Rolling bearing vibration signal frequency domain resampling method and system

Abstract

The invention provides a rolling bearing vibration signal frequency domain resampling method and system, and relates to the technical field of fault detection, and the method comprises the steps: collecting a vibration signal of a rolling bearing; through a Gabor expansion algorithm, the vibration signal is expanded, and a Gabor time-frequency graph is obtained; through a C-ICEF algorithm, extracting a target order component from the Gabor time-frequency diagram, and performing polynomial fitting on the target order component to obtain a filtering center frequency line; calculating a Gabor coefficient of the target order component through a masking algorithm according to the filtering center frequency line, and obtaining a time-frequency spectrum of the target order component; performing instantaneous frequency estimation on the target order component through a local extremum search algorithm according to the time-frequency spectrum, and performing quadratic fitting to obtain an instantaneous frequency function; and calculating a key phase time scale of equal-angle resampling through the instantaneous frequency function, and carrying out equal-angle resampling according to the key phase time scale.

Description

Technical Field

[0001] This invention relates to the field of fault detection technology, and in particular to a method and system for frequency domain resampling of rolling bearing vibration signals. Background Art

[0002] There is increasing focus on equipment health management and fault diagnosis in industrial production. Rolling bearings, as a crucial component widely used in industrial production, directly impact production efficiency, operating costs, and safety. Especially with the diversification, complexity, and scale of industrial equipment, the application scenarios for rolling bearings are becoming more extensive.

[0003] In practical applications, there are complex vibration interferences between rolling bearings or between their own components, which causes the collected vibration signals to contain a large amount of interference noise. When the fault is not obvious, the fault characteristic vibration signals are often submerged by noise and difficult to separate.

[0004] In addition, the operating conditions of rolling bearings are often in flux, with irregular variable speed operation being more common than stable operation, which brings great difficulties to vibration signal analysis. Summary of the Invention

[0005] To address the complex vibration interference between traditional rolling bearings or their components, which results in a large amount of noise in the collected vibration signals, making it difficult to separate fault characteristic vibration signals when the fault is not obvious, and because the operating conditions of rolling bearings are often variable, with irregular variable speed operation being more common than stable operation, this presents a significant technical challenge for vibration signal analysis. This invention provides a frequency domain resampling method and system for rolling bearing vibration signals.

[0006] The technical solutions provided by the embodiments of the present invention are as follows: First aspect: This invention provides a frequency domain resampling method for rolling bearing vibration signals, comprising: S1: Acquire vibration signals from rolling bearings; S2: The vibration signal is expanded using the Gabor expansion algorithm to obtain the Gabor time-frequency diagram; S3: Using the C-ICEF algorithm, extract the target order component from the Gabor time-frequency diagram, and perform polynomial fitting on the target order component to obtain the filter center frequency line; S4: Based on the filter center frequency line, calculate the Gabor coefficients of the target order component using a masking algorithm to obtain the time spectrum of the target order component; S5: Using a local extremum search algorithm, based on the time spectrum, the instantaneous frequency of the target order component is estimated, and a second fitting is performed to obtain the instantaneous frequency function; S6: Calculate the key phase timescale for equal-angle resampling using the instantaneous frequency function, and perform equal-angle resampling based on the key phase timescale.

[0007] Second aspect: This invention provides a frequency domain resampling system for rolling bearing vibration signals, comprising: a memory and one or more processors; The memory stores one or more application programs, which are adapted to be executed by the one or more processors to implement the above-described frequency domain resampling method for rolling bearing vibration signals.

[0008] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following: In this invention, the collected vibration signal is expanded using the Gabor expansion algorithm to obtain a Gabor time-frequency diagram. The target order component is extracted from the Gabor time-frequency diagram using the C-ICEF algorithm, and polynomial fitting is performed on the target order component to obtain the filtered center frequency line. This effectively solves the complex vibration interference problem between rolling bearings or between their own components. It can eliminate a large amount of interference noise in the collected vibration signal, avoiding the situation where the fault characteristic vibration signal is submerged by noise and difficult to separate when the fault is not obvious. By calculating the key phase time scale of equal-angle resampling and performing equal-angle resampling according to the key phase time scale, the problem of difficulty in vibration signal analysis caused by irregular variable speed operation of rolling bearings under varying operating conditions is solved. Attached Figure Description

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A flowchart illustrating a frequency domain resampling method for rolling bearing vibration signals provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a frequency domain resampling system for rolling bearing vibration signals provided in an embodiment of the present invention. Detailed Implementation

[0011] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0012] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0013] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0014] Reference Manual Figure 1 , shows a flow chart of a method for frequency domain resampling of rolling bearing vibration signals provided by an embodiment of the present invention.

[0015] This invention provides a frequency domain resampling method for rolling bearing vibration signals. This method can be implemented using a rolling bearing vibration signal frequency domain resampling device, which can be a terminal or a server. The processing flow of the rolling bearing vibration signal frequency domain resampling method may include the following steps: S1: Collect vibration signals from rolling bearings.

[0016] Optionally, the vibration signal of the rolling bearing is collected by an acceleration sensor.

