On-line monitoring method for bearing abnormal sound based on voiceprint frequency characteristic analysis

Abstract

This invention belongs to the field of fault diagnosis technology, specifically relating to an online monitoring method for bearing abnormal noise based on acoustic signature frequency feature analysis. The method includes the following steps: acquiring a continuous time-domain acoustic signature signal of the bearing during operation; performing detrending and pre-whitening processing on the continuous time-domain signal to obtain a preprocessed discrete signal sequence; performing a Fourier transform on the discrete signal sequence to obtain the amplitude spectrum; calculating the spectral fluctuation complexity index based on the amplitude spectrum; determining the optimal decomposition level for variational mode decomposition based on the spectral fluctuation complexity index; and using the optimal decomposition level to decompose the discrete signal sequence to obtain several intrinsic mode components. This invention effectively distinguishes between mechanical fault impacts and random environmental interference through adaptive parameter optimization and a dual feature interlocking mechanism, significantly improving the accuracy and robustness of online monitoring.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis technology, specifically relating to an online monitoring method for bearing abnormal noise based on acoustic frequency characteristic analysis. Background Technology

[0002] As a core component of rotating machinery, the health of bearings directly affects the safe, stable operation, and production efficiency of the entire production system. In modern industrial continuous and intelligent production scenarios, using acoustic or vibration sensors to collect signals for abnormal noise monitoring has become a primary means of judging the operating status of bearings and achieving early fault warning. Currently, the industry generally adopts the variational mode decomposition (VMD) algorithm to decompose complex non-stationary acoustic signals into several intrinsic mode components, and then extracts fault features through envelope demodulation analysis to achieve bearing condition identification.

[0003] However, existing VMD-based monitoring technologies still face significant challenges in practical industrial applications. First, the decomposition effect of VMD algorithms is highly sensitive to parameters such as the number of decomposition levels K and the penalty factor. Under conditions of large fluctuations in workshop speed, variable loads, and complex background noise, fixed parameter settings often lead to under-decomposition of signals, causing weak fault characteristics to be drowned out by noise, or over-decomposition, generating spurious modal components. There is a lack of an effective mechanism that can adaptively adjust the number of decomposition levels based on the real-time frequency characteristics of the signal.

[0004] Secondly, in the component selection and fault identification stages, existing technologies typically rely solely on kurtosis as a single indicator for component optimization and fault determination. However, industrial sites are subject to numerous non-fault interferences such as random impacts and airflow noise. These interferences also generate extremely high kurtosis values, causing monitoring systems to frequently misclassify normal conditions as bearing failures, resulting in a persistently high false positive rate. This makes it difficult to effectively distinguish between genuine mechanical fault impacts and random environmental interferences, severely limiting the practicality and reliability of the monitoring system. Summary of the Invention

[0005] This invention provides an online monitoring method for bearing abnormal noise based on acoustic frequency feature analysis, in order to solve the technical problems in the prior art where VMD decomposition parameters are difficult to adaptively match and the false alarm rate is high due to single feature recognition.

[0006] This invention provides an online monitoring method for bearing abnormal noise based on acoustic frequency feature analysis, comprising the following steps: Acquire the continuous time-domain acoustic signature signal of the bearing during operation, perform detrending and pre-whitening processing on the continuous time-domain signal to obtain the pre-processed discrete signal sequence; The discrete signal sequence is subjected to Fourier transform to obtain the amplitude spectrum. The spectral fluctuation complexity index is calculated based on the amplitude spectrum. The optimal number of decomposition layers for variational mode decomposition is determined based on the spectral fluctuation complexity index. The discrete signal sequence is then decomposed using the optimal number of decomposition layers to obtain several intrinsic mode components. The fault characteristic period is determined based on the bearing geometric parameters and rotational speed. The impact period coordination factor of each intrinsic modal component is calculated. The impact period coordination factor is used to characterize the degree to which the intrinsic modal component simultaneously possesses high-energy impact characteristics and fault period repetition characteristics. The intrinsic mode component with the largest impact cycle coordination factor is selected as the optimal component. The fault severity index is calculated based on the envelope spectrum energy distribution of the optimal component. The fault severity index is compared with a preset threshold. When the preset threshold is exceeded, an alarm is triggered to realize online monitoring of bearing abnormal noise.

