A method for rapid extraction of bearing fault characteristics

Abstract

The present invention belongs to the technical field of bearing fault detection, and particularly relates to a method for rapidly extracting bearing fault characteristics. In this invention, by using autoregressive model decomposition and combining with the RKSR optimization index, the optimal model order can be adaptively selected to ensure the effective separation of the steady-state signal and the residual signal, so that the fault characteristics are easier to detect, reducing the risks of false alarms and missed alarms. Through the self-nested iterative residual strategy, the steady-state part is peeled off multiple times, enabling the gradual accumulation and manifestation of the weak fault characteristics initially hidden in the noise, thus realizing the timely diagnosis of early faults. The improved spectral amplitude modulation algorithm effectively suppresses random pulses and other interference signals, enabling the entire fault extraction process to still operate stably under complex working conditions and high background noise conditions. Combining with the fast spectral kurtosis processing method, key spectral characteristics can be extracted from the conditioned signal in a short time, meeting the requirements of real-time monitoring and fault warning.

Description

Technical Field

[0001] The present invention belongs to the technical field of bearing fault detection, and particularly relates to a method for quickly extracting bearing fault characteristics. Background Art

[0002] With the continuous improvement of industrial automation and equipment intelligence levels, rolling bearings, as key components in mechanical equipment, their operating states are directly related to the safety and reliability of the equipment, and their timely diagnosis is crucial for ensuring the safe operation of the equipment. Rolling bearings are an important part of industrial equipment such as trains. However, under complex operating conditions, especially in the presence of high background noise, traditional signal processing techniques often have difficulty effectively extracting early weak fault characteristics.

[0003] In traditional technologies, although the collected bearing vibration signals are preliminarily processed through preprocessing means such as low-pass filtering and normalization, due to the often superposition of noise components and fault signals in the signals, it is difficult to clearly distinguish the fault characteristics during subsequent analysis, affecting the accuracy and real-time performance of diagnosis, and the steady-state part is not gradually peeled off in a self-nested iterative manner, resulting in insufficiently prominent fault characteristics in the residual signals and limiting the sensitive detection of early faults. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for quickly extracting bearing fault characteristics, which can effectively amplify early fault signals and reduce the risks of false alarms and missed alarms under complex working conditions and high background noise.

[0005] The technical solutions adopted by the present invention are specifically as follows:

[0006] A method for quickly extracting bearing fault characteristics, comprising:

[0007] Obtain a bearing vibration signal, and perform preprocessing on the bearing vibration signal to obtain a preprocessed signal;

[0008] Obtain an autoregressive model, and obtain a steady-state signal and a residual signal according to the autoregressive model and the preprocessed signal;

[0009] Obtain the root mean square of the steady-state signal and the kurtosis of the residual signal, obtain an optimized evaluation index according to the root mean square of the steady-state signal and the kurtosis of the residual signal, traverse and calculate the optimized evaluation indexes at each order within a preset autoregressive order range, obtain a target order according to multiple optimized evaluation indexes, and obtain preliminary fault characteristics according to the target order, wherein the preliminary fault characteristics include a preliminary steady-state signal and a preliminary residual signal;

[0010] Obtain the preset number of iterations, and return the preliminary steady-state signal as the preprocessed signal to the step of obtaining the steady-state signal and the residual signal according to the autoregressive model and the preprocessed signal. Until the preset number of iterations is reached, obtain the cumulative residual signal according to the preset number of iterations;

[0011] Perform spectral amplitude modulation processing on the cumulative residual signal to obtain the conditioned fault signal;

[0012] Perform fast spectral kurtosis processing on the conditioned fault signal to obtain the bearing fault characteristics.

[0013] In a preferred scheme, the steps of obtaining the bearing vibration signal and preprocessing the bearing vibration signal to obtain the preprocessed signal include:

[0014] Obtain the bearing vibration signal;

[0015] Filter the bearing vibration signal according to the low-pass filter to obtain the vibration filtered signal;

[0016] Normalize the vibration filtered signal to obtain the preprocessed signal.

[0017] In a preferred scheme, the steps of obtaining the autoregressive model and obtaining the steady-state signal and the residual signal according to the autoregressive model and the preprocessed signal include:

[0018] Obtain the autoregressive model;

[0019] Obtain the bearing sequence vibration signals at multiple moments according to the bearing vibration signal;

[0020] Obtain the autoregressive coefficients corresponding to each bearing sequence vibration signal;

[0021] Obtain the steady-state signal according to the autoregressive model, each bearing sequence vibration signal at multiple moments, and the autoregressive coefficients corresponding to each bearing sequence vibration signal;

[0022] Obtain the residual signal according to the autoregressive model and the steady-state signal.