[0017] It should be noted that the theoretical fault characteristic order of rolling bearings specifically includes: Theoretical fault characteristic order of bearing outer ring:

[0018] in, O bpo This indicates the order of theoretical fault characteristics of the bearing outer ring. n ’’ Indicates the number of balls. d Indicates the diameter of the rolling element. D Indicates the bearing pitch diameter. β Indicates the rolling element contact angle.

[0019] Theoretical fault characteristic order of bearing inner ring:

[0020] in, O bpi Indicates the theoretical fault characteristic order of the bearing inner ring.

[0021] Theoretical failure characteristic order of bearing rolling elements:

[0022] in, O bs This indicates the order of theoretical fault characteristics of the bearing rolling elements.

[0023] Theoretical failure characteristic order of bearing cage:

[0024] in, O c This indicates the order of theoretical fault characteristics of the bearing cage.

[0025] In this invention, the analysis of characteristic orders can identify early fault signals. These signals are often weak in the initial stages, but through theoretical calculations, they can be separated from vibration signals, thereby achieving early fault detection and avoiding significant losses.

[0026] S2: The vibration signal is expanded using the Gabor expansion algorithm to obtain the Gabor time-frequency diagram.

[0027] The Gabor expansion algorithm is as follows:

[0028] in, x L ( k )express k Time in the cycle L Vibration signal after unfolding L Indicates the period of the vibration signal. M Represents the number of samples in the time domain. N Represents the number of frequency domain samples. C m,n Indicates the first m The time-domain sampling location and the first n Gabor coefficients of vibration signals at each frequency domain sampling location h L ( k )express k Time in the cycle L The Gabor basis functions.

[0029] It's important to note that Gabor unrolling is a technique for decomposing a signal in time and frequency. It generates a time-frequency representation of the signal by convolving it with a set of Gabor basis functions. A Gabor basis function is the product of a window function and a sine wave, used to extract features of the signal within a specific time and frequency range. Gabor unrolling effectively captures local features of a signal and is suitable for analyzing non-stationary signals and detecting frequency changes; therefore, it is widely used in signal processing, image processing, and feature extraction.

[0030] Specifically, the Gabor coefficients of the vibration signal are:

[0031] in, C m,n ( k )express k At the moment m The time-domain sampling location and the first n Gabor coefficients of vibration signals at each frequency domain sampling location x ( k ) indicates in k Vibration signal at a given time, express k Time in the cycle L The conjugate function of the Gabor analytic function under the given conditions. γ ( k ) indicates in k Gabor analysis function at time t, express k Time in the cycle L The conjugate function of the Gabor basis functions.

[0032] The Gabor basis functions are as follows:

[0033] in, h Represents Gabor basis functions. ΔM represents the time domain sampling interval, ΔN This indicates the frequency domain sampling interval.

[0034] The Gabor analysis function is as follows:

[0035] in, γ L ( k )express k Time in the cycle L The Gabor analysis function under, γ This represents the Gabor analysis function.

[0036] Specifically, the bioorthogonality between Gabor basis functions and Gabor analysis functions is as follows:

[0037] in, δ ( m )( n ) represents the Kronecker delta function.

[0038] It should be noted that, in the case of critical sampling, .

[0039] In this invention, the Gabor expansion algorithm is a tool that combines time-domain and frequency-domain analysis, providing both time-domain and frequency-domain information of a signal simultaneously. Using the Gabor expansion algorithm, the signal can be analyzed at different time points to obtain a detailed time-frequency diagram. This is particularly effective for processing non-stationary signals (such as rolling bearing vibration signals) because the frequency components of non-stationary signals change over time. By appropriately selecting the time-domain sampling interval (…), ΔM ) and frequency domain sampling interval ( ΔN This allows for high resolution in time-frequency plots, providing detailed characterization of signal spectral changes. This is particularly important for detecting subtle frequency variations and fault signal characteristics in vibration signals. The basis functions and analysis functions in the Gabor expansion algorithm can be adjusted according to actual needs to adapt to different types of vibration signals. This flexibility allows for effective fault detection and analysis of various complex mechanical systems.

[0040] S3: Using the C-ICEF algorithm, extract the target order component from the Gabor time-frequency diagram, and perform polynomial fitting on the target order component to obtain the filter center frequency line.

[0041] In this invention, the C-ICEF algorithm can accurately extract the target order component from complex Gabor time-frequency plots. For characteristic orders in rolling bearing vibration signals, this precise extraction capability helps isolate and analyze signal components related to specific fault modes, thereby improving the accuracy of fault diagnosis. By identifying and extracting the target order component from the time-frequency plot, the C-ICEF algorithm can effectively filter out signal components and noise irrelevant to the target. This process improves signal purity and ensures the reliability of subsequent analysis.

[0042] It should be noted that the C-ICEF algorithm is a combination of the Canny edge detection algorithm and the ICEF algorithm.

[0043] The Canny edge detection algorithm is a classic image processing technique used to detect edges in images. Its steps include: first, smoothing the image using a Gaussian filter to reduce noise; then calculating the gradient intensity and direction of each pixel in the image; next, performing non-maximum suppression to preserve local maxima and refine the edges; and finally, applying double threshold detection to classify edges into strong and weak edges, connecting them to obtain a complete edge image. This algorithm is widely used due to its excellent edge detection performance and low false alarm rate.