[0007] Compared to existing technologies where fixed VMD parameters can lead to signal under-decomposition or spurious components under varying speeds and loads, this invention achieves adaptive matching of the number of VMD decomposition layers by constructing a spectral fluctuation complexity index and introduces an impact cycle synergy factor to establish a dual-feature interlocking mechanism. In complex industrial application scenarios filled with random impacts and electromagnetic noise, it can effectively distinguish between real mechanical fault impacts and environmental interference, significantly reducing the false alarm rate of online monitoring and improving the accuracy of early weak fault detection.

[0008] Furthermore, the complexity index of spectral fluctuations calculated based on the amplitude spectrum includes: Based on the degree of deviation of the amplitude at each frequency point in the amplitude spectrum from the average amplitude of the entire frequency band, the macroscopic dispersion term reflecting the non-uniformity of the spectrum energy distribution is calculated. Based on the sum of amplitude differences between adjacent frequency points in the amplitude spectrum, a micro-roughness term reflecting the abundance of local spectral spikes is calculated. By combining the macroscopic dispersion term with the microscopic roughness term after logarithmic processing, the spectral fluctuation complexity index is obtained. The magnitude of the spectral fluctuation complexity index reflects the complexity of the signal's frequency domain components and the richness of its independent components.

[0009] By introducing the spectral fluctuation complexity index, this invention abandons simple energy statistics and instead deeply analyzes frequency domain information from two dimensions: macroscopic dispersion and microscopic roughness. This index is positively correlated with the number of independent components contained in the signal, thus providing a precise physical basis for the adaptive selection of the VMD decomposition layer. It effectively avoids the problems of under-decomposition or over-decomposition caused by fixed parameters and ensures the integrity of feature extraction.

[0010] Furthermore, the impact period coordination factor for each intrinsic modal component is calculated separately, including: An impulse characteristic term is constructed, which is determined based on the ratio of the peak absolute value to the effective value of the intrinsic mode component time-domain waveform, and is used to characterize the impulse significance of the signal. A periodic feature term is constructed, which is determined based on the ratio of the sum of the amplitudes of the autocorrelation functions of the intrinsic mode components at the fault characteristic period and its harmonics to the total energy at zero delay, and is used to characterize the periodic repeatability of the signal. The impact characteristic term and the periodic characteristic term are multiplied to obtain the impact period coordination factor. The interlocking logic of the two is used to eliminate random interference or periodic noise with only a single characteristic.

[0011] By constructing an impact cycle coordination factor, this invention establishes a feature interlocking mechanism. This factor only exhibits a high value when the signal simultaneously possesses high impact energy and specific fault cycle repeatability. This mechanism acts like a filter, accurately eliminating common workshop noises such as random material dropping noise and motor electromagnetic noise, thus fundamentally solving the problem of high false alarm rates caused by existing technologies relying on a single kurtosis index.

[0012] Furthermore, the calculation of the fault severity index based on the envelope spectrum energy distribution of the optimal components includes: Calculate the energy integral value of the envelope spectrum of the optimal component within the fault characteristic frequency and its neighborhood bandwidth; Calculate the total energy integral value of the envelope spectrum of the optimal component over the entire analysis frequency band; Calculate the ratio of the energy integral value to the total energy integral value to obtain the frequency domain energy proportion; The fault severity index is obtained by weighting the frequency domain energy proportion with the impact period synergy factor of the optimal component, thereby comprehensively characterizing the energy concentration of the fault in the frequency domain and the impact period characteristic intensity in the time domain.

[0013] By calculating the fault severity index, this invention no longer relies solely on features of a single dimension, but instead integrates the energy concentration in the frequency domain with the impact period characteristics in the time domain for evaluation. This multi-dimensional comprehensive consideration makes the fault determination results more robust, maintaining the objectivity and accuracy of the evaluation results even under extreme conditions such as sensor signal fluctuations or sudden changes in background noise, providing reliable data support for predictive maintenance of equipment.