[0023] In a preferred scheme, the steps of obtaining the root mean square of the steady-state signal and the kurtosis of the residual signal, obtaining the optimization evaluation index according to the root mean square of the steady-state signal and the kurtosis of the residual signal, traversing and calculating the optimization evaluation index at each order within the preset autoregressive order range, and obtaining the target order according to multiple optimization evaluation indexes and obtaining the preliminary fault characteristics according to the target order include:

[0024] Obtain the preset autoregressive order range;

[0025] Obtain the root mean square of the steady-state signal and the kurtosis of the residual signal for each autoregressive order within the preset autoregressive order range;

[0026] Obtain the optimized evaluation index for each autoregressive order based on the root mean square of the steady-state signal and the kurtosis of the residual signal for each autoregressive order;

[0027] Arrange the optimized evaluation indexes of multiple autoregressive orders in descending order to obtain an order ranking table;

[0028] Select the first optimized evaluation index according to the order ranking table, and obtain the corresponding autoregressive order according to the first optimized evaluation index, and mark it as the target order;

[0029] Obtain the preliminary steady-state signal and the preliminary residual signal according to the target order and the autoregressive model.

[0030] In a preferred scheme, obtain the preset number of iterations, return the preliminary steady-state signal as the preprocessed signal to the step of obtaining the steady-state signal and the residual signal according to the autoregressive model and the preprocessed signal until the preset number of iterations is reached. The step of obtaining the cumulative residual signal according to the preset number of iterations includes:

[0031] Obtain the preset number of iterations;

[0032] Return the preliminary steady-state signal as the preprocessed signal to the step of obtaining the steady-state signal and the residual signal according to the autoregressive model and the preprocessed signal, and obtain the number of returns;

[0033] Obtain the iteration period when the number of returns meets the preset number of iterations;

[0034] Obtain multiple residual signals obtained each time after returning to the step of obtaining the steady-state signal and the residual signal according to the autoregressive model and the preprocessed signal during the iteration period;

[0035] Summarize the multiple residual signals to obtain the cumulative residual signal.

[0036] In a preferred scheme, the step of obtaining the iteration period when the number of returns meets the preset number of iterations includes:

[0037] Obtain the time node when the preliminary steady-state signal is first returned as the preprocessed signal to the step of obtaining the steady-state signal and the residual signal according to the autoregressive model and the preprocessed signal, and mark it as the start time of the iteration period;

[0038] Obtain the time node when the step of returning the preliminary steady-state signal as the preprocessed signal to the step of obtaining the steady-state signal and the residual signal according to the autoregressive model and the preprocessed signal is completed when the number of returns is equal to the preset number of iterations, and mark it as the end time of the iteration period;

[0039] Obtain the iteration period according to the start time and the end time of the iteration period.

[0040] In a preferred embodiment, the steps of performing spectral amplitude modulation processing on the accumulated residual signal to obtain the conditioned fault signal include:

[0041] Perform Fourier transform on the accumulated residual signal to obtain the amplitude spectrum and phase spectrum of the accumulated residual signal;

[0042] Perform exponential operation on the amplitude spectrum to obtain the corrected amplitude spectrum;

[0043] Obtain the corrected signal frequency spectrum according to the corrected amplitude spectrum and the phase spectrum;

[0044] Perform inverse Fourier transform on the corrected signal frequency spectrum to obtain the time-domain corrected signal;

[0045] Obtain the exponential range of the exponential operation process, and combine the time-domain corrected signal to obtain multiple corrected optimization evaluation indexes;

[0046] Arrange the multiple corrected optimization evaluation indexes in descending order to obtain the corrected index sorting table;

[0047] Select the first corrected optimization evaluation index according to the corrected index sorting table, and obtain the corresponding index according to the first corrected optimization evaluation index, and mark it as the target index;

[0048] Obtain the conditioned fault signal according to the target index and the autoregressive model.

[0049] In a preferred embodiment, the exponential range is from 0.8 to 1.0.

[0050] In a preferred embodiment, the steps of performing fast spectral kurtosis processing on the conditioned fault signal to obtain the bearing fault characteristics include:

[0051] Perform frame division processing on the conditioned fault signal to obtain multiple frequency band signals;

[0052] Obtain the signal kurtosis of each frequency band signal and generate a fast spectral kurtosis diagram;

[0053] Obtain the highest frequency band in the fast spectral kurtosis diagram as the target frequency band;

[0054] Obtain the filtered and conditioned signal according to the target frequency band and the conditioned fault signal;

[0055] Perform Hilbert transform on the filtered and conditioned signal to obtain the envelope signal;

[0056] Perform fast Fourier transform on the envelope signal to obtain the envelope spectrum;

[0057] Obtain the fault characteristic frequency in the envelope spectrum, and obtain the bearing fault type according to the fault characteristic frequency.

[0058] And a terminal for quickly extracting bearing fault characteristics, comprising:

[0059] One or more processors;

[0060] A storage device having one or more programs stored thereon;

[0061] When the one or more programs are executed by the one or more processors, the one or more processors implement a method for quickly extracting bearing fault characteristics.

