index function; calculating scale evaluation indices for each scale amplitude map based on the scale evaluation index function; selecting the scale with the highest scale evaluation index value as the optimal scale; performing phase-based motion estimation on the selected optimal scale based on the phase map to obtain the vibration signal of the rolling bearing; using the rolling bearing vibration signal as input to optimized IMCKD to output a fault period enhancement signal; wherein, optimized IMCKD is based on improved maximum correlation kurtosis deconvolution and introduces an improved period estimation method and a two-stage search method; the improved period estimation method is used to estimate the number of period samples of the rolling bearing vibration signal to obtain the estimated number of period samples, and the two-stage search method is used to select the optimal filter length of the improved maximum correlation kurtosis deconvolution; and envelope analysis is performed on the fault period enhancement signal to identify the characteristic frequency of the fault.

[0009] Furthermore, the scale evaluation index function is expressed as follows:

[0010] .

[0011] In the formula, Indicates the first Scale evaluation index for scale amplitude map; median is the median; j is the scale index; k is the number of directions for complex-to-oriented pyramid decomposition; For the first Frame image frame at scale j, orientation The weights of the subbands on the surface; For the first Frame image frame at scale j, orientation The score of the sub-band; This is the set of frames used for evaluation.

[0012] Furthermore, the scale evaluation index function is defined by measurement points. Centered, scale-dependent Window expanded proportionally to wavelength :

[0013] .

[0014] In the formula, Let be the window radius at scale j; ; , The maximum scale number; The radius of the initial window.

[0015] Further, the improved period estimation method specifically involves: calculating the square envelope and autocorrelation spectrum of the rolling bearing vibration signal; selecting the top V local maxima with the largest peak amplitude within the interval after the first zero crossing of the autocorrelation spectrum, and using the abscissa as the set of candidate period samples; calculating the period-enhanced harmonic signal-to-noise ratio (SNR) for each candidate period sample in the set; first calculating the mean SNR of all candidate period samples in the set, and then retaining the number of candidate period samples whose SNR is greater than the mean, thus obtaining a retention set; checking whether there is an integer multiple relationship between the candidate period samples in the retention set: if so, selecting the smallest among the candidate period samples that satisfy the integer multiple relationship as the estimated period sample number; if no integer multiple relationship is detected, directly selecting the candidate period sample number with the largest SNR in the retention set as the estimated period sample number.

[0016] Furthermore, the expression for the periodically enhanced harmonic signal-to-noise ratio is established based on the autocorrelation spectrum and the number of candidate periodic samples.

[0017] Furthermore, the two-stage search method specifically involves: firstly, performing a coarse search within the first filter length interval with a first preset step size, calculating the fusion index one by one, and recording the filter length corresponding to the maximum value of the fusion index within the first filter length interval as... Then with A fine search is performed within the second filter length interval, centered on a second preset step size. The filter length corresponding to the maximum value of the fusion index within the second filter length interval is denoted as . and will The optimal filter length is defined as follows: the first preset step size is greater than the second preset step size.

[0018] Furthermore, the fusion index expression is established based on the correlation kurtosis and the square envelope harmonic noise signal-to-noise ratio.

[0019] According to a second aspect of the present invention, a rolling bearing fault diagnosis system based on improved phase motion estimation and optimized IMCKD is provided, comprising modules of any of the methods described above.

[0020] According to a third aspect of the present invention, a terminal is provided, comprising a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the steps of any of the methods described above.

[0021] The beneficial effects of this invention are:

[0022] 1. This invention can promote the application of non-contact visual measurement in the field of bearing fault diagnosis, and provides a new method for mechanical fault diagnosis in complex industrial environments.

[0023] 2. This invention proposes a method to accurately extract phase information by adaptively selecting an appropriate scale level based on the amplitude map and the phase map, thereby improving the accuracy of phase-based motion estimation methods.

[0024] 3. This invention proposes a more robust period estimation method within the improved maximum correlation kurtosis deconvolution framework to enhance anti-interference capability under noise and impulse interference conditions; finally, to achieve adaptive optimization of filter length, a two-stage search strategy is further proposed to reduce dependence on prior knowledge and improve computational efficiency. Attached Figure Description

[0025] Figure 1 This is a flowchart of the steps of the present invention.

[0026] Figure 2 This is a schematic diagram of the experimental platform for an example of the present invention.