[0014] Furthermore, the specific preprocessing steps include: using polynomial fitting to eliminate DC bias and extremely low frequency trend terms in the signal; and using a linear prediction filter to pre-whiten the signal to flatten the background noise spectrum and highlight periodic narrowband components.

[0015] To address the unavoidable baseline drift and background colored noise interference during acoustic sensor acquisition, this invention utilizes a combination of polynomial fitting and linear prediction filters for preprocessing. This effectively flattens the background noise spectrum without compromising useful features, enhances the visibility of periodic narrowband fault components in the original signal, and lays a clean data foundation for subsequent feature extraction.

[0016] Furthermore, determining the optimal number of decomposition layers for variational mode decomposition based on the spectral fluctuation complexity index includes: calculating the optimal number of decomposition layers through a linear mapping function, establishing a positive correlation between the spectral fluctuation complexity index and the number of decomposition layers through the linear mapping function, and rounding the calculation result to a preset integer range.

[0017] Furthermore, the fault characteristic period is calculated based on the number of bearing balls, diameter, contact angle, and real-time rotational speed, and the time unit is converted into the number of sampling points.

[0018] Furthermore, the energy integral value is obtained by integrating the envelope spectrum of the optimal component over a frequency range from the fault characteristic frequency minus the preset bandwidth to the fault characteristic frequency plus the preset bandwidth.

[0019] Furthermore, the preset threshold is set based on the statistical characteristics of historical normal operation data. When the calculated fault severity index continuously exceeds the preset threshold, the system determines that there is a bearing fault.

[0020] Furthermore, a high-frequency acoustic-vibration composite sensor is installed in the vertical direction of the bearing housing or in the load area to collect the acoustic time-domain continuous signal of the bearing during operation.

[0021] The beneficial effects are as follows: This invention proposes a monitoring mechanism that adaptively optimizes VMD parameters based on spectral fluctuation complexity and accurately locks fault components using an impact cycle coordination factor. A mathematical mapping between signal frequency domain complexity and VMD decomposition parameters is established using the spectral fluctuation complexity index. This allows for finding the optimal decomposition level under different speeds and noise environments without manual intervention, avoiding feature omissions caused by fixed parameters. Simultaneously, the impact cycle coordination factor of this invention mandates that the signal must simultaneously satisfy two conditions: high impact and a specific fault cycle. This logical interlocking mechanism effectively filters out common workshop noises such as random material dropping noise and motor electromagnetic noise, significantly improving the monitoring's anti-interference capability. The formula design fully considers mathematical boundary conditions, ensuring that the algorithm will not experience logical collapse under extreme operating conditions, making it suitable for long-term stable operation of industrial-grade systems. Attached Figure Description

[0022] Figure 1 This is a flowchart of the online monitoring method for bearing abnormal noise based on acoustic frequency feature analysis in this invention.

[0023] Figure 2 This is a schematic diagram illustrating the current state of sample distribution in the traditional time-domain feature space in existing technologies.

[0024] Figure 3 This is a schematic diagram of the VMD parameter adaptive optimization space based on complexity index in this invention.

[0025] Figure 4 This is a schematic diagram of the characteristic two-dimensional clustering distribution based on the shock cycle synergy factor in this invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] An embodiment of the bearing noise online monitoring method based on acoustic frequency feature analysis provided by this invention: like Figure 1 As shown, the online monitoring method for bearing abnormal noise based on acoustic frequency feature analysis includes the following steps: S1, Voiceprint Data Acquisition and Basic Preprocessing.

[0028] First, a high-frequency acoustic-vibration composite sensor is installed in the vertical direction of the bearing housing or in the load area, and the sampling rate is set to... The above describes the acquisition of continuous time-domain acoustic signature signals during bearing operation. Since the original signal often contains DC components and colored background noise, preprocessing is required.

[0029] A polynomial fitting method is used to eliminate DC bias and extremely low-frequency trend terms in the signal. Next, pre-whitening processing is performed. Since industrial noise is mostly colored noise, a linear predictive filter is used to filter the detrended signal to flatten the background noise spectrum and highlight periodic narrowband fault components, ultimately obtaining the preprocessed discrete signal sequence. .