[0044] Among them, the ICEF algorithm (Iterative Clustering and Estimation of Frequency) is a frequency extraction algorithm used in signal processing. This algorithm iteratively clusters and estimates the frequency of features in a signal to extract the target order components. It combines frequency domain analysis with adaptive filtering techniques, and through an iterative optimization process, it can more accurately identify and extract periodic features in signals while maintaining high robustness in noisy environments.

[0045] In one possible implementation, S3 specifically includes sub-steps S301 to S304: S301: Use the averaging method to convert the Gabor time-frequency graph into a grayscale image:

[0046] in, Gray This represents the pixel values ​​after grayscale conversion of the Gabor time-frequency map. R This represents the pixel value of the red channel in the Gabor time-frequency plot. G This represents the pixel value of the green channel in the Gabor time-frequency plot. B This represents the pixel value of the blue channel in the Gabor time-frequency plot.

[0047] One common image processing method is the averaging method, used to convert color images to grayscale images. This method calculates the weighted average of the red, green, and blue channel values ​​for each pixel in the color image to obtain the grayscale value. This weighted average reflects the differences in human eye sensitivity to different colors, thus generating a grayscale image that better matches visual perception.

[0048] In this invention, Gabor time-frequency graphs are typically presented as color images, containing information from three channels: red, green, and blue (RGB). Converting them to grayscale retains only information from a single channel, significantly simplifying subsequent calculations and processing. This simplification makes the algorithm more efficient in image processing, reducing computational resource consumption.

[0049] S302: Extract the curved edges in the grayscale image using the Canny edge detection algorithm to obtain the edge image.

[0050] In one possible implementation, S302 specifically includes sub-steps S3021 to S3025: S3021: Use a Gaussian filter to filter the grayscale image.

[0051] in, G ( x, y ) represents the Gaussian function in coordinates ( x , y The pixel value at () x , y ) represents the coordinates of a pixel. x Represents the x-coordinate of a pixel. y Represents the ordinate of a pixel. σ This represents the standard deviation of the Gaussian filter.

[0052] S3022: Using the Sobel operator, calculate the gradient intensity and gradient direction of each pixel in the filtered grayscale image:

[0053] in, M ( x , y ) indicates that at coordinates ( x , y The gradient intensity of the pixel at position ) G x This represents the gradient value in the horizontal direction. G y This represents the gradient value in the vertical direction. θ ( x , y ) indicates that at coordinates ( x , y The gradient direction of the pixel at position ) This represents the convolution operator.

[0054] The Sobel operator is a commonly used edge detection operator used in image processing to calculate the gradient intensity and direction of pixels. It uses two 3x3 convolution kernels to perform convolution operations on the image in the horizontal and vertical directions respectively to detect image edges. Using the gradients in these two directions, the Sobel operator can calculate the gradient intensity and direction of each pixel, which can then be used for edge detection and feature extraction. Its advantages include good noise suppression and high accuracy in edge localization.

[0055] S3023: Perform non-maximum suppression on the calculated gradient intensity of the pixels. Check along the gradient direction of each pixel, retain the pixels with local maxima, and set the pixels that are not local maxima to 0.

[0056] Non-maximum suppression (NVS) is an image processing technique used for edge detection. Its main purpose is to refine edges, making the detected edges more accurate and clearer. Its basic principle is to examine the gradient intensity of each pixel and its neighborhood along the gradient direction. Only when the gradient intensity of a pixel is the largest in its neighborhood (i.e., a local maximum) is the pixel retained; otherwise, its gradient intensity is set to 0. This process effectively removes stray responses that are not edge-related, preserving true edge pixels and resulting in a finer, more accurate edge image.

[0057] It's important to note that a local maximum (LoM) refers to a pixel and its neighborhood where the gradient intensity of that pixel is greater than the gradient intensity of all surrounding pixels. Specifically, a LoM is a "highest point" within a specific region, where the gradient intensity reaches a local maximum relative to the surrounding pixels. During non-maximum suppression, the gradient direction of each pixel is checked to ensure that only pixels with the highest gradient intensity in that direction are preserved, while the gradient intensity of other pixels is set to 0.

[0058] S3024: Perform dual threshold detection on the gradient intensity of pixels after non-maximum suppression. Pixels with gradient intensity greater than or equal to the first threshold are considered strong edges, pixels with gradient intensity less than or equal to the second threshold are ignored, and pixels with gradient intensity between the first and second thresholds are considered weak edges.

[0059] It should be noted that those skilled in the art can set the size of the first threshold and the second threshold according to actual needs, and the present invention does not limit them.

[0060] Dual-threshold detection is a step in edge detection algorithms, primarily used to distinguish between strong edges, weak edges, and non-edges. In this process, the algorithm sets two thresholds: a higher threshold and a lower threshold. Pixels with gradient strength greater than or equal to the higher threshold are considered strong edges and are directly retained. Pixels with gradient strength less than the lower threshold are ignored and set to 0. Pixels with gradient strength between the two thresholds are considered weak edges. Subsequently, weak edge pixels are only retained if they are connected to strong edge pixels; otherwise, they are discarded. Through this method, dual-threshold detection can effectively reduce noise while preserving important edge information.

[0061] S3025: Link strong edges with weak edges to obtain an edge image.

[0062] Furthermore, to ensure the continuity and integrity of the edges, the set of all data points in the edge image is defined as U.

