A fault early warning method combining vibration signal envelope demodulation and statistical characteristics
By combining vibration signal envelope demodulation with statistical features, the shortcomings of early fault identification in rotating machinery are addressed, enabling effective monitoring of the fault's development from its initial stage, and improving the accuracy and reliability of predictive maintenance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN JIAHE INTEGRITY TECH
- Filing Date
- 2026-04-23
- Publication Date
- 2026-07-31
AI Technical Summary
Existing technologies are insufficient to effectively reflect the gradual process of early failures in rotating machinery. Traditional methods are easily affected by individual equipment differences and fluctuations in operating conditions, leading to false alarms or missed alarms, and thus failing to meet the requirements of predictive maintenance.
By combining vibration signal envelope demodulation and statistical feature analysis, and extracting the amplitude time series of fault characteristic frequencies, and utilizing dynamic control limits and statistical distribution characteristic parameters within the analysis window, early fault identification and stability judgment can be achieved.
It can identify abnormal changes in the early stages of a fault, improve the stability and reliability of early warnings, reduce false alarms and missed alarms, and enable tiered alerts and timely maintenance for early faults.
Smart Images
Figure CN122486779A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of fault early warning, and in particular to a fault early warning method that combines vibration signal envelope demodulation with statistical features. Background Technology
[0002] Rotating machinery is widely used in industrial settings such as wind power, metallurgy, petrochemicals, and rail transportation. Its critical components, such as bearings and gearboxes, are prone to early damage such as pitting, spalling, and cracking during long-term operation. Among existing equipment condition monitoring technologies, vibration signal analysis is one of the most commonly used fault identification methods. Traditional methods typically involve manual analysis of the spectral characteristics of vibration signals or setting fixed alarm thresholds for single indicators such as the effective value and peak value of vibration to determine whether the equipment is abnormal.
[0003] However, the above methods still have obvious shortcomings in engineering applications: on the one hand, the impact component generated by early faults is relatively weak and is often submerged in background noise and operating condition fluctuations. Relying only on overall indicators such as effective values or peak values, alarms are usually triggered only after the fault has developed to a certain extent, making it difficult to detect potential risks in a timely manner; on the other hand, the fixed threshold method is more sensitive to changes in rotational speed, load fluctuations, and individual equipment differences, which can easily lead to false alarms or missed alarms, affecting the reliability of the early warning results.
[0004] In addition, most existing methods focus on judging the instantaneous amplitude and lack the utilization of the statistical distribution change law of characteristic amplitude over a period of time. Therefore, they are difficult to effectively reflect the gradual process of fault development from its inception and cannot meet the requirements of predictive maintenance for early detection, stable judgment and continuous tracking.
[0005] Therefore, it is necessary to provide a new fault early warning method that combines vibration signal envelope demodulation with statistical characteristics to solve the above-mentioned technical problems. Summary of the Invention
[0006] To address the aforementioned technical problems, this invention provides a fault early warning method that combines vibration signal envelope demodulation with statistical features. This method solves the problem that existing technologies are unable to effectively reflect the gradual process of a fault from its inception to its development, and cannot meet the requirements of predictive maintenance for early detection, stable judgment, and continuous tracking.
[0007] The fault early warning method combining vibration signal envelope demodulation and statistical characteristics provided by this invention includes the following steps: Step 1: Acquire the vibration signal of the monitored equipment under operating conditions, perform envelope demodulation analysis on the vibration signal, and extract the amplitude time series corresponding to the preset fault characteristic frequency; Step 2: Perform statistical distribution analysis on the amplitude time series according to the preset analysis window, divide the amplitude time series into at least two preset amplitude intervals, and calculate the probability value of the number of data points in each preset amplitude interval relative to the total number of data points in the analysis window, so as to obtain the statistical characteristic parameters of the current analysis window; Step 3: Determine the changing trend of the statistical characteristic parameters based on the statistical characteristic parameters from multiple consecutive analysis windows; Step 4: When the probability of the high amplitude interval in the statistical feature parameters exceeds the corresponding dynamic control limit, or when the change trend meets the preset upward criterion, output a device fault warning signal.
[0008] Preferably, in step 1, before performing envelope demodulation analysis, the vibration signal is further bandpass filtered with the resonant frequency band of the monitored component as the center; The envelope demodulation analysis employs either Hilbert transform envelope demodulation or rectifier-low-pass filter envelope demodulation.
[0009] Preferably, in step 1, the preset fault characteristic frequency is at least one of the following: bearing outer ring fault characteristic frequency, bearing inner ring fault characteristic frequency, rolling element fault characteristic frequency, cage fault characteristic frequency, or gear meshing frequency.