[0027] Figure 3 The faulty bearing used in this embodiment of the invention.

[0028] Figure 4 This is the first frame captured by the high-speed camera in this embodiment of the invention.

[0029] Figure 5 The vibration signal is an embodiment of the present invention. Figure 5 (a) is a time-domain diagram of the vibration signal extracted from the video by the method of the present invention. Figure 5 (b) is the vibration signal envelope spectrum extracted from the video by the method of the present invention. Figure 5 (c) is a time-domain plot of the vibration signal acquired by the vibration sensor. Figure 5 (d) is the envelope spectrum of the vibration signal collected by the vibration sensor.

[0030] Figure 6 The graph shows the changes in the original period estimate and the period estimate in this embodiment of the invention with the number of iterations.

[0031] Figure 7 This is a filter length search graph in an example of the present invention. Figure 7 (a) Coarse search phase, Figure 7 (b) Fine search phase.

[0032] Figure 8 This is a comparison diagram of examples of the present invention and different methods. Figure 8 (a) is the time-domain plot of IMCKD. Figure 8 (b) is the envelope spectrum of IMCKD. Figure 8(c) Time-domain plot of ACYCBD Figure 8 (d) Envelope spectrum of ACYCBD, Figure 8 (e) Time-domain diagram of the method of the present invention, Figure 8 (f) Envelope spectrum of the method of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other.

[0034] Example 1: As Figures 1-8 As shown, according to a first aspect of the present invention, a method for diagnosing rolling bearing faults based on improved phase motion estimation and optimized IMCKD is provided, comprising:

[0035] S1. Use a high-speed industrial camera to capture video of the rolling bearing in operation, and obtain image frames of the rolling bearing.

[0036] S2. Perform complex steerable pyramid (CSP) decomposition on the image frames to obtain multi-scale amplitude and phase maps; construct a scale evaluation index function; calculate the scale evaluation index for each scale amplitude map based on the scale evaluation index function; select the scale with the highest scale evaluation index value as the optimal scale.

[0037] The Complex Turnable Pyramid Decomposition (CSP) decomposes an image into subbands of several scales and orientations, each subband being equivalent to a filter responding at a specific central spatial frequency and orientation. When the central frequency and orientation of a subband coincide with the local structure, the amplitude of that subband increases significantly at that location; otherwise, the response weakens. Therefore, comparing subband amplitudes across scales at the same pixel location can reflect which scale best matches the target structure (measurement point). Thus, a method is proposed that uses the amplitude map from the CSP decomposition process as a reference to obtain the appropriate scale for motion estimation.

[0038] The scale evaluation index function is expressed as follows:

[0039] ;

[0040] In the formula, Indicates the first Scale evaluation index for scale amplitude map; median is the median; j is the scale index; k is the number of directions in the complex turnable pyramid decomposition (CSP). For the first Frame image frame at scale j, orientation The weights of the subbands on the surface; For the first Frame image frame at scale j, orientation The score of the sub-band; The set of frames used for evaluation (in this embodiment of the invention, the number of frames in the set is 100) is used as the median of multiple frames as the final evaluation index to achieve the purpose of suppressing the anomalies of occasional frames.

[0041] The weights and scores of the sub-bands are expressed as follows:

[0042] ;

[0043] ;

[0044] in, For the first Measurement points on the frame of the image In scale j, direction The amplitude of the sub-band; mean is the average value; This represents the absolute deviation of the median.

[0045] Considering that CSP decomposition achieves multi-scale response through bandpass filters with different center frequencies, its spatial domain supports scaling with increasing scale; to maintain comparable local statistics at different scales, a method is defined with measurement points... Centered, scale-dependent Window expanded proportionally to wavelength :

[0046] ;

[0047] In the formula, Let be the window radius at scale j; ; , This is the maximum scale number (determined based on the image resolution); The radius of the initial window is [missing information] in this invention. Take 3.

[0048] S3. Based on the phase diagram, perform phase-based motion estimation on the optimal scale selected in S2 to obtain the vibration signal of the rolling bearing;

[0049] The phase-based motion estimation is specifically as follows:

[0050] Assuming that the amplitude and phase of an image frame are obtained by CSP decomposition, respectively, they are expressed as follows: and The same measuring point along its pixel trajectory The phase remains constant over time.