[0063] In this invention, smoothing the grayscale image using a Gaussian filter effectively suppresses noise interference. Noise affects the accuracy of edge detection, while the Gaussian filter smooths the image, reducing the impact of noise and thus obtaining more accurate edge information in subsequent gradient calculations. The Sobel operator highlights salient features in the image when calculating gradient strength and direction. By calculating the gradient of each pixel in the image, edges in the image, i.e., significant curves or contours, can be accurately identified. These features are crucial for extracting target order components. Non-maximum suppression is one of the key steps in the Canny algorithm. It ensures that only local maxima points in the gradient direction are retained, filtering out unimportant edge information, thereby improving the accuracy of edge detection. This processing method ensures that the edge detection results are more accurate and clear. Through the steps of double threshold detection and edge linking, the Canny algorithm can not only distinguish between strong and weak edges but also connect weak edges to strong edges, thereby generating a complete edge image. This edge linking process ensures that the edges in the image have good coherence and integrity, making subsequent feature extraction more stable and reliable.

[0064] S303: Cluster all data points in the edge image using the DBSCAN algorithm to extract target order components.

[0065] Specifically, after edge detection by the Canny edge detection algorithm, multiple clusters of data points will appear in the edge image. In order to filter the most significant data points, the DBSCAN algorithm is used to cluster all data points in the edge image and identify the main set of data points (target order components).

[0066] DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density-based clustering algorithm capable of discovering clusters of arbitrary shapes while identifying noisy points. The algorithm determines the density of data points by defining a radius parameter `eps` and a minimum number of points parameter `minPts`. Specifically, DBSCAN starts with an unvisited data point. If the number of points within its neighborhood of the `eps` radius is not less than `minPts`, then that point is considered a core point, forming a cluster. This cluster is then expanded to include all density-connected data points until it cannot be expanded further. Data points that do not belong to any cluster are marked as noise points. The advantage of DBSCAN is that it does not require pre-specifying the number of clusters and can handle noisy datasets.

[0067] In a possible implementation, S303 specifically includes sub-steps S3031 to S3036: S3031: Initialize all data points in the edge image and mark each data point as unvisited.

[0068] S3032: Randomly select data points from unvisited data points p Mark as visited and inspect data points. p of eps The number of data points in the neighborhood, of which, eps Representing data points p The radius of the neighborhood.

[0069] Optionally, eps The Euclidean distance is set to 0.5.

[0070] S3033: At data point p of eps The number of data points in the domain is greater than or equal to minPts In the case of data points p Create a new cluster as the core point and divide the data points p Add it to the newly created cluster, where, minPts express eps The threshold for the number of samples in the neighborhood.

[0071] Optionally, minPts Set to 5.

[0072] S3034: Traverse data points p Check each unvisited data point within the neighborhood. eps The number of data points in the neighborhood, and eps The number of data points in the neighborhood is greater than or equal to minPts The unvisited data points are classified as data points p As the core point of the newly created cluster.

[0073] S3035: Repeat sub-steps S3032 to S3034 until all data points in the edge image have been accessed, resulting in multiple clusters containing data points of the edge image, and marking data points not assigned to a cluster as noise points.

[0074] S3036: By counting the number of data points in each cluster, the cluster with the largest number of data points is used as the target order component.

[0075] In this invention, DBSCAN exhibits strong noise resistance. During clustering, DBSCAN marks data points that do not meet the clustering criteria (i.e., the number of data points in the neighborhood of eps is less than minPts) as noise points, thereby eliminating irrelevant or isolated noise points in the image. This characteristic makes the extracted target order components more accurate and reliable. The DBSCAN algorithm can identify clusters of arbitrary shapes, not just spherical clusters. Since the target order components in edge images may exhibit complex shapes, DBSCAN can flexibly handle these diverse data distributions, thus extracting the target order components more accurately.

[0076] S304: The target order components are polynomially fitted using the ICEF algorithm to obtain the filter center frequency line.

[0077] In one possible implementation, S304 specifically includes sub-steps S3041 to S3044: S3041: Initialize fitting parameters, set iteration radius θ Noise point set L 1 and the set of fitted points L 2. Set of fitting points L 2 represents the set of all data points in the target order component.

[0078] It should be noted that .

[0079] S3042: Determine the fitting order n ’ And using the minimum squares method, with the objective of minimizing the error between the fitted curve and the actual data points, the fitted point set is... L The data points in step 2 are fitted with a polynomial to obtain the initial curve:

[0080] in, Represents a polynomial function. This represents the order constant obtained after polynomial fitting. error This represents the fitting error. y i Indicates the i The true value of each data point Represents the polynomial function at the th i Predicted values ​​for each data point x i Indicates the i The independent variable values ​​for each data point m Represents the set of fitted points L The total number of data points in 2.

[0081] It should be noted that the order of the polynomial fitting... n ’ The order of fitting determines the complexity of the fitted curve. The higher the order of fitting, the greater the flexibility of fitting, but it may also lead to overfitting.

[0082] Specifically, the present invention selects Given data points The minimum squares method aims to minimize the error between the fitted curve and the actual data points, and applies this method to the set of fitted points. L The data points in step 2 are subjected to polynomial fitting to obtain the initial curve.