[0030] Through the above data acquisition and preprocessing steps, baseline drift and colored noise interference in the signal can be effectively removed, providing a cleaner data foundation for subsequent feature extraction.

[0031] S2, adaptive calculation of VMD decomposition layer number based on spectral fluctuation complexity.

[0032] For VMD decomposition layer number For problems that are difficult to determine, this step constructs an index that reflects the richness of frequency domain information, enabling automatic parameter matching. Specifically, it involves processing discrete signal sequences... Perform a Fast Fourier Transform and take the modulus to obtain the amplitude spectrum sequence. When a bearing has early failures or complex noise interference, its spectrum will show a lot of irregular spikes; while the spectrum of a healthy bearing is relatively smooth.

[0033] Based on this, a spectrum fluctuation complexity index is constructed. The calculation formula is as follows:

[0034] In the formula, The first in the amplitude spectrum The amplitude at each frequency point; This is the arithmetic mean of the amplitude spectrum across the entire frequency band. ; This represents the total number of frequency domain sampling points; It is a non-zero constant, taking the value of This is used to prevent the denominator from being zero. It should be noted that, to meet strict mathematical operation standards, the cumulative summation term of the amplitude difference in the above formula... Non-zero constants, before being substituted into the natural logarithm function for calculation, have all been implicitly divided by the unit reference parameter corresponding to the amplitude to achieve dimensionlessness, participating in the logarithmic mapping and subsequent addition with the constant 1 in pure numerical form. Similarly, non-zero constants added to physical quantities in subsequent formulas all implicitly contain the same physical dimensions as their corresponding terms.

[0035] This formula constructs a spectral fluctuation complexity index from two dimensions: macroscopic dispersion and microscopic roughness. Its core logic is to reflect the signal complexity by quantifying the distribution characteristics of the frequency domain amplitude spectrum. The first part of the formula uses the ratio of the standard deviation to the mean of the amplitude spectrum to characterize macroscopic dispersion, measuring the non-uniformity of spectral energy distribution across the entire frequency band. A non-zero constant is added to the denominator to avoid division by zero; a larger value indicates stronger concentration of spectral energy and more significant differences in frequency domain components. The second part performs a logarithmic transformation on the sum of the absolute values ​​of amplitude differences between adjacent frequency points and adds 1 to characterize microscopic roughness. The logarithmic processing weakens the influence of extreme values ​​while amplifying the variation characteristics of local spectral glitches, accurately reflecting the complexity of frequency domain details. By coupling the two dimensions through multiplication, a comprehensive quantification of the macroscopic distribution and microscopic details of the signal in the frequency domain is achieved. The final index value is positively correlated with the richness of independent signal components, providing a quantitative basis that fits the physical characteristics of the signal for the adaptive selection of the VMD decomposition layer, ensuring that the decomposition parameters match the actual frequency domain characteristics.

[0036] To facilitate understanding, a calculation example is given here: assuming frequency domain sampling points arithmetic mean of amplitude spectrum Assuming the calculated amplitude standard deviation is 2, then the first half is... Assuming the sum of the absolute values ​​of the first-order differences between adjacent points on the spectrum is 150, then the latter half is... .final If the signal spectrum is more complex and the differential sum increases, this index will also increase accordingly.

[0037] Calculate Then, through a linear mapping function The optimal decomposition level for the current signal is obtained, where, The scaling factor controls the impact of spectral fluctuation complexity on the number of decomposition layers K, determining approximately how many layers K changes for every unit change in complexity. This is the bias coefficient, used in the overall translation calculation results to limit K within the usable range of the project, avoiding under-decomposition due to too small a number of layers or over-decomposition due to too large a number of layers. For example... The value is limited to between 3 and 8, and the signal is decomposed using VMD to obtain... Each intrinsic mode component .

[0038] By calculating the spectral fluctuation complexity index and establishing its mapping relationship with the number of decomposition layers, the VMD parameters can be adaptively adjusted according to the current frequency domain characteristics of the signal, ensuring the effectiveness of the decomposition and avoiding the loss of signal features or the generation of spurious components due to fixed parameters.