[0062] The technical effects achieved by the present invention are as follows:

[0063] In the present invention, by using the autoregressive model decomposition and combining with the RKSR optimization index, the optimal model order can be adaptively selected to ensure the effective separation of the steady-state signal and the residual signal, so that the fault characteristics are more easily detected, reducing the risk of false alarms and missed detections. Through the self-nested iterative residual strategy, the steady-state part is peeled off multiple times, enabling the weak fault characteristics initially hidden in the noise to gradually accumulate and appear, thus realizing the timely diagnosis of early faults. The improved spectral amplitude modulation algorithm effectively suppresses random pulses and other interference signals, enabling the entire fault extraction process to still operate stably under complex working conditions and high background noise conditions, with high robustness. Combining with the fast spectral kurtosis processing method, key spectral characteristics can be extracted from the conditioned signal in a short time, meeting the requirements of real-time monitoring and fault warning, and improving the efficiency of equipment maintenance and fault response. Description of the Drawings

[0064] Figure 1 is the flowchart of the method provided by the present invention. Detailed Embodiments

[0065] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following detailed description of the specific embodiments of the present invention will be given in conjunction with the accompanying drawings of the specification.

[0066] Many specific details are set forth in the following description in order to fully understand the present invention, but the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0067] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The "in a preferred embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that mutually excludes other embodiments.

[0068] Next, the present invention will be described in detail with reference to the schematic diagrams. When describing the embodiments of the present invention in detail, for the sake of convenience of explanation, the schematic diagrams are only examples and should not limit the scope of protection of the present invention herein.

[0069] Please refer to the attached Figure 1 As shown, a method for quickly extracting bearing fault characteristics is provided, including:

[0070] S1. Obtain the bearing vibration signal and preprocess the bearing vibration signal to obtain a preprocessed signal;

[0071] S2. Obtain an autoregressive model, and obtain a steady-state signal and a residual signal according to the autoregressive model and the preprocessed signal;

[0072] S3. Obtain the root mean square of the steady-state signal and the kurtosis of the residual signal, obtain an optimized evaluation index according to the root mean square of the steady-state signal and the kurtosis of the residual signal, traverse and calculate the optimized evaluation index at each order within the preset autoregressive order range, and obtain the target order according to multiple optimized evaluation indexes, and obtain preliminary fault characteristics according to the target order, where the preliminary fault characteristics include a preliminary steady-state signal and a preliminary residual signal;

[0073] S4. Obtain the preset number of iterations, return the preliminary steady-state signal as the preprocessed signal to the step of obtaining the steady-state signal and the residual signal according to the autoregressive model and the preprocessed signal until the preset number of iterations is reached, and obtain an accumulated residual signal according to the preset number of iterations;

[0074] S5. Perform spectral amplitude modulation processing on the accumulated residual signal to obtain a conditioned fault signal;

[0075] S6. Perform fast spectral kurtosis processing on the conditioned fault signal to obtain bearing fault characteristics.

[0076] In the above steps S1 to S6, the collected bearing vibration signals are first preprocessed to remove high-frequency noise and irrelevant interference, ensuring that subsequent processing is based on a cleaner signal. The autoregressive model is used to decompose the preprocessed signal into a steady-state signal and a residual signal. The steady-state signal reflects the normal operating state of the device, while the residual signal contains latent fault information. This decomposition principle relies on the predictive ability of the autoregressive model for time-series signals, which can separate the predictable part from the unpredictable (abnormal or faulty) part. By calculating the kurtosis of the residual signal and the root mean square (RMS) of the steady-state signal, and using the ratio of the two as the optimization evaluation index (RKSR), each candidate order is traversed within the preset autoregressive order range, and the target order that maximizes the evaluation index is selected, thereby obtaining a preliminary steady-state signal and a preliminary residual signal. The steady-state signal obtained from the preliminary decomposition is fed back as the new input, and the AR model decomposition and optimization evaluation are repeatedly executed until the preset number of iterations is reached. In each iteration, the proportion of the fault characteristics in the residual signal accumulates and strengthens continuously, and finally an accumulated residual signal is formed. The improved spectral amplitude modulation algorithm (ISAM) is used for the accumulated residual signal. By performing exponential correction and reconstruction on the amplitude spectrum, the interference of random pulses and other irrelevant signals is effectively suppressed. The fast spectral kurtosis processing method is used to perform spectral processing on the conditioned fault signal, and the center frequency and related characteristic frequency bands of the fault signal are extracted therefrom to accurately locate the fault information and achieve the rapid extraction of bearing fault characteristics. Using the autoregressive model decomposition and combining with the RKSR optimization index, the best model order can be adaptively selected to ensure the effective separation of the steady-state signal and the residual signal, making the fault characteristics easier to detect and reducing the risks of false alarms and missed alarms. Through the self-nested iterative residual strategy, the steady-state part is peeled off multiple times, enabling the weak fault characteristics initially hidden in the noise to gradually accumulate and appear, thereby realizing the timely diagnosis of early faults. The improved spectral amplitude modulation algorithm effectively suppresses random pulses and other interference signals, enabling the entire fault extraction process to still operate stably under complex working conditions and high background noise conditions, with high robustness. Combining with the fast spectral kurtosis processing method, the key spectral characteristics can be extracted from the conditioned signal in a short time, meeting the requirements of real-time monitoring and fault warning and improving the efficiency of equipment maintenance and fault response.