[0051] ;

[0052] in, It is a constant. Taking the derivative of the above equation with respect to time t, we get:

[0053] ;

[0054] in, and These represent the velocities in the horizontal and vertical directions of the phase profile, respectively; further, the velocity projected onto the spatial phase gradient direction can be obtained:

[0055] ;

[0056] in, The spatial gradient representing the phase, It is the projection of velocity onto the phase direction. Then, it is performed over time. The pixel displacement can be obtained by integrating the integral. If the obtained image is not distorted, the actual displacement can be obtained by using the appropriate scaling factor.

[0057] S4. The rolling bearing vibration signal obtained in S3 is used as the input of the optimized IMCKD, and the fault period enhancement signal is output. The optimized IMCKD is based on the improved maximum correlation kurtosis deconvolution and introduces an improved period estimation method and a two-stage search method. The improved period estimation method is used to estimate the number of period samples of the rolling bearing vibration signal to obtain the estimated number of period samples. The two-stage search method is used to select the optimal filter length of the improved maximum correlation kurtosis deconvolution.

[0058] S5. Perform envelope analysis on the fault cycle enhancement signal obtained in S4 to identify the characteristic frequency of the fault.

[0059] Furthermore, the improved period estimation method is specifically as follows:

[0060] Calculate the square envelope and autocorrelation spectrum of the rolling bearing vibration signal. Within the interval after the first zero crossing of the autocorrelation spectrum, select the top V local maxima with the largest peak amplitude, and use the abscissa as the set of candidate periodic sample numbers; for example, V=5.

[0061] The periodic enhanced harmonic signal-to-noise ratio is proposed as an evaluation index of periodic effectiveness; the periodic enhanced harmonic signal-to-noise ratio is calculated for each candidate period sample in the candidate period sample set; the periodic enhanced harmonic signal-to-noise ratio is established based on the autocorrelation spectrum and the number of candidate period samples.

[0062] Considering the possibility of mistakenly selecting an overtone of the true period as the final number of period samples, the mean of the period-enhanced harmonic signal-to-noise ratio (PEHNR) for all candidate period samples in the candidate period sample set is first calculated. Then, the number of candidate period samples whose PEHNR is greater than the mean is retained to obtain the retained set. It is then checked whether there is an integer multiple relationship between the number of candidate period samples in the retained set: if there is, the smallest one among the candidate period samples that satisfy the integer multiple relationship is selected as the estimated number of period samples; if no integer multiple relationship is detected, the number of candidate period samples with the largest PEHNR in the retained set is directly selected as the estimated number of period samples.

[0063] Since random pulses are not periodic, this invention proposes Periodically Enhanced Harmonic Signal-to-Noise Ratio (PEHNR) as an evaluation index for periodicity effectiveness. The PEHNR uses a squared envelope instead of a normal envelope to enhance fault characteristics. The PEHNR is defined as follows:

[0064] ;

[0065] in, , Harmonic order; The square envelope of the vibration signal is in Corresponding delay The autocorrelation spectrum at the octaves, The square envelope of the vibration signal is in The autocorrelation spectrum at the time delay, denoted as the number of candidate periodic samples, and T as the autocorrelation length involved in the measurement.

[0066] Furthermore, the two-stage search method is specifically as follows:

[0067] First, in the first filter length interval [ L m i n , L m a x ] Within the first preset step size A coarse search is performed, and the fusion index is calculated one by one. The filter length corresponding to the maximum value of the fusion index within the first filter length range is denoted as... Then with Centered in the second filter length interval [ L o − S 1 , L o + S 1 ] A fine search is performed using a second preset step size, and the filter length corresponding to the maximum value of the fusion index within the second filter length range is denoted as . and will The optimal filter length is defined as follows: the first preset step size is larger than the second preset step size; the coarse-to-fine search strategy described above can effectively improve selection efficiency. For example, the first preset step size... Set the value to 10, and the second preset step size to 1.

[0068] This invention proposes a fusion metric driven by both correlated kurtosis (CK) and squared envelope harmonic noise signal-to-noise ratio (SEHNR).

[0069] ;

[0070] in, This represents the normalization operator. To optimize the correlation kurtosis value obtained during IMCKD convergence; the squared envelope harmonic noise signal-to-noise ratio (SEHNR) is defined as:

[0071] ;

[0072] in, The number of periodic samples estimated using the improved period estimation method. The autocorrelation spectrum of the squared envelope of the vibration signal at a time delay of 0 is given. The square envelope of the vibration signal is in The autocorrelation spectrum at the corresponding time delay.