[0083] It's important to note that the Least Squares Method (LSM) is a mathematical optimization method for data fitting, widely used in regression analysis. Its core idea is to find the best-fitting parameters or function by minimizing the sum of squared errors between the observed data and the model's predicted values. Specifically, the LSM adjusts the model parameters to minimize the sum of squared differences (i.e., residuals) between the model's predicted values ​​and the actual observed values, thus obtaining an optimal fit. This method is very effective, especially in dealing with linear regression problems, as it can provide a definite analytical solution.

[0084] Furthermore, the initial curve is extended through sub-steps S3043 and S3044 to obtain the filter center frequency line.

[0085] S3043: Find the left and right endpoints of the initial curve, and traverse the set of noise points. L All data points in 1, including the set of noise points. L In step 1, data points within the iteration radius between the left and right endpoints are moved to the set of fitted points. L In step 2, the curve is refitted using the minimum squares method.

[0086] S3044: Repeat sub-step S3043 until the noise point set is reached. L In step 1, no data points were moved to the set of fitted points. L The fitting process ends in step 2, yielding the filter center frequency line.

[0087] It should be noted that when the noise point set L In step 1, no data points were moved to the set of fitted points. L In step 2, this indicates the set of noise points at that time. L There are no data points in step 1 that are located within the iteration radius between the left and right endpoints.

[0088] In this invention, polynomial fitting effectively fits data curves that conform to the target order components. By using the minimum squares method for fitting, the goal is to minimize the error between the fitted curve and the actual data points, ensuring that the fitted curve approximates the true target data as closely as possible, thus improving the fitting accuracy and precision. Throughout the fitting process, adaptive processing of complex data is achieved by continuously adjusting the fitted point set and the noise point set. Even with a large amount of noise or outliers in the data, the algorithm can flexibly adjust to ensure the final filtered center frequency line has high robustness. By gradually incorporating data points from the noise point set into the fitted point set, data utilization is maximized, avoiding the loss of valuable information due to directly discarding noise points. This process preserves the effective information in the data to the greatest extent, resulting in a more accurate final filtered center frequency line.

[0089] S4: Based on the filter center frequency line, calculate the Gabor coefficients of the target order component using the masking algorithm, and obtain the time spectrum of the target order component.

[0090] Masking algorithms are a common technique in signal and image processing used to selectively preserve or suppress data in specific regions. By constructing a binary mask array (mask), the masking algorithm retains regions with a mask value of 1, while regions with a mask value of 0 are ignored or suppressed. In signal processing, masking algorithms are often used to highlight specific frequencies or time-frequency components of the target signal, thereby helping to eliminate noise or interference and extract useful signal information. This method can significantly improve the accuracy and effectiveness of analysis when processing complex signals and images.

[0091] In one possible implementation, in step S4, the Gabor coefficients of the target order component are calculated using a masking algorithm based on the filter center frequency line, specifically as follows: By using a time-varying filtering neighborhood, a binary masking array with the same dimension as the Gabor coefficients of the vibration signal is set, and the Gabor coefficients of the target order component are calculated:

[0092] in, The Gabor coefficients representing the target order components. This represents a binary masking array.

[0093] Among them, the calculation methods for the time-varying filter neighborhood include: the equal frequency method and the equal order method.

[0094] When using the equal-frequency method, the time-varying filtering neighborhood is specifically as follows:

[0095] in, f q ( t ) indicates the first q The center frequency of the first step is the center frequency line of the filter. This indicates equal frequency bandwidth.

[0096] When using the equal-order method, the time-varying filtering neighborhood is specifically as follows:

[0097] in, Indicates equal bandwidth. f 1( t ) represents the first-order center frequency.

[0098] In this invention, the Gabor coefficients calculated using the masking algorithm can form a more concise and clear time spectrum. Since only the target order component is retained, the generated time spectrum removes interference from other irrelevant frequency components, significantly improving its readability and interpretability. This is highly beneficial for subsequent signal analysis and fault diagnosis, as engineers can more clearly see the frequency components related to the target characteristics, making it easier to identify fault modes. The use of a time-varying filtering neighborhood allows the filtering process to be dynamically adjusted according to the characteristics of the target order component. This adaptive filtering method can effectively handle the complexity of frequency variations in vibration signals. When using the equal-frequency method, a fixed frequency bandwidth Δ is set... f This allows for the extraction of signal components related to a specific frequency range from the time-frequency plot. When using the equal-order method, a bandwidth Δ related to the order is set. o This allows for a better capture of the relationships between different orders. This flexibility enables the algorithm to adapt to the needs of different signal characteristics, improving the accuracy of target order component extraction.

[0099] S5: Using a local extremum search algorithm, the instantaneous frequency of the target order component is estimated based on the time spectrum, and a second fitting is performed to obtain the instantaneous frequency function.

[0100] The local extremum search algorithm is as follows:

[0101] in, IF q ( t +1) indicates that in t The instantaneous frequency at time +1 t Indicates time, f Indicates frequency, Arg Represents the average operator. max Represents the maximum value operator.M q ( t , f ) indicates in t Time and f Amplitude at frequency.