[0077] In a preferred embodiment, the steps of obtaining a bearing vibration signal and preprocessing the bearing vibration signal to obtain a preprocessed signal include:

[0078] S101. Obtain a bearing vibration signal;

[0079] S102. Filter the bearing vibration signal according to a low-pass filter to obtain a vibration filtered signal;

[0080] S103. Normalize the vibration filtered signal to obtain a preprocessed signal.

[0081] In the above steps S101 to S106, the vibration signal of the bearing during operation is collected in real time by a sensor, the mechanical motion state is recorded in the form of an electrical signal, and the collected original vibration signal is filtered by a low-pass filter. The main purpose is to remove high-frequency noise and other interference components above the normal operating frequency range of the bearing. By setting an appropriate cut-off frequency, the low-pass filter retains the low-frequency components that reflect the main vibration characteristics of the bearing, thereby improving the signal-to-noise ratio of the signal. The vibration signal after filtering is normalized to unify the signal amplitude range. The low-pass filtering effectively reduces the high-frequency noise interference, making the signal on which the subsequent processing is based more pure, ensuring that the true information in the vibration signal is not submerged by the noise, thereby improving the accuracy of fault feature extraction. The normalization process makes the vibration signals collected at different times and under different conditions on the same numerical scale, reduces the influence of data fluctuations on the results, and improves the consistency and stability of data processing.

[0082] In a preferred embodiment, the steps of obtaining an autoregressive model and obtaining a steady-state signal and a residual signal according to the autoregressive model and the preprocessed signal include:

[0083] S201. Obtain an autoregressive model;

[0084] S202. Obtain the bearing sequence vibration signals at multiple moments according to the bearing vibration signal;

[0085] S203. Obtain the autoregressive coefficients corresponding to each bearing sequence vibration signal;

[0086] S204. Obtain a steady-state signal according to the autoregressive model, each bearing sequence vibration signal at multiple moments, and the autoregressive coefficients corresponding to each bearing sequence vibration signal;

[0087] S205. Obtain a residual signal according to the autoregressive model and the steady-state signal.

[0088] In the above steps S201 to S205, by constructing an autoregressive (AR) model, the expression of the autoregressive model is , using historical signal data to predict the vibration signal at the current moment, forming a mathematical description of the signal. For the preprocessed signal, the corresponding signal sequences are extracted at different sampling moments so as to apply the AR model in the context of the time series. Using methods such as the Yule-Walker equation or the least squares method, each signal sequence is fitted, and the corresponding autoregressive coefficients are solved. These coefficients characterize the influence weight of the historical data on the current signal and are important parameters for constructing the prediction signal later. According to the constructed AR model and the solved autoregressive coefficients, the bearing vibration signal at each moment is predicted to obtain a predicted value (steady-state signal), and the calculation formula of the steady-state signal is , where \(s(t)\) represents the steady-state signal, \(i\) represents the bearing sequence vibration signals at multiple moments, \(i = 1, 2, 3, \cdots, p\), \(a(i)\) represents the autoregressive coefficient corresponding to the \(i\)-th bearing sequence vibration signal, \(x(t - i)\) represents the bearing sequence vibration signal at the \(t - i\) moment. The steady-state signal represents the relatively stable part of the signal that can be predicted by historical data, usually reflecting the vibration characteristics of the bearing under normal working conditions. By subtracting the steady-state signal from the original preprocessed signal, the residual signal can be obtained. The calculation formula of the residual signal is , where \(e(t)\) represents the residual signal, \(x(t)\) is the current signal value, and \(s(t)\) represents the steady-state signal. The residual signal mainly contains those parts that do not conform to the prediction of the AR model, usually containing the vibration changes caused by faults or abnormalities, thus providing a direct basis for fault feature extraction. Through AR model decomposition, the vibration signal is divided into a steady-state signal and a residual signal, realizing the effective separation of normal operation characteristics and abnormal fault characteristics. By processing the signal sequences at multiple moments and dynamically obtaining the autoregressive coefficients, the AR model can better adapt to the changes in bearing vibration under different working conditions, improving the robustness and prediction accuracy of the model.

[0089] In a preferred embodiment, the root mean square of the steady-state signal and the kurtosis of the residual signal are obtained, and an optimized evaluation index is obtained according to the root mean square of the steady-state signal and the kurtosis of the residual signal. Within the preset autoregressive order range, the optimized evaluation indexes at each order are calculated by traversing, and the target order is obtained according to multiple optimized evaluation indexes. The steps of obtaining the preliminary fault features according to the target order include:

[0090] S301. Obtain the preset autoregressive order range;

[0091] S302. Obtain the root mean square of the steady-state signal and the kurtosis of the residual signal for each autoregressive order within the preset autoregressive order range;

[0092] S303. Obtain the optimized evaluation index for each autoregressive order according to the root mean square of the steady-state signal and the kurtosis of the residual signal for each autoregressive order;

[0093] S304. Arrange the optimized evaluation indexes of multiple autoregressive orders in descending order to obtain an order ranking table;

[0094] S305. Select the first optimized evaluation index according to the order ranking table, and obtain the corresponding autoregressive order according to the first optimized evaluation index, and mark it as the target order;

[0095] S306. Obtain the preliminary steady-state signal and the preliminary residual signal according to the target order and the autoregressive model.

