Bearing early fault self-adaptive detection method based on sequential probability ratio test
By adaptively determining the amount of sampled data in each rotation period of the bearing based on sequential probability ratio inspection method, the problem of high leakage detection rate and redundancy of bearing failure detection in the prior art is solved, and efficient and economical bearing failure detection is achieved.
Patent Information
- Application Number
- CN202411556992.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-04
- Publication Date
- 2025-03-28
AI Technical Summary
The existing bearing fault detection methods have problems with high fault miss detection rate and redundant data, and are difficult to apply to general rotating mechanical equipment.
Adaptive detection method of early bearing failure based on sequential probability ratio detection is adopted, and the sampling data amount in each rotation period of the rotating shaft is adaptively determined to realize the adaptive detection of bearing failures.
It reduces the fault miss detection rate, reduces data redundancy and detection costs, and is suitable for bearing failure detection of general rotating mechanical equipment.
Smart Images
Figure CN119714891A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of early fault detection and diagnosis of rotating machinery, and in particular to an adaptive detection method for early fault of bearings based on sequential probability ratio test. Background Art
[0002] Rotating machinery and equipment are important components of industrial equipment. The operating conditions of rotating machinery and equipment directly affect the economic benefits of enterprises. As a key component of rotating machinery, bearings bear important functions such as supporting rotating shafts and reducing friction. Due to the high working intensity of bearings and relatively harsh environment, most of the failures of rotating machinery are caused by bearing failures. Bearing failures not only affect equipment performance, but may also cause serious safety accidents. Therefore, in order to prevent the negative impact caused by bearing failures, rotating machinery and equipment should be equipped with a certain bearing fault detection system to implement bearing fault detection in a timely and accurate manner, identify the early fault status of bearings, and then manage the health of bearings and reduce maintenance costs.
[0003] At present, the existing bearing fault detection methods are mainly divided into periodic detection methods and continuous detection methods. The periodic detection method refers to the implementation of bearing fault detection at a specified time node. Periodic detection can greatly reduce the cost of bearing detection, but the failure rate is high. Bearing failures can occur at any time during its operation, not just at the time of detection. Untimely bearing fault detection may cause excessive maintenance costs or even cause major safety accidents.
[0004] In order to improve the reliability of bearing fault detection and reduce the missed detection rate of faults, some researchers have proposed to implement continuous detection of bearings. At present, the continuous detection method refers to the use of a fixed sampling frequency to continuously and uninterruptedly detect bearings. This type of method is mainly used for bearing detection of key rotating machinery, such as major chemical equipment or power equipment. Although continuous detection can reliably implement fault detection and effectively reduce the missed detection rate of bearing faults, it will lead to data redundancy and high detection costs, and is not suitable for bearing fault detection of general rotating machinery equipment. Summary of the invention
[0005] Purpose of the invention: In order to overcome the shortcomings of the background technology, the present invention requests to disclose a method for adaptive detection of early bearing faults based on sequential probability ratio test. The method adaptively determines the amount of sampling data within one rotation period of the shaft according to the health status of the bearing, thereby realizing adaptive detection of bearing faults. While reducing the missed fault detection rate, data redundancy and detection costs can be reduced.
[0006] Technical solution: The method for adaptively detecting early bearing faults based on sequential probability ratio test disclosed in the present invention comprises the following steps:
[0007] S1. Determine prior knowledge;
[0008] Including: the installation position of the vibration sensor and the speed sensor, and the sticking position of the sampling reflective strip that matches the speed sensor; the minimum amount N1, maximum amount N2 and initial amount N0 of the periodic synchronous sampling data within each rotation of the shaft, the number of rotations k of the shaft sampling within the time window, the probability density function f (std) and g (std) of the fault characteristics when the bearing is fault-free and faulty, and the threshold T1 and threshold T2;
[0009] S2, obtaining periodic synchronous sampling data X of the bearing;
[0010] S3, constructing N groups of periodic synchronous single-point sampling data based on the periodic synchronous sampling data amount N per revolution;
[0011] S4, calculating the standard deviation Std of N groups of periodic synchronous single-point sampling data, and using it as the fault feature;
[0012] S5, calculating the log-likelihood ratio L of the standard deviation of N groups of periodically synchronized single-point sampling data;
[0013] S6, accumulating the log-likelihood ratio of the standard deviation of N-period synchronous single-point sampling data to obtain the cumulative sum S, and using it as a fault indicator;
[0014] S7. Determine whether the bearing is faulty based on the threshold value, and adaptively determine the amount of synchronous sampling data N per rotation of the shaft for fault detection in the next time window.