[0102] It's important to note that the Local Extremum Search Algorithm (LOS) is a commonly used algorithm in signal processing and optimization problems. It's used to find local maxima or minima within a specific range. In signal processing, LOS is typically used to analyze the frequency or time characteristics of a signal. By searching for extreme points within a given window or interval, it determines the location of the maximum or minimum value of the signal within that interval. This algorithm is crucial for capturing the instantaneous characteristics of a signal and identifying important frequency components or time points, and it is particularly useful in dynamic signal analysis.

[0103] The quadratic fitting is specifically as follows:

[0104] in, IF q ( t ) indicates in t Instantaneous frequency at a given moment a , b and c Both represent constants obtained after fitting.

[0105] In this invention, the local extremum search algorithm estimates the instantaneous frequency by searching for local extrema of amplitude in the time-frequency spectrum. This method can find the point of maximum amplitude within the frequency range, ensuring accurate estimation of the instantaneous frequency. Because local extrema in the time-frequency spectrum typically represent the main frequency components of the signal, this search method can effectively capture instantaneous frequency changes. Quadratic fitting smooths the instantaneous frequency estimation result by performing a quadratic fit on the instantaneous frequency obtained from the local extremum search. Quadratic fitting can handle the nonlinear characteristics of frequency changes, making the instantaneous frequency function smoother and reducing fluctuations caused by noise or data discreteness. This helps to obtain a more stable and accurate frequency estimate, especially when the signal noise is high or the frequency changes rapidly.

[0106] S6: Calculate the key phase time scale for equal-angle resampling using the instantaneous frequency function, and perform equal-angle resampling based on the key phase time scale.

[0107] In one possible implementation, S6 involves calculating the bond phase timescale for equal-angle resampling and performing equal-angle resampling based on the bond phase timescale, specifically including: Calculate the bond phase timescale of equal-angle resampling using the following formula:

[0108] in, T n The key phase timescale represents equal-angle resampling. Q max Indicates the highest order of analysis. T 0 Indicates the start time of time domain sampling, N Indicates the total number of resampling times.

[0109] According to the following formula, the Lagrange linear interpolation algorithm is used to perform equal angle resampling based on the bond phase time scale:

[0110] in, x ( T n () represents the time-domain signal of the key phase timescale resampled at equal angles. x ( t i ) indicates in t i The time-domain signal at time t, x ( t i+1 ) indicates in t i+1 The time domain signal at the moment.

[0111] In the present invention, equiangular resampling resamples the signal based on instantaneous frequency changes, ensuring that each sampling point reflects the signal's actual frequency change. This method provides more accurate frequency information when analyzing non-uniform or variable-speed signals, reducing errors caused by frequency changes. Using the Lagrange linear interpolation algorithm for equiangular resampling smoothly interpolates the signal, eliminating noticeable artifacts or discontinuities during the resampling process. This method provides continuous, smooth resampling results, reducing errors introduced by interpolation.

[0112] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least: In this invention, the collected vibration signal is expanded using the Gabor expansion algorithm to obtain a Gabor time-frequency diagram. The target order component is extracted from the Gabor time-frequency diagram using the C-ICEF algorithm, and polynomial fitting is performed on the target order component to obtain the filtered center frequency line. This effectively solves the complex vibration interference problem between rolling bearings or between their own components. It can eliminate a large amount of interference noise in the collected vibration signal, avoiding the situation where the fault characteristic vibration signal is submerged by noise and difficult to separate when the fault is not obvious. By calculating the key phase time scale of equal-angle resampling and performing equal-angle resampling according to the key phase time scale, the problem of difficulty in vibration signal analysis caused by irregular variable speed operation of rolling bearings under varying operating conditions is solved.

[0113] Reference Manual Figure 2 The diagram shows a schematic of the structure of a frequency domain resampling system for rolling bearing vibration signals provided by the present invention.

[0114] The present invention also provides a frequency domain resampling system 30 for rolling bearing vibration signals, comprising: a memory 303 and one or more processors 301.

[0115] The memory 303 stores one or more application programs, which are adapted to be executed by the one or more processors 301 to implement the frequency domain resampling method for rolling bearing vibration signals as described in the method embodiment.

[0116] The rolling bearing vibration signal frequency domain resampling system 30 includes a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302.

[0117] The structure of the frequency domain resampling system 30 for the rolling bearing vibration signal does not constitute a limitation on the embodiments of the present invention.

[0118] Processor 301 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, a transistor logic device, a hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the present disclosure. Processor 301 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like.

[0119] Bus 302 may include a pathway for transmitting information between the aforementioned components. Bus 302 may be a PCI bus or an EISA bus, etc. Bus 302 may be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the figure, but this does not mean that there is only one bus or one type of bus.

[0120] The memory 303 may be a ROM or other type of static storage device capable of storing static information and instructions, RAM or other type of dynamic storage device capable of storing information and instructions, or it may be an EEPROM, CD-ROM or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0121] It should be noted that the rolling bearing vibration signal frequency domain resampling system 30 can realize the above-mentioned rolling bearing vibration signal frequency domain resampling method and achieve the same or similar technical effects. To avoid repetition, the present invention will not elaborate further.