[0096] In the above steps S301 to S306, a range of autoregressive model orders that may effectively describe the bearing vibration signal is preset in advance. This range covers various possible orders and provides a basis for subsequent model parameter selection. For each autoregressive order within the preset order range, the preprocessed signal is decomposed using the corresponding AR model to obtain the steady-state signal and the residual signal respectively. Then, the root mean square (RMS) of the steady-state signal is calculated to reflect the normal vibration energy; at the same time, the kurtosis of the residual signal is calculated to reflect the spike characteristics in the signal. Usually, fault information will increase the kurtosis value. According to the RMS of the steady-state signal and the kurtosis of the residual signal calculated at each order, an optimized evaluation index is constructed (such as using the ratio of the kurtosis of the residual signal to the RMS of the steady-state signal, i.e., RKSR). The calculation formula of the optimized evaluation index is , where RKSR represents the optimized evaluation index, represents the kurtosis of the residual signal, The calculation formula of is , where t represents the number of the residual signal at multiple moments, t = 1, 2, 3…T, and e(t) represents the residual signal at the t-th moment, represents the average value of the vibration signal x(t), represents the root mean square of the steady-state signal, The calculation formula of , represents the standard deviation of the steady-state signal x(t), t represents the number of the steady-state signal at multiple moments, t = 1, 2, 3…T, represents the steady-state signal at the t-th moment, which is used to measure the extraction effect of fault features at this order. The higher this index, the more it usually means that the model can more effectively distinguish normal signals and fault features. Sort the optimized evaluation indexes of each order calculated within the preset order range from largest to smallest to form an order ranking table. This ranking helps to identify which order of the model has the best fault feature extraction effect. According to the order ranking table, select the optimized evaluation index ranked first and determine the corresponding autoregressive order as the target order. Using the selected target order and the corresponding AR model parameters, reconstruct the autoregressive model to obtain the preliminary steady-state signal and the preliminary residual signal respectively. The residual signal contains obvious fault features. By traversing and comparing the optimized evaluation indexes of each order within the preset order range, the best AR model order is adaptively selected, making the model parameters more in line with the characteristics of the actual vibration signal and improving the accuracy of fault feature extraction. The optimized evaluation index constructed using the RMS of the steady-state signal and the kurtosis of the residual signal can effectively distinguish normal vibration and abnormal fault information, making the fault features more obvious in the preliminary residual signal extracted at the best order, thus reducing the risk of false alarms and missed alarms.

[0097] In a preferred embodiment, obtaining a preset number of iterations, returning the preliminary steady-state signal as a preprocessed signal to the step of obtaining a steady-state signal and a residual signal according to the autoregressive model and the preprocessed signal until the preset number of iterations is reached, and obtaining the accumulated residual signal according to the preset number of iterations includes:

[0098] S401, obtaining a preset number of iterations;

[0099] S402, returning the preliminary steady-state signal as a preprocessed signal to the step of obtaining a steady-state signal and a residual signal according to the autoregressive model and the preprocessed signal, and obtaining the number of returns;

[0100] S403, obtaining an iteration period whose return number meets the preset iteration number;

[0101] S404, obtaining multiple residual signals obtained after returning to the step of obtaining a steady-state signal and a residual signal according to the autoregressive model and the preprocessed signal each time during the iteration period;

[0102] S405: Aggregate the multiple residual signals to obtain an accumulated residual signal.

[0103] As in the above steps S401 to S405, a fixed number of iterations is pre-set to ensure that the entire signal decomposition process can be iterated enough, so that the fault information remaining in the preliminary steady-state signal is gradually stripped off and accumulated into the residual signal. The obtained preliminary steady-state signal is used as a new preprocessing signal and re-input into the signal decomposition process based on the autoregressive model. Each return is equivalent to further stripping of the original signal. The purpose is to extract purer and more concentrated fault features through continuous iteration. At the same time, the number of returns is recorded. According to the preset number of iterations, the time period that meets the iteration conditions is selected, that is, after completing the specified number of returns, the iterative processing is terminated. In each iteration, the residual signal obtained by decomposition according to the autoregressive model is extracted. These signals The signal contains the fault information after gradual stripping, records the residual signals obtained in all iteration cycles, and summarizes or accumulates the residual signals obtained in each iteration process to form an accumulated residual signal. This signal integrates the results of multiple iterations. The fault characteristics are significantly enhanced due to multiple stripping, making it easier to be identified by subsequent processing steps. By feeding the preliminary steady-state signal back to the autoregressive model decomposition process and iterating multiple times, each iteration will further strip out the weak fault signal. The accumulated residual signal can effectively amplify the early fault characteristics and enhance the detection sensitivity. The iterative process continuously eliminates the steady-state (normal) components through multiple decompositions, gradually reducing the influence of noise and irrelevant information, making the fault components in the final accumulated residual signal more prominent and reducing the risk of misjudgment.