[0010] Preferably, in step 2, the statistical distribution analysis includes setting a low-amplitude interval and a high-amplitude interval, and calculating, within each analysis window, the proportion P of data points in the amplitude time series falling into the high-amplitude interval relative to the total number of data points within that analysis window. H , will the P H As a statistical characteristic parameter characterizing the change in the impact intensity of a fault.
[0011] Preferably, in step 4, the dynamic control limit is determined based on the statistical results of the corresponding statistical characteristic parameters in the historical health operation data of the equipment, and is updated as new health operation data is introduced.
[0012] Preferably, step 4 further includes calculating the effective value of the vibration signal within the same analysis window; When the statistical characteristic parameter exceeds the dynamic control limit or the change trend meets the preset rising criterion, and the effective value does not reach the preset effective value alarm threshold, an early warning signal is output.
[0013] The beneficial effects of this invention are as follows: 1. This invention extracts the amplitude time series corresponding to the fault characteristic frequencies by envelope demodulating the vibration signal, and then analyzes it in conjunction with statistical distribution characteristics, which can effectively highlight early weak impact information. Compared with traditional methods that rely solely on overall indicators such as vibration RMS value and peak value, this invention can identify abnormal changes in the early stages of fault development, thereby achieving earlier fault warning.
[0014] 2. This invention does not rely solely on the instantaneous amplitude at a single moment for judgment. Instead, it calculates statistical characteristic parameters such as the probability of high amplitude intervals within the analysis window and combines them with the changing trends of multiple consecutive analysis windows. Therefore, it can reflect the gradual change pattern of fault characteristics from weak to strong and from few to many, which is more consistent with the actual fault development process of equipment.
[0015] 3. Traditional fixed threshold methods are easily affected by individual equipment differences, load variations, and fluctuations in operating conditions. This invention establishes dynamic control limits based on historical health operation data, enabling the warning thresholds to adapt to the equipment's own operating characteristics, thereby improving the stability and reliability of warning results and reducing false alarm and false negative rates.
[0016] 4. This invention can output an early warning signal when the statistical characteristic parameters exceed the dynamic control limit or meet the continuous rise criterion. It can also be combined with the vibration effective value alarm threshold for hierarchical judgment, thereby realizing a hierarchical prompting method of early warning-fault alarm, which can buy more time for equipment maintenance, repair arrangements and spare parts preparation. Attached Figure Description
[0017] Figure 1 This is a flowchart of the fault early warning method of the present invention. Figure 2 This is a schematic diagram of the signal processing and feature extraction chain of the present invention. Figure 3 This is a schematic diagram illustrating the probability and statistical principle of the high-amplitude interval of the present invention. Figure 4 This is a schematic diagram of the early warning determination logic of the present invention. Figure 5 This is a time-series diagram illustrating the early warning effect of the present invention. Detailed Implementation
[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0019] The invention will now be further described with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating the overall process of the fault early warning method of the present invention. Figure 2 This is a schematic diagram of the signal processing and feature extraction link of the present invention; Figure 3 This is a schematic diagram illustrating the probability and statistics principle of the high-amplitude range of the present invention; Figure 4 This is a schematic diagram of the early warning determination logic of the present invention; Figure 5 This is a time-series diagram illustrating the early warning effect of the present invention.
[0020] In the specific implementation process, such as Figures 1-5 As shown, this invention uses rolling bearings in rotating machinery as the monitoring object, and illustrates a fault early warning method that combines vibration signal envelope demodulation with statistical characteristics. The monitored equipment can be wind turbine bearings, motor bearings, gearbox bearings, etc. Early pitting corrosion faults on the outer ring of a bearing are used as an example for illustration.
[0021] The early warning method consists of a signal acquisition unit, a signal processing unit, a statistical analysis unit, and an early warning output unit. The signal acquisition unit collects vibration signals from the monitored equipment during operation; the signal processing unit performs bandpass filtering, envelope demodulation, and extraction of fault characteristic frequency amplitudes; the statistical analysis unit calculates statistical characteristic parameters within each analysis window; and the early warning output unit outputs an early warning signal based on the judgment results.
[0022] The method of the present invention includes the following steps.
[0023] I. Vibration Signal Acquisition and Envelope Demodulation During the normal operation of the monitored equipment, vibration acceleration signals are continuously collected using vibration sensors installed at the bearing housing to obtain the original vibration signal x(t).