[0122] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following: In this invention, the collected vibration signal is expanded using the Gabor expansion algorithm to obtain a Gabor time-frequency diagram. The target order component is extracted from the Gabor time-frequency diagram using the C-ICEF algorithm, and polynomial fitting is performed on the target order component to obtain the filtered center frequency line. This effectively solves the complex vibration interference problem between rolling bearings or between their own components. It can eliminate a large amount of interference noise in the collected vibration signal, avoiding the situation where the fault characteristic vibration signal is submerged by noise and difficult to separate when the fault is not obvious. By calculating the key phase time scale of equal-angle resampling and performing equal-angle resampling according to the key phase time scale, the problem of difficulty in vibration signal analysis caused by irregular variable speed operation of rolling bearings under varying operating conditions is solved.

[0123] The present invention also provides a computer-readable storage medium having a computer program stored thereon, the computer program being loaded and executed by a processor to perform the frequency domain resampling method for rolling bearing vibration signals as described in the first aspect.

[0124] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0125] The following points need to be explained: (1) The accompanying drawings of the embodiments of the present invention only involve the structures involved in the embodiments of the present invention. Other structures can refer to the general design.

[0126] (2) For clarity, the thickness of layers or regions is enlarged or reduced in the drawings used to describe embodiments of the invention, i.e., these drawings are not drawn to scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being “above” or “below” another element, the element may be “directly” located “above” or “below” the other element or there may be intermediate elements.

[0127] (3) Where there is no conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

[0128] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A frequency domain resampling method for rolling bearing vibration signals, characterized in that: include: S1: Collect vibration signals of rolling bearings; S2: Expand the vibration signal by using a Gabor expansion algorithm to obtain a Gabor time-frequency graph; S3: extracting target order components from the Gabor time-frequency diagram by using a C-ICEF algorithm, and performing polynomial fitting on the target order components to obtain a filtering center frequency line; S4: Calculate the Gabor coefficient of the target order component according to the filtering center frequency line through a masking algorithm to obtain the time-frequency spectrum of the target order component; S5: using a local extreme value search algorithm, based on the time-frequency spectrum, the instantaneous frequency of the target order component is estimated, and a quadratic fitting is performed to obtain an instantaneous frequency function; S6: Calculate the key phase time scale of equal angle resampling through the instantaneous frequency function, and perform equal angle resampling according to the key phase time scale.

2. The rolling bearing vibration signal frequency domain resampling method according to claim 1, characterized in that: The Gabor expansion algorithm is specifically: in, x L ( k )express k Moments in the cycle L The vibration signal after expansion is shown below. L represents the period of the vibration signal, M represents the number of time domain samples, N represents the number of frequency domain samples, C m,n Indicated in m The time domain sampling position and n Gabor coefficients of the vibration signal at the frequency domain sampling position, h L ( k )express k Moments in the cycle L Gabor basis function under ; Among them, the Gabor coefficient of the vibration signal is specifically: in, C m,n ( k )express k At the moment m The time domain sampling position and n Gabor coefficients of the vibration signal at the frequency domain sampling position, x ( k ) means in k The vibration signal at the moment, express k Moments in the cycle L The conjugate function of the Gabor analysis function under , γ ( k ) means in k Gabor analysis function at the moment, express k Moments in the cycle L The conjugate function of the Gabor basis function under ; Wherein, the Gabor basis function is specifically: in, h represents the Gabor basis function, ΔM represents the time domain sampling interval, ΔN represents the frequency domain sampling interval; Wherein, the Gabor analysis function is specifically: in, γ L ( k )express k Moments in the cycle L Gabor analysis function under γ represents Gabor analysis function; The biorthogonal relationship between the Gabor basis function and the Gabor analysis function is specifically: in, δ ( m )( n ) represents the Kronecker delta function.

3. The rolling bearing vibration signal frequency domain resampling method according to claim 1, characterized in that: The S3 specifically includes: S301: Using the average method, convert the Gabor time-frequency image into a grayscale image: in, Gray Represents the pixel value after the Gabor time-frequency map is grayed out. R Represents the pixel value of the red channel of the Gabor time-frequency map, G represents the pixel value of the green channel of the Gabor frequency spectrum map, B Represents the pixel value of the blue channel of the Gabor time-frequency map; S302: extracting the curve edges in the grayscale image by using the Canny edge detection algorithm to obtain an edge image; S303: clustering all data points in the edge image using the DBSCAN algorithm to extract target order components; S304: Perform polynomial fitting on the target order component by ICEF algorithm to obtain a filtering center frequency line.

4. The rolling bearing vibration signal frequency domain resampling method according to claim 3, characterized in that: The S302 specifically includes: S3021: Use a Gaussian filter to filter the grayscale image: in, G ( x , y ) represents the Gaussian function at coordinates ( x , y ), ( x , y ) represents the coordinates of the pixel point, x Represents the horizontal coordinate of the pixel point, y Represents the vertical coordinate of the pixel point. σ represents the standard deviation of the Gaussian filter; S3022: Use the Sobel operator to calculate the gradient intensity and gradient direction of each pixel in the grayscale image after filtering: in, M ( x , y ) means that at the coordinate ( x , y ), G x Represents the gradient value in the horizontal direction. G y Represents the gradient value in the vertical direction. θ ( x , y ) means that at the coordinate ( x , y ), represents the convolution operator; S3023: performing non-maximum suppression on the calculated gradient strength of the pixel points, checking along the gradient direction of each pixel point, retaining the pixel points with the local maximum gradient strength, and setting the pixel points that are not the local maximum gradient strength to 0; S3024: performing double threshold detection on the gradient strength of the pixel points after non-maximum suppression, taking the pixel points whose gradient strength is greater than or equal to the first threshold as strong edges, ignoring the pixel points whose gradient strength is less than or equal to the second threshold, and taking the pixel points whose gradient strength is between the first threshold and the second threshold as weak edges; S3025: edge-linking the strong edge and the weak edge to obtain an edge image.