[0104] In a preferred embodiment, the step of obtaining an iterative period in which the number of returns meets a preset number of iterations includes:

[0105] S4031. Obtain the time node when the preliminary steady-state signal is first returned as a preprocessed signal to the step of obtaining a steady-state signal and a residual signal according to the autoregressive model and the preprocessed signal, and mark it as the start time of the iterative period;

[0106] S4032. Obtain the time node when the number of returns is equal to the preset number of iterations and the step of returning the preliminary steady-state signal as a preprocessed signal to the step of obtaining a steady-state signal and a residual signal according to the autoregressive model and the preprocessed signal is completed, and mark it as the end time of the iterative period;

[0107] S4033. Obtain the iterative period according to the start time and the end time of the iterative period.

[0108] In the above steps S4031 to S4033, when the preliminary steady-state signal is first returned as a preprocessed signal to the step of signal decomposition based on the autoregressive model, record the time node at this time and mark this time point as the start time of the iterative period. When the number of returns reaches the preset number of iterations, record the time node at this time and mark this time point as the end time of the iterative period. Use the recorded start time and end time to determine the entire iterative period, that is, the time interval experienced from the first feedback to reaching the preset number of iterations. This iterative period is used to subsequently summarize and accumulate the residual signals obtained in each iterative process to ensure the integrity and data consistency of the iterative process.

[0109] In a preferred embodiment, the step of performing spectral amplitude modulation processing on the accumulated residual signal to obtain a conditioned fault signal includes:

[0110] S501. Perform a Fourier transform on the accumulated residual signal to obtain the amplitude spectrum and phase spectrum of the accumulated residual signal;

[0111] S502. Perform an exponential operation on the amplitude spectrum to obtain a corrected amplitude spectrum;

[0112] S503. Obtain a corrected signal frequency spectrum according to the corrected amplitude spectrum and the phase spectrum;

[0113] S504. Perform an inverse Fourier transform on the corrected signal frequency spectrum to obtain a time-domain corrected signal;

[0114] S505. Obtain the exponential range of the exponential operation and combine it with the time-domain corrected signal to obtain multiple corrected optimization evaluation indicators;

[0115] S506. Arrange the multiple corrected optimization evaluation indicators in descending order to obtain a corrected index ranking table;

[0116] S507. Select the first corrected optimization evaluation index according to the corrected index sorting table, obtain the corresponding index according to the first corrected optimization evaluation index, and mark it as the target index;

[0117] S508. Obtain the conditioning fault signal according to the target index and the autoregressive model.

[0118] It should be noted that the index range is from 0.8 to 1.0.

[0119] As in the above steps S501 to S508, perform Fourier transform on the cumulative residual signal, convert the time-domain signal to the frequency domain, respectively obtain the amplitude spectrum and phase spectrum of the signal, perform exponential operation on the amplitude spectrum (the index range is set from 0.8 to 1.0) to obtain the corrected amplitude spectrum. The exponential operation can adjust the attenuation characteristics of the signals in each frequency band of the amplitude spectrum, highlight the fault characteristics, and at the same time suppress the noise interference. Combine the corrected amplitude spectrum with the original phase spectrum to construct the spectrum of the corrected signal. Perform inverse Fourier transform on the spectrum of the corrected signal to convert the frequency-domain signal back to the time domain, so as to obtain the time-domain corrected signal. At this time, the corrected signal has initially completed the enhancement of the fault characteristics and the suppression of the interference signal. Using the preset index range (0.8 to 1.0), combined with the time-domain corrected signal, calculate multiple corrected optimization evaluation indicators. These indicators reflect the contrast between the fault characteristics and the noise in the signal under different exponential operations, providing a quantitative basis for selecting the best index. Arrange all the calculated corrected optimization evaluation indicators in descending order to obtain the corrected index sorting table. The sorting table intuitively shows the processing effects under each index, facilitating comparison and selection of the best parameters. Select the first (i.e., the optimal) corrected optimization evaluation index from the sorting table, determine the corresponding index value, and mark it as the target index. This target index is considered to make the fault characteristics in the conditioned signal the most prominent and the interference the smallest. Using the target index and the autoregressive model, combined with the foregoing processing results, further adjust the signal to finally obtain the conditioning fault signal. This signal shows enhanced fault characteristics and obvious noise suppression in the time domain. The exponential operation processing can perform non-linear modulation on the amplitude spectrum, so that the fault signal is amplified in the frequency domain, while the noise and random pulse interference are relatively suppressed, thereby improving the recognition rate of the fault characteristics. By modulating the amplitude spectrum, the amplitudes of the noise and irrelevant signals are reduced. After further restoring to the time domain through inverse Fourier transform, the noise interference in the conditioning fault signal is effectively controlled, which helps to achieve more accurate fault diagnosis. Using the corrected optimization evaluation index and the index sorting method, the optimal modulation index can be automatically selected within the index range of 0.8 to 1.0, ensuring the best fault signal conditioning effect under different working conditions.