[0073] As can be seen from the above, the method of the present invention can be used for fault diagnosis of faulty rolling bearings.

[0074] The improved maximum correlation kurtosis deconvolution is specifically as follows:

[0075] First, the initial estimated number of period samples of the vibration signal is calculated. Then, the maximum number of iterations is set, the filter coefficients are initialized based on the filter length, and the iteration process begins.

[0076] ;

[0077] in, This is the output of the inverse filter. For input signal, denoted as , where L is the filter coefficient and L is the filter length.

[0078] ;

[0079] Where P is the fault cycle that needs to be detected; Sampling rate, This represents the number of periodic samples, i.e., the corresponding number of sampling points.

[0080] In each iteration, the enhanced fault cycle signal output after filtering is calculated and relevant parameters are obtained. Then, based on the updated estimated number of cycle samples and the corresponding signal matrix, new filter coefficients are calculated. This process is repeated until the maximum number of iterations is reached. Finally, the signal with the largest kurtosis value is selected as the output signal from all filtered signals. Finally, the output signal is subjected to spectral and envelope spectrum analysis. The iterative closed-form general expression for the filter coefficients is (in the general expression...) Right now ; Right now ):

[0081] ;

[0082] in, Z r = [ z 1 − r z 2 − r ⋯ z N − r 0 z 1 − r ⋯ z N − 1 − r ⋮ ⋮ ⋱ ⋮ 0 0 ⋯ z N − L + 1 − r ] L × N ， r = m Q s , m = 0 , 1 , … , M ； The shift order is 1 in this embodiment of the invention. For the number of periodic samples, N The number of samples in the signal; α m = [ u 1 − m Q s − 1 ( u 1 2 u 1 − Q s 2 ⋯ u 1 − M Q s 2 ) u 2 − m Q s − 1 ( u 2 2 u 2 − Q s 2 ⋯ u 2 − M Q s 2 ) ⋮ u N − m Q s − 1 ( u N 2 u N − Q s 2 ⋯ u N − M Q s 2 ) ] , β = [ u 1 u 1 − Q s ⋯ u 1 − M Q s u 2 u 2 − Q s ⋯ u 2 − M Q s ⋮ ⋮ ⋱ ⋮ u N u N − Q s ⋯ u N − M Q s ] .

[0083] Furthermore, experiments were conducted on a rotating machinery simulation fault test bench.

[0084] I. The test bench consists of a drive motor, the rolling bearing under test, and a control console, such as... Figure 2 As shown. The faulty bearing was simulated using electrical discharge machining, such as... Figure 3 As shown, after the experimental platform stabilized, a high-speed industrial camera was used to collect video of the bearing's operation. The first two seconds of video were used for analysis in the actual study. Simultaneously, a vibration sensor was used to synchronously collect vibration signals as a reference, and the light source was adjusted to ensure image quality. Detailed experimental parameters are shown in Table 1. The geometric parameters of the tested bearing N204 EM are shown in Table 2. Based on the parameters in Table 2, the theoretical outer ring fault characteristic frequency is calculated to be approximately 169.44 Hz. Considering the slight fluctuations in actual rotational speed, the measured characteristic frequency may deviate slightly from the theoretical value, which is acceptable. The following is the expression for the theoretical outer ring fault characteristic frequency:

[0085] ;

[0086] in, The number of rollers, Where is the diameter of the roller. The diameter of the pitch circle. Contact angle, This refers to the rotational frequency of the drive motor.

[0087] Table 1 Experimental parameters

[0088]

[0089] Table 2 Basic Parameters of N 204 EM

[0090]

[0091] Second, the amplitude map and phase map of the video frame are obtained through CSP decomposition; then the scale evaluation index of each scale is calculated based on the amplitude map; finally, the scale with the highest scale evaluation index value is selected as the scale for motion estimation, i.e., the optimal scale.

[0092] The first frame obtained from the video is as follows Figure 4 As shown, in this experiment, the measurement points were expanded into a region, namely the area selected by the green box. The green box represents the region for subsequent vibration measurements. After CSP decomposition, the acquired bearing fault video was divided into four effective scales. The scores of each scale are shown in Table 3. The fourth scale had the highest evaluation index value, with a score of 2.5472.