[0024] To highlight the impact response caused by localized bearing damage, the original vibration signal x(t) was first bandpass filtered. The filtering frequency band was centered on the resonant frequency band of the monitored component to suppress low-frequency speed fluctuation components and background vibration components unrelated to the fault, resulting in the bandpass filtered signal x_b(t).
[0025] Then, envelope demodulation is performed on the bandpass-filtered signal x_b(t). Envelope demodulation can be implemented using the Hilbert transform, i.e., by calculating:
[0026] in, Representing the Hilbert transform This is the envelope signal. In other alternative implementations, the envelope signal can also be obtained by rectifying and then low-pass filtering.
[0027] For envelope signal Spectral analysis was performed to obtain the envelope spectrum. E(f) Based on the bearing structural parameters and rotational speed information, the target fault characteristic frequency f_0 is pre-calculated. In this embodiment, f_0 is the bearing outer ring fault characteristic frequency. Subsequently, the characteristic amplitude is extracted within a preset frequency band centered on f_0, serving as the characteristic amplitude corresponding to the current sampling segment. Specifically, within the frequency band... The maximum spectral line amplitude or integral amplitude is taken and denoted as a_k. The above process is repeated for consecutive sampling segments to form an amplitude time series arranged by time. This amplitude time series serves as the input for subsequent statistical analysis.
[0028] II. Analysis Window Division and Statistical Feature Calculation In order to reflect the overall changes in the characteristics of fault impact over a period of time, instead of directly using the amplitude at a single moment for early warning, statistical distribution analysis of the amplitude time series A is performed according to a preset analysis window.
[0029] In this embodiment, the analysis window is divided by day, meaning that all feature amplitude samples acquired within a day constitute one analysis window. Let the i-th analysis window contain... N(i) Each amplitude sample is denoted as: For each analysis window, at least two amplitude ranges are pre-defined: a low amplitude range and a high amplitude range. The high amplitude range is used to characterize anomalous amplitude samples with strong impact.
[0030] Let the lower threshold of the high amplitude range be... T_H The number of data points falling into the high-amplitude range within the i-th analysis window is denoted as . N_H(i) ,satisfy: The probability of a high-amplitude interval within the i-th analysis window is defined as: ,in, P_H(i) This represents the proportion of high-amplitude samples in the i-th analysis window out of all samples.
[0031] In this embodiment, P_H(i) As a statistical characteristic parameter characterizing the change in the intensity of a fault impact. Unlike directly comparing a single peak value, P_H(i) It reflects the statistical frequency of high-amplitude samples, and is therefore more sensitive to weak but persistent early shocks.
[0032] III. Establishment of Thresholds and Dynamic Control Limits for High-Amplitude Ranges To adapt the statistical criteria to individual differences among different devices, a period of historical health data is collected in advance at the initial stage when the device is in a healthy operating state, and historical characteristic amplitude samples are extracted using the same method as described above to establish a health sample set.
[0033] Statistical analysis was performed on the amplitude data corresponding to the healthy sample set to determine the threshold of the high amplitude interval. T_H .
[0034] In this embodiment, the high quantile value of the amplitude distribution of the healthy sample set can be used as the threshold for the high amplitude interval. For example, the 95th quantile value or the 98th quantile value can be selected as the threshold. T_H .
[0035] In obtaining T_H Subsequently, the probabilities of high-amplitude intervals within multiple consecutive analysis windows during the healthy period were statistically analyzed to obtain the probability sequence under the healthy state: Based on this health probability sequence, dynamic control limits are determined. U .
[0036] In this embodiment, the following formula can be used for calculation:
[0037] in, In a healthy state The mean, In a healthy state The standard deviation.
[0038] In other implementations, the healthy state can also be... The 99th percentile value is directly used as the dynamic control limit. U The dynamic control limits obtained using the above method are not fixed constants, but are derived from the equipment's own health history data, thus better adapting to the differences in the inherent vibration levels of different equipment.
[0039] IV. Determining the Trend of Change After completing each analysis window After calculation, the trend of statistical characteristic parameters of multiple consecutive analysis windows is determined.
[0040] In this embodiment, the preset ascent criterion can be one of the following methods: (1) Three consecutive analysis windows satisfy ; Or (2) The slope of the linear fit of P_H in the most recent m analysis windows is greater than the preset slope threshold η; Alternatively, (3) the moving average of the most recent m analysis windows continues to rise, and the cumulative increment exceeds the preset increment threshold.
[0041] For ease of implementation, this embodiment preferably adopts the first method, that is, when three consecutive analysis windows... P_H As the values increase sequentially, it is determined that they satisfy a continuous upward trend. This criterion is simple to calculate and easy to implement online.