[0120] In a preferred embodiment, the steps of performing fast spectral kurtosis processing on the conditioned fault signal to obtain bearing fault characteristics include:

[0121] S601. Perform frame segmentation processing on the conditioned fault signal to obtain multiple frequency band signals;

[0122] S602. Obtain the signal kurtosis of each frequency band signal and generate a fast spectral kurtosis diagram;

[0123] S603. Obtain the highest frequency band in the fast spectral kurtosis diagram as the target frequency band;

[0124] S604. Obtain a filtered and conditioned signal according to the target frequency band and the conditioned fault signal;

[0125] S605. Perform Hilbert transform on the filtered and conditioned signal to obtain an envelope signal;

[0126] S606. Perform fast Fourier transform on the envelope signal to obtain an envelope spectrum;

[0127] S607. Obtain the fault characteristic frequency in the envelope spectrum and obtain the bearing fault type according to the fault characteristic frequency.

[0128] In the above steps S601 to S607, the conditioning fault signal is frame - processed, and the entire signal is divided into several short time periods or multiple frequency - band signals, so that the signal characteristics within each frame are relatively stable. For each frequency - band signal after frame - division, its signal kurtosis is calculated. By plotting the kurtosis values of each frequency band into a fast spectral kurtosis diagram, the concentration degree of signal energy distribution in different frequency bands can be visually displayed. The frequency band with the highest kurtosis value is selected from the fast spectral kurtosis diagram as the target frequency band. This frequency band usually corresponds to the abnormal signal or impact characteristics caused by the fault. According to the target frequency - band information, the conditioning fault signal is band - pass filtered to filter out the interference of other frequency bands and only retain the signal components within the target frequency band. In the signal after filtering and conditioning, the fault characteristics are more prominent. The filtered and conditioned signal is subjected to Hilbert transform to calculate its envelope signal. The envelope signal reflects the trend of the amplitude of the original signal changing with time. The envelope signal is subjected to fast Fourier transform to convert it to the frequency domain to obtain the envelope spectrum. The amplitudes of the various frequency components in the envelope spectrum can visually reflect the periodic impact or vibration characteristics in the fault signal. The fault characteristic frequency is extracted from the envelope spectrum. This frequency is often closely related to the bearing structure and defect characteristics. By comparing the characteristic frequency with the pre - established fault - characteristic database or empirical model, the bearing fault type is finally determined. Frame - processing and fast spectral kurtosis processing can quickly locate the abnormal frequency band in the signal, realize the rapid extraction of the fault characteristic frequency, improve the signal - processing speed. By band - pass filtering and Hilbert transform to extract the envelope signal, the fault impact signal is further enhanced, making the fault characteristics more obvious in the frequency spectrum, which helps to reduce the risks of false alarms and missed alarms. The fast Fourier transform and the fault - frequency determination process enable the entire signal - processing and fault - diagnosis process to be carried out in real time, providing a timely basis for equipment maintenance and preventive maintenance.

[0129] And, a terminal for rapid extraction of bearing fault characteristics, comprising:

[0130] One or more processors;

[0131] A storage device having one or more programs stored thereon;

[0132] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for rapid extraction of bearing fault characteristics.

[0133] The above - mentioned is only the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention. The structures, devices, and operation methods not specifically described and explained in the present invention, unless otherwise specified and limited, are implemented according to the conventional means in the art.

Claims

1. A method for extracting bearing fault characteristics, characterized in that, Including: Obtain the bearing vibration signal, and preprocess the bearing vibration signal to obtain a preprocessed signal; Obtain an autoregressive model, and obtain a steady-state signal and a residual signal according to the autoregressive model and the preprocessed signal; Obtain the root mean square of the steady-state signal and the kurtosis of the residual signal, obtain an optimized evaluation index according to the root mean square of the steady-state signal and the kurtosis of the residual signal, traverse and calculate the optimized evaluation indexes at each order within a preset autoregressive order range, and obtain a target order according to multiple optimized evaluation indexes, and obtain preliminary fault characteristics according to the target order, where the preliminary fault characteristics include a preliminary steady-state signal and a preliminary residual signal; Obtain a preset number of iterations, return the preliminary steady-state signal as the preprocessed signal to the step of obtaining the steady-state signal and the residual signal according to the autoregressive model and the preprocessed signal until the preset number of iterations is reached, and obtain an accumulated residual signal according to the preset number of iterations; Perform spectral amplitude modulation processing on the accumulated residual signal to obtain a conditioned fault signal; Perform fast spectral kurtosis processing on the conditioned fault signal to obtain bearing fault characteristics.

2. The method for extracting bearing fault characteristics according to claim 1, wherein The step of obtaining the bearing vibration signal and preprocessing the bearing vibration signal to obtain a preprocessed signal includes: Obtain the bearing vibration signal; Filter the bearing vibration signal according to a low-pass filter to obtain a vibration filtered signal; Normalize the vibration filtered signal to obtain a preprocessed signal.

3. The method for extracting bearing fault characteristics according to claim 1, wherein The step of obtaining an autoregressive model and obtaining a steady-state signal and a residual signal according to the autoregressive model and the preprocessed signal includes: Obtain an autoregressive model; Obtain the bearing sequence vibration signals at multiple moments according to the bearing vibration signal; Obtain the autoregressive coefficients corresponding to each bearing sequence vibration signal; Obtain a steady-state signal according to the autoregressive model, each bearing sequence vibration signal at multiple moments, and the autoregressive coefficients corresponding to each bearing sequence vibration signal; Obtain a residual signal according to the autoregressive model and the steady-state signal.