[0039] S3, based on the impact cycle synergy factor for precise screening of fault components.

[0040] To eliminate random interference from the environment, screening criteria are designed based on the physical characteristics that real bearing fault signals must simultaneously possess high-energy impact and fault cycle repeatability. First, the theoretical fault characteristic cycle is calculated based on the bearing's geometric parameters and rotational speed. .

[0041] Next, a shock cycle synergistic factor is constructed. For each decomposed component The evaluation is conducted using the following calculation formula:

[0042] In the formula, For the first The absolute value of the peak value of the time-domain waveform of each IMF component; For the first The effective values ​​of each IMF component; For the autocorrelation function in delay The value at; This indicates the peak value of the autocorrelation at the 1st, 2nd, and 3rd times the fault cycle. This is the value of the autocorrelation function at zero delay.

[0043] The core logic of this formula is to comprehensively characterize the discriminability of a target signal from two dimensions: signal peak characteristics and autocorrelation energy features, providing a quantitative basis for signal feature extraction or anomaly identification. The formula consists of two key multiplicative coupling parts, taking into account both macroscopic peak differences and microscopic energy correlations. The first part uses a logarithmic transformation, using the maximum peak value of the signal... With root mean square value The ratio is the core, and a constant is added. To avoid the denominator being zero, The method involves transforming the compressible numerical range and weakening the interference of extreme peaks. Simultaneously, adding 1 ensures the result is non-negative. This part primarily quantifies the deviation between the signal peak value and the average energy; the greater the deviation, the more significant the characteristic of the signal peak. The latter part calculates the autocorrelation values ​​under three different fault cycles. The sum of the signals and the energy of the signal To calculate the ratio, we also introduce... Anomaly avoidance primarily involves quantifying the autocorrelation characteristics of signals at different time scales, reflecting the temporal correlation of signal energy. After coupling these two parts, the final index... It can comprehensively reflect the peak identification degree and energy correlation of the signal. The larger the value, the more prominent the signal characteristics, providing reliable quantitative support for subsequent signal analysis.

[0044] The following calculation example demonstrates the filtering logic of this formula: Assuming a certain component represents a real fault signal: its peak value... , effective value Then the left-hand side item Meanwhile, due to its periodicity, its autocorrelation value is high at the fault cycle. Assuming the sum of the autocorrelation values ​​over three cycles is 30, the total energy... Then the right-hand side is ,final .

[0045] Assume a certain component is random impulse noise: although the peak value is very high. , The left-hand term remains 2.58, but due to its lack of periodicity, it... The autocorrelation value at a given point is extremely low. Assuming the sum of the autocorrelation values ​​over three periods is only 1, then the term on the right-hand side is: ,final It is evident that the score for real faults is much higher than that for random noise.

[0046] By calculating the impact cycle coordination factor, the fault signal and random interference signal can be effectively distinguished. By utilizing the interlocking mechanism of physical characteristics, the false alarm rate is greatly reduced, ensuring that the selected component does indeed contain fault information.

[0047] S4, quantitative assessment and early warning of fault severity.

[0048] Based on the optimal component selected in step S3, i.e. The component with the largest value is denoted as Assess the severity of the current fault. First, assess the... Envelope spectrum analysis was performed to search for the actual spectral peak energies near the theoretical fault characteristic frequencies and their harmonics.

[0049] Constructing a fault severity index The calculation formula is as follows:

[0050] in In the optimal component envelope spectrum, at the fault characteristic frequency Left and right bandwidth Energy integral within the range; The optimal component envelope spectrum represents the total energy of the entire analysis frequency band. This is the maximum synergy factor value calculated in S3.