[0093] Table 3 Scores at different scales in bearing failure videos

[0094]

[0095] Third, based on the phase diagram, phase-based motion estimation is performed on the selected optimal scale to obtain the vibration signal of the rolling bearing.

[0096] IV. To verify the accuracy of the vibration signal extracted by this invention, the vibration signal obtained by this invention is compared with the signal collected by the vibration sensor, such as... Figure 5 As shown, the outer ring fault characteristic frequency extracted by the method of the present invention is 169.00Hz, while the sensor result is 169.67Hz, with a deviation of 0.67Hz. The error is within a controllable range, indicating that the method of the present invention can accurately identify the bearing fault characteristic frequency.

[0097] 5. Using the rolling bearing vibration signal as input to optimize IMCKD, the output fault cycle enhancement signal is used; specifically: the vibration signal extracted from the video is used to estimate the initial number of cycle samples, and the final number of cycle samples is estimated to be 59 (as the number of iterations increases, the original cycle estimation method failed to find the correct cycle even after 30 iterations, while the method proposed in this invention found the final number of cycle samples after the 10th iteration), as shown. Figure 6 As shown. A two-stage search is performed to select a suitable filter length, and the search results are as follows. Figure 7 As shown, the appropriate filter length obtained in the final coarse search stage is 140, and the final filter length determined in the fine search stage is 141. The search process is as follows: Figure 7 As shown.

[0098] VI. Envelope analysis is performed on the fault periodic enhancement signal to identify the characteristic frequencies of the fault. To highlight the effectiveness of the proposed method, the currently mainstream blind deconvolution algorithms, Improved Maximum Correlation Kurtosis Deconvolution (IMCKD) and Adaptive Maximum Cyclic Stationary Blind Deconvolution (ACYCBD), are selected for comparison. The results obtained are... Figure 8 As shown. For quantitative evaluation, the Hilbert envelope spectrum fault eigenvalue ratio was used. As an evaluation index, different methods are compared and analyzed. This index measures the proportion of the fault fundamental frequency and its harmonics in the envelope spectrum; a higher value indicates clearer fault characteristics and a cleaner background. Its definition is:

[0099] ;

[0100] in, The number of harmonics included (7 in this invention), For 1, 2, 3, ..., ; The amplitude of the envelope spectrum. Pick ; The outer ring fault characteristic frequency identified by this invention is shown in Table 4. The envelope spectrum of the original signal and the envelope spectrum output after the original signal has undergone three blind deconvolution methods (IMCKD, ACYCBD, and the method of this invention) are quantitatively evaluated.

[0101] Table 4. Based on visual vibration conversion signals Value comparison

[0102]

[0103] According to a second aspect of the present invention, a rolling bearing fault diagnosis system based on improved phase motion estimation and optimized IMCKD is provided, comprising modules of any of the methods described above. Specifically, the module includes: a first module for acquiring video of the rolling bearing in operation, obtaining image frames of the rolling bearing; a second module for performing complex-turnable pyramid decomposition on the image frames, obtaining multi-scale amplitude and phase maps; constructing a scale evaluation index function; calculating scale evaluation indices for each scale amplitude map based on the scale evaluation index function; and selecting the scale with the highest scale evaluation index value as the optimal scale; a third module for performing phase-based motion estimation on the selected optimal scale based on the phase map, obtaining the vibration signal of the rolling bearing; a fourth module for using the rolling bearing vibration signal as input to an optimized IMCKD, outputting a fault period enhancement signal; wherein, the optimized IMCKD uses an improved maximum correlation kurtosis deconvolution as a framework, and introduces an improved period estimation method and a two-stage search method, wherein the improved period estimation method is used to estimate the number of period samples of the rolling bearing vibration signal to obtain the estimated number of period samples, and the two-stage search method is used to select the optimal filter length for the improved maximum correlation kurtosis deconvolution; and a fifth module for performing envelope analysis on the fault period enhancement signal to identify the characteristic frequencies of the fault. For any parts of the modules not described in detail above, please refer to the relevant descriptions in the embodiments.

[0104] According to a third aspect of the present invention, a terminal is provided, comprising a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the steps of any of the methods described above.