4. The method for extracting bearing fault characteristics according to claim 1, characterized in that, The step of obtaining the root mean square of the steady-state signal and the kurtosis of the residual signal, obtaining an optimized evaluation index according to the root mean square of the steady-state signal and the kurtosis of the residual signal, traversing and calculating the optimized evaluation indexes at each order within a preset autoregressive order range, and obtaining preliminary fault characteristics according to multiple optimized evaluation indexes includes: Obtain a preset autoregressive order range; Obtain the root mean square of the steady-state signal and the kurtosis of the residual signal at each autoregressive order within the preset autoregressive order range; Obtain the optimized evaluation index at each autoregressive order according to the root mean square of the steady-state signal and the kurtosis of the residual signal at each autoregressive order; Arrange the optimized evaluation indexes of multiple autoregressive orders in descending order to obtain an order sorting table; Select the first optimized evaluation index according to the order sorting table, and obtain the corresponding autoregressive order according to the first optimized evaluation index, and mark it as the target order; Obtain a preliminary steady-state signal and a preliminary residual signal according to the target order and the autoregressive model.

5. The method for extracting bearing fault characteristics according to claim 1, wherein Obtain a preset number of iterations, return the preliminary steady-state signal as a preprocessed signal to the step of obtaining the steady-state signal and the residual signal according to the autoregressive model and the preprocessed signal, until the preset number of iterations is reached. The step of obtaining the cumulative residual signal according to the preset number of iterations includes: Obtain a preset number of iterations; Return the preliminary steady-state signal as a preprocessed signal to the step of obtaining the steady-state signal and the residual signal according to the autoregressive model and the preprocessed signal, and obtain the return count; Obtain the iteration period when the return count meets the preset number of iterations; Obtain multiple residual signals obtained each time after returning to the step of obtaining the steady-state signal and the residual signal according to the autoregressive model and the preprocessed signal during the iteration period; Summarize the multiple residual signals to obtain the cumulative residual signal.

6. The method for extracting bearing fault characteristics according to claim 5, wherein The step of obtaining the iteration period when the return count meets the preset number of iterations includes: Obtain the time node of the first time the preliminary steady-state signal is returned as a preprocessed signal to the step of obtaining the steady-state signal and the residual signal according to the autoregressive model and the preprocessed signal, and mark it as the start time of the iteration period; Obtain the time node when the return count is equal to the preset number of iterations and the step of returning the preliminary steady-state signal as a preprocessed signal to the step of obtaining the steady-state signal and the residual signal according to the autoregressive model and the preprocessed signal is completed, and mark it as the end time of the iteration period; Obtain the iteration period according to the start time and the end time of the iteration period.

7. The method for extracting bearing fault characteristics according to claim 1, wherein The step of performing spectral amplitude modulation processing on the cumulative residual signal to obtain the conditioned fault signal includes: Perform Fourier transform on the cumulative residual signal to obtain the amplitude spectrum and phase spectrum of the cumulative residual signal; Perform exponential operation processing on the amplitude spectrum to obtain the corrected amplitude spectrum; Obtain the corrected signal spectrum according to the corrected amplitude spectrum and the phase spectrum; Perform inverse Fourier transform on the corrected signal spectrum to obtain the time-domain corrected signal; Obtain the exponential range of the exponential operation processing, and combine the time-domain corrected signal to obtain multiple corrected optimization evaluation indexes; Arrange the multiple corrected optimization evaluation indexes in descending order to obtain the corrected index sorting table; Select the first corrected optimization evaluation index according to the corrected index sorting table, and obtain the corresponding index according to the first corrected optimization evaluation index, and mark it as the target index; Obtain the conditioned fault signal according to the target index and the autoregressive model.

8. The method for extracting bearing fault characteristics according to claim 7, characterized in that, The exponential range is from 0.8 to 1.

0.

9. The method for extracting bearing fault characteristics according to claim 1, wherein The step of performing fast spectral kurtosis processing on the conditioned fault signal to obtain the bearing fault characteristics includes: Perform frame division processing on the conditioned fault signal to obtain multiple frequency band signals; Obtain the signal kurtosis of each frequency band signal and generate a fast spectral kurtosis diagram; Obtain the highest frequency band in the fast spectral kurtosis diagram as the target frequency band; Obtain the filtered conditioned signal according to the target frequency band and the conditioned fault signal; Perform Hilbert transform on the filtered conditioned signal to obtain the envelope signal; Perform fast Fourier transform on the envelope signal to obtain the envelope spectrum; Obtain the fault characteristic frequency in the envelope spectrum, and obtain the bearing fault type according to the fault characteristic frequency.

10. A terminal for bearing fault feature extraction, characterized in that Includes: One or more processors; A storage device on which one or more programs are stored; When one or more programs are executed by one or more processors such that the one or more processors implement the method for bearing fault feature extraction according to any one of claims 1 to 8.

Images