[0051] The core logic of this formula is to couple the energy proportion of the signal frequency band with the optimal feature identification degree to achieve accurate quantification of the fault degree, providing a reliable basis for fault diagnosis and assessment. The formula consists of the multiplication of two key parts, taking into account both energy distribution and feature effectiveness. The first part uses the energy of the target fault frequency band... With total energy The ratio is the core, and a constant is introduced. To avoid a denominator of zero, the main focus is on quantifying the proportion of fault-related energy in the total energy. A higher proportion indicates a more concentrated fault-related energy and more prominent fault characteristics. (The latter part...) The optimal feature identification index represents the selected feature parameters that best characterize the fault; the larger the value, the stronger the identification and reliability of the fault features. After the two parts are coupled, It can comprehensively reflect the degree of energy concentration and the effectiveness of the characteristics of the fault. The index value is positively correlated with the severity of the fault. It avoids the limitations of a single energy index and ensures that the quantitative results are consistent with the actual fault characteristics, providing scientific support for fault level classification.

[0052] Calculation example: Assuming the optimal component is selected Envelope analysis revealed the energy integral near the fault frequency. The total energy across the entire frequency band The energy percentage is: Final Fault Severity Index If the preset alarm threshold is 2.0, then at this time... The system will trigger an alarm.

[0053] By calculating the fault severity index, which combines the frequency domain energy proportion with the time domain co-operation characteristics, the health status of the bearing can be objectively and quantitatively reflected, providing reliable data support for equipment maintenance decisions.

[0054] To verify the effectiveness of the present invention, the following description is provided in conjunction with the accompanying drawings.

[0055] like Figure 2 As shown, the current sample distribution of existing technologies in the traditional time-domain feature space is illustrated. The horizontal axis of the figure represents the peak factor of the traditional time-domain signal, and the vertical axis represents the effective value of the traditional time-domain vibration energy. It can be seen that the circular points representing background environmental noise samples and the triangular points representing early weak fault samples are severely mixed in the central region, forming a feature overlap and confusion area. This indicates that it is impossible to distinguish between weak faults and noise by simply relying on peak values ​​or energy.

[0056] like Figure 3 As shown, the VMD parameter adaptive optimization space based on the complexity index of this invention is illustrated. The horizontal axis represents the number of variational mode decomposition layers, and the vertical axis represents the mode decomposition penalty factor. The fixed parameter points of the prior art fall into the inefficient region, while the optimal point of the adaptive optimization of this invention finds the efficient region through automatic climbing by the algorithm, which proves that this invention can automatically find the parameter combination that can extract the most complete features.

[0057] like Figure 4 As shown, the two-dimensional clustering distribution based on the impact cycle synergy factor of the present invention is illustrated. The horizontal axis represents the intensity of the impact feature extracted adaptively, and the vertical axis represents the impact cycle synergy factor. The sample points are clearly divided into three categories: noise in the lower left corner, non-fault impact interference in the lower right corner, and real faults in the upper right corner. The pentagram region is marked as the fault area, with clear boundaries from other regions, which strongly proves that the index of the present invention has a very strong classification ability and can effectively eliminate false alarms.

[0058] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for online monitoring of bearing abnormal noise based on acoustic frequency characteristic analysis, characterized in that, Includes the following steps: Acquire the continuous time-domain acoustic signature signal of the bearing during operation, perform detrending and pre-whitening processing on the continuous time-domain signal to obtain the pre-processed discrete signal sequence; The discrete signal sequence is subjected to Fourier transform to obtain the amplitude spectrum. The spectral fluctuation complexity index is calculated based on the amplitude spectrum. The optimal number of decomposition layers for variational mode decomposition is determined based on the spectral fluctuation complexity index. The discrete signal sequence is then decomposed using the optimal number of decomposition layers to obtain several intrinsic mode components. The fault characteristic period is determined based on the bearing geometric parameters and rotational speed. The impact period coordination factor of each intrinsic modal component is calculated. The impact period coordination factor is used to characterize the degree to which the intrinsic modal component simultaneously possesses high-energy impact characteristics and fault period repetition characteristics. The intrinsic mode component with the largest impact cycle coordination factor is selected as the optimal component. The fault severity index is calculated based on the envelope spectrum energy distribution of the optimal component. The fault severity index is compared with a preset threshold. When the preset threshold is exceeded, an alarm is triggered to realize online monitoring of bearing abnormal noise.