[0105] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for diagnosing rolling bearing faults based on improved phase motion estimation and optimized IMCKD, characterized in that, include: Acquire video of the rolling bearing in operation to obtain image frames of the rolling bearing; The image frames are decomposed into a complex turnable pyramid to obtain multi-scale amplitude and phase maps; a scale evaluation index function is constructed; based on the scale evaluation index function, the scale evaluation index is calculated for the amplitude maps at each scale; the scale with the highest scale evaluation index value is selected as the optimal scale. Based on the phase diagram, phase-based motion estimation is performed on the selected optimal scale to obtain the vibration signal of the rolling bearing; The rolling bearing vibration signal is used as the input of the optimized IMCKD, and the output is a fault period enhancement signal. The optimized IMCKD is based on the improved maximum correlation kurtosis deconvolution and introduces an improved period estimation method and a two-stage search method. The improved period estimation method is used to estimate the number of period samples of the rolling bearing vibration signal to obtain the estimated number of period samples. The two-stage search method is used to select the optimal filter length of the improved maximum correlation kurtosis deconvolution. Envelope analysis is performed on the fault periodic enhancement signal to identify the characteristic frequency of the fault.

2. The rolling bearing fault diagnosis method based on improved phase motion estimation and optimized IMCKD according to claim 1, characterized in that, The scale evaluation index function is expressed as follows: ； In the formula, Indicates the first Scale evaluation index for scale amplitude map; median is the median; j is the scale index; k is the number of directions for complex-to-oriented pyramid decomposition; For the first Frame image frame at scale j, orientation The weights of the subbands on the surface; For the first Frame image frame at scale j, orientation The score of the sub-band; This is the set of frames used for evaluation.

3. The rolling bearing fault diagnosis method based on improved phase motion estimation and optimized IMCKD according to claim 2, characterized in that, The scale evaluation index function is defined by measurement points. Centered, scale-dependent Window expanded proportionally to wavelength : ； In the formula, Let be the window radius at scale j; ; , The maximum scale number; The radius of the initial window.

4. The rolling bearing fault diagnosis method based on improved phase motion estimation and optimized IMCKD according to claim 1, characterized in that, The improved period estimation method is specifically as follows: Calculate the square envelope and autocorrelation spectrum of the rolling bearing vibration signal. In the interval after the first zero crossing of the autocorrelation spectrum, select the top V local maxima with the largest peak amplitude, and use the abscissa as the set of candidate period sample numbers. Calculate the periodic enhanced harmonic signal-to-noise ratio for each candidate period sample in the candidate period sample set; First, calculate the mean signal-to-noise ratio (SNR) of the periodic enhanced harmonics (HMR) of all candidate periodic sample numbers in the candidate periodic sample number set. Then, retain the number of candidate periodic sample numbers that satisfy the condition that the SNR of the periodic enhanced harmonics is greater than the mean value to obtain the retention set. Check whether there is an integer multiple relationship between the number of candidate periodic sample numbers in the retention set. If there is, select the smallest number among the candidate periodic sample numbers that satisfy the integer multiple relationship as the estimated number of periodic sample numbers. If no integer multiple relationship is detected, the candidate period sample number with the largest period-enhanced harmonic signal-to-noise ratio in the retained set is directly selected as the estimated number of period samples.

5. The rolling bearing fault diagnosis method based on improved phase motion estimation and optimized IMCKD according to claim 4, characterized in that, The expression for the periodically enhanced harmonic signal-to-noise ratio is established based on the autocorrelation spectrum and the number of candidate periodic samples.

6. The rolling bearing fault diagnosis method based on improved phase motion estimation and optimized IMCKD according to claim 1, characterized in that, The two-stage search method specifically involves: firstly, performing a coarse search within the first filter length interval with a first preset step size, calculating the fusion index one by one, and recording the filter length corresponding to the maximum value of the fusion index within the first filter length interval as... Then with A fine search is performed within the second filter length interval, centered on a second preset step size. The filter length corresponding to the maximum value of the fusion index within the second filter length interval is denoted as . and will The optimal filter length is defined as follows: the first preset step size is greater than the second preset step size.

7. The rolling bearing fault diagnosis method based on improved phase motion estimation and optimized IMCKD according to claim 6, characterized in that, The expression for the fusion index is established based on the correlation kurtosis and the square envelope harmonic noise signal-to-noise ratio.

8. A rolling bearing fault diagnosis system based on improved phase motion estimation and optimized IMCKD, characterized in that, The module includes the method described in any one of claims 1-7.

9. A terminal, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-7.

Images