5. The rolling bearing vibration signal frequency domain resampling method according to claim 3, characterized in that: The S303 specifically includes: S3031: Initialize all data points in the edge image and mark each data point as unvisited; S3032: Randomly select data points from unvisited data points p Mark as visited and check the data point p of eps The number of data points in the neighborhood, where eps Represents data points p The radius of the neighborhood; S3033: At the data point p of eps The number of data points in the domain is greater than or equal to minPts In the case of the data point p Create a new cluster as the core point and add the data point p Join the newly created cluster, where the minPts express eps The sample number threshold of the neighborhood; S3034: Traverse the data points p All unvisited data points in the neighborhood, check each unvisited data point eps The number of data points in the neighborhood, and eps The number of data points in the neighborhood is greater than or equal to minPts The unvisited data points are classified as the data points p In the newly created cluster as the core point; S3035: repeating sub-steps S3032 to S3034 until all data points in the edge image are visited, obtaining a plurality of clusters containing data points of the edge image, and marking data points that are not included in the clusters as noise points; S3036: By counting the number of data points in each cluster, the cluster with the largest number of data points is used as the target order component.

6. The rolling bearing vibration signal frequency domain resampling method according to claim 3, characterized in that: The S304 specifically includes: S3041: Initialize fitting parameters and set iteration radius θ , noise point set L 1 and the set of fitting points L 2. The fitting point set L 2 is the set of all data points in the target order component; S3042: Determine the fitting order n ’ , and use the minimum square method to minimize the error between the fitting curve and the actual data points. L The data points in 2 are fitted with a polynomial to obtain the initial curve: in, represents a polynomial function, represents the order constant obtained after polynomial fitting, error represents the fitting error, y i Indicates i The true value of the data point, Represents the polynomial function in i The predicted value of the data point, x i Indicates i The independent variable value for each data point, m Represents a set of fitting points L The total number of data points in 2; S3043: Find the left endpoint and the right endpoint of the initial curve, and traverse the noise point set L All data points in 1 are the noise points set L The data points within the iteration radius between the left and right endpoints in 1 are moved to the fitting point set L 2, and refit the curve using the minimum square method; S3044: Repeat sub-step S3043 until the noise point set L 1 No data points are moved to the set of fitting points L The fitting is completed in 2 to obtain the filtering center frequency line.

7. The rolling bearing vibration signal frequency domain resampling method according to claim 1, characterized in that: The Gabor coefficient of the target order component is calculated according to the filtering center frequency line by using a masking algorithm in S4, specifically: Through the time-varying filtering neighborhood, a binary masking array with the same dimension as the Gabor coefficient of the vibration signal is set to calculate the Gabor coefficient of the target order component: in, represents the Gabor coefficient of the target order component, Represents a binary masked array; Wherein, the calculation method of the time-varying filtering neighborhood includes: equal frequency method and equal class method; When the equal frequency method is used, the time-varying filtering neighborhood is specifically: in, f q ( t ) indicates the q The order center frequency is the filter center frequency line. Indicates equal frequency bandwidth; When the equal-order method is used, the time-varying filtering neighborhood is specifically: in, represents equal class bandwidth, f 1( t ) represents the first-order center frequency.

8. The rolling bearing vibration signal frequency domain resampling method according to claim 1, characterized in that: The local extremum search algorithm is specifically: in, IF q ( t +1) indicates that t The instantaneous frequency at time +1, t Indicates time, f Indicates frequency, Arg represents the average operator, max represents the maximum operator, M q ( t , f ) means in t Moment and f Amplitude at frequency; The quadratic fitting is specifically: in, IF q ( t ) means in t The instantaneous frequency at the moment, a , b and c They all represent the constants obtained after fitting.

9. The rolling bearing vibration signal frequency domain resampling method according to claim 1, characterized in that: The step of calculating the key phase time scale of equal angle resampling in S6, and performing equal angle resampling according to the key phase time scale, specifically includes: The key phase time scale of equal angle resampling is calculated according to the following formula: in, T n represents the key phase time scale of equal angle resampling, Q max represents the maximum analysis order, T 0 Indicates the start time of time domain sampling, N Indicates the total number of resampling times; According to the following formula, equal angle resampling is performed according to the bond phase time scale by Lagrange linear interpolation algorithm: in, x ( T n ) represents the time domain signal at the key phase time scale with equal angle resampling, x ( t i ) means in t i The time domain signal at time, x ( t i+1 ) means in t i+1 The time domain signal at the moment.

10. A rolling bearing vibration signal frequency domain resampling system, characterized in that: include: memory and one or more processors; One or more application programs are stored in the memory, and the one or more application programs are suitable for being executed by the one or more processors to implement the rolling bearing vibration signal frequency domain resampling method according to any one of claims 1 to 9.

Images