2. The online monitoring method for bearing abnormal noise based on acoustic frequency feature analysis according to claim 1, characterized in that, The complexity index for calculating spectral fluctuations based on amplitude spectra includes: Based on the degree of deviation of the amplitude at each frequency point in the amplitude spectrum from the average amplitude of the entire frequency band, the macroscopic dispersion term reflecting the non-uniformity of the spectrum energy distribution is calculated. Based on the sum of amplitude differences between adjacent frequency points in the amplitude spectrum, a micro-roughness term reflecting the abundance of local spectral spikes is calculated. By combining the macroscopic dispersion term with the microscopic roughness term after logarithmic processing, the spectral fluctuation complexity index is obtained. The magnitude of the spectral fluctuation complexity index reflects the complexity of the signal's frequency domain components and the richness of its independent components.

3. The online monitoring method for bearing abnormal noise based on acoustic frequency feature analysis according to claim 1, characterized in that, The impact period synergy factor for each intrinsic modal component was calculated separately, including: An impulse characteristic term is constructed, which is determined based on the ratio of the peak absolute value to the effective value of the intrinsic mode component time-domain waveform, and is used to characterize the impulse significance of the signal. A periodic feature term is constructed, which is determined based on the ratio of the sum of the amplitudes of the autocorrelation functions of the intrinsic mode components at the fault characteristic period and its harmonics to the total energy at zero delay, and is used to characterize the periodic repeatability of the signal. The impact characteristic term and the periodic characteristic term are multiplied to obtain the impact period coordination factor. The interlocking logic of the two is used to eliminate random interference or periodic noise with only a single characteristic.

4. The online monitoring method for bearing abnormal noise based on acoustic frequency feature analysis according to claim 1, characterized in that, The fault severity index is calculated based on the envelope spectrum energy distribution of the optimal components, including: Calculate the energy integral value of the envelope spectrum of the optimal component within the fault characteristic frequency and its neighborhood bandwidth; Calculate the total energy integral value of the envelope spectrum of the optimal component over the entire analysis frequency band; Calculate the ratio of the energy integral value to the total energy integral value to obtain the frequency domain energy proportion; The fault severity index is obtained by weighting the frequency domain energy proportion with the impact period synergy factor of the optimal component, thereby comprehensively characterizing the energy concentration of the fault in the frequency domain and the impact period characteristic intensity in the time domain.

5. The online monitoring method for bearing abnormal noise based on acoustic frequency feature analysis according to claim 1, characterized in that, The specific preprocessing steps include: using polynomial fitting to eliminate DC bias and extremely low frequency trend terms in the signal; and using a linear prediction filter to pre-whiten the signal in order to flatten the background noise spectrum and highlight periodic narrowband components.

6. The online monitoring method for bearing abnormal noise based on acoustic frequency feature analysis according to claim 2, characterized in that, Determining the optimal number of decomposition layers for variational mode decomposition based on the spectral fluctuation complexity index includes: calculating the optimal number of decomposition layers through a linear mapping function, establishing a positive correlation between the spectral fluctuation complexity index and the number of decomposition layers through the linear mapping function, and rounding the calculation result to a preset integer range.

7. The online monitoring method for bearing abnormal noise based on acoustic frequency feature analysis according to claim 3, characterized in that, The fault characteristic cycle is calculated based on the number of bearing balls, diameter, contact angle, and real-time rotation speed, and the time unit is converted into the number of sampling points.

8. The online monitoring method for bearing abnormal noise based on acoustic frequency feature analysis according to claim 4, characterized in that, The energy integral value is obtained by integrating the envelope spectrum of the optimal component over a frequency range from the fault characteristic frequency minus the preset bandwidth to the fault characteristic frequency plus the preset bandwidth.

9. The online monitoring method for bearing abnormal noise based on acoustic frequency feature analysis according to claim 4, characterized in that, The preset threshold is set based on the statistical characteristics of historical normal operation data. When the calculated fault severity index continues to exceed the preset threshold, the system determines that there is a bearing fault.

10. The online monitoring method for bearing abnormal noise based on acoustic frequency feature analysis according to claim 1, characterized in that, A high-frequency acoustic-vibration composite sensor is installed in the vertical direction of the bearing housing or in the load area to collect the continuous time-domain acoustic signal of the bearing during operation.

Images