[0042] V. Early Warning Judgment and Output After completing the statistics in the i-th analysis window, the following two types of judgments are executed simultaneously.
[0043] 1. Statistical Feature Early Warning Judgment A statistical anomaly is determined to occur when any of the following conditions are met: ; It does not exceed U, but meets the preset rising criterion.
[0044] When any of the above conditions are met, the statistical analysis unit sends an early warning trigger command to the early warning output unit.
[0045] 2. Comparison and judgment with RMS alarm values To reflect the early warning characteristics, the effective value R(i) of the original vibration signal is also calculated within the same analysis window. The calculation formula is as follows:
[0046] in, This represents the original vibration signal sample value within the analysis window. Let the traditional effective value alarm threshold of the equipment be... When the following conditions are met: or The criteria for continuous increase are met. And simultaneously satisfy: If so, an early warning signal will be output. When problems occur during subsequent operation... When this occurs, a traditional fault alarm signal is output.
[0047] In a practical monitoring process, the system calculates the proportion of high-amplitude intervals in the time series of amplitude corresponding to the characteristic frequencies of bearing outer ring faults on a daily basis. P_H .
[0048] During the healthy operation phase of the equipment P_H It fluctuates slightly around 1%.
[0049] Once early pitting occurs on the equipment, starting from day 10, P_H It begins to rise slowly to about 2%, but at this point the effective value of vibration does not show a significant change.
[0050] As the fault progressed, by day 25, P_H If the value continues to rise to approximately 5% and has already met the above dynamic control limit judgment or the continuous rise criterion, the system will issue an early warning signal.
[0051] It wasn't until about the 40th day of operation that the effective vibration value increased significantly and triggered the traditional effective value alarm.
[0052] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.
Claims
1. A fault early warning method combining vibration signal envelope demodulation and statistical characteristics, characterized in that, Includes the following steps: Step 1: Acquire the vibration signal of the monitored equipment under operating conditions, perform envelope demodulation analysis on the vibration signal, and extract the amplitude time series corresponding to the preset fault characteristic frequency; Step 2: Perform statistical distribution analysis on the amplitude time series according to the preset analysis window, divide the amplitude time series into at least two preset amplitude intervals, and calculate the probability value of the number of data points in each preset amplitude interval relative to the total number of data points in the analysis window, so as to obtain the statistical characteristic parameters of the current analysis window; Step 3: Determine the changing trend of the statistical characteristic parameters based on the statistical characteristic parameters from multiple consecutive analysis windows; Step 4: When the probability of the high amplitude interval in the statistical feature parameters exceeds the corresponding dynamic control limit, or when the change trend meets the preset upward criterion, output a device fault warning signal.
2. The fault early warning method combining vibration signal envelope demodulation and statistical characteristics according to claim 1, characterized in that, In step 1, before performing envelope demodulation analysis, the vibration signal is further bandpass filtered with the resonant frequency band of the monitored component as the center. The envelope demodulation analysis employs either Hilbert transform envelope demodulation or rectifier-low-pass filter envelope demodulation.
3. The fault early warning method combining vibration signal envelope demodulation and statistical characteristics according to claim 1, characterized in that, In step 1, the preset fault characteristic frequency is at least one of the following: bearing outer ring fault characteristic frequency, bearing inner ring fault characteristic frequency, rolling element fault characteristic frequency, cage fault characteristic frequency, or gear meshing frequency.
4. The fault early warning method combining vibration signal envelope demodulation and statistical characteristics according to claim 1, characterized in that, In step 2, the statistical distribution analysis includes setting low-amplitude intervals and high-amplitude intervals, and calculating, within each analysis window, the proportion P of data points falling into the high-amplitude interval out of the total number of data points in that analysis window. H , will the P H As a statistical characteristic parameter characterizing the change in the impact intensity of a fault.
5. The fault early warning method combining vibration signal envelope demodulation and statistical characteristics according to claim 1, characterized in that, In step 4, the dynamic control limit is determined based on the statistical results of the corresponding statistical characteristic parameters in the historical health operation data of the equipment, and is updated as new health operation data is introduced.
6. The fault early warning method combining vibration signal envelope demodulation and statistical characteristics according to claim 1, characterized in that, Step 4 also includes calculating the effective value of the vibration signal within the same analysis window; When the statistical characteristic parameter exceeds the dynamic control limit or the change trend meets the preset rising criterion, and the effective value does not reach the preset effective value alarm threshold, an early warning signal is output.