[0015] Among them, in S1:
[0016] The vibration sensor is used to obtain the bearing vibration signal and is installed near the bearing to be tested; the speed sensor is used to measure the shaft speed signal and is installed near the shaft, and the sampling reflective strip used in conjunction with the speed sensor is pasted on the shaft;
[0017] The minimum amount N1 and maximum amount N2 of synchronous sampling data within each rotation of the shaft are determined according to the bearing fault form. The selection range of the minimum amount N1 is [1,10], and the selection range of the maximum amount N2 is [50,100];
[0018] The initial quantity N0 is determined according to the health status of the bearing being tested. If the bearing being tested is a brand new bearing, the initial quantity N0 is set to the minimum quantity N1; if the bearing being tested is close to failure, the initial quantity N0 is set to the maximum quantity N2; if the health status of the bearing being tested is unknown, the initial quantity N0 is set to Among them, the function round() is a rounding function;
[0019] The sampling shaft speed k in the time window determines the time window length of fault detection. The probability density functions f(std) and g(std) of the fault characteristics when the bearing is fault-free and faulty are obtained based on statistical data.
[0020] The threshold T1 and the threshold T2 are determined according to the first risk level α and the second risk level β. The first risk level α and the second risk level β determine the reliability of bearing fault detection. In the process of bearing fault detection, there are two detection risks. The first risk is that the bearing is faulty but a fault is detected. The probability of this risk is called the first risk level α. The second risk is that the bearing is faulty but a fault is not detected. The probability of this risk is called the second risk level β. The two thresholds T1 and T2 are shown in the following formulas:
[0021]
[0022] Furthermore, the amount of periodic synchronous sampling data per rotation of the rotating shaft in S2 is N, the number of rotations of the sampling rotating shaft is k, the length of the periodic synchronous sampling data X is kN, and the periodic synchronous sampling data X is expressed as [x1, x2, x3, ..., x kN ];
[0023] In S3, the periodic synchronous sampling data X obtained in S2 is constructed into N groups of periodic synchronous single-point sampling data [X1, X2, X3, ..., X i ,…,X N ], the i-th group of periodic synchronous single-point sampling data X i (i=1,2,3,…,N) is expressed as [x i ,x (i+N) ,x (i+2N) ,...,x (i+(k-1)N) ].
[0024] Furthermore, in S4, the standard deviation Std = [Std1, Std2, Std3, ..., Std i ,…,Std N ] and use it as a fault feature;
[0025] For the i-th group of periodic synchronous single-point sampling data X i , its standard deviation is
[0026]
[0027] Among them, μ i Represents the i-th group of periodic synchronous single-point sampling data X i The mean of .
[0028] Furthermore, in S5:
[0029] According to the probability density distribution functions f(std) and g(std) of the pre-set fault characteristics when the bearing is fault-free and faulty, the logarithmic likelihood ratios of the standard deviations of N groups of periodic synchronous single-point sampling data are calculated respectively. i ,L,L N ],in
[0030]
[0031] Furthermore, in S6:
[0032]
[0033] Based on the sequential probability ratio test theory, the log-likelihood ratio of the standard deviation of N-period synchronous single-point sampling data is accumulated to obtain the cumulative sum S, which is used as the fault indicator.
[0034] Furthermore, in S7:
[0035] When the cumulative sum S>T2, it indicates that the bearing has an early fault, and the bearing early fault detection process is stopped;
[0036] When the cumulative sum S < T1, it indicates that the bearing is in a healthy state, then the synchronous sampling data amount N of each rotation of the shaft for the next time window fault detection is adjusted to the minimum value N1, and the process goes to S2 to continue the early bearing fault detection process;
[0037] When the cumulative sum T1≤S≤T2, it means that it is temporarily impossible to determine whether the bearing is faulty. At this time, the synchronous sampling data volume N per rotation of the shaft for the next time window fault detection is adjusted to Among them, the function ceil() is a rounding up function, and goes to S2 to continue the bearing fault detection process.
[0038] Beneficial effects: Compared with the prior art, the advantages of the present invention are: the fault information of N groups of periodic synchronous single-point sampling data can be accumulated to determine the bearing degradation state, and then determine whether an early fault occurs according to the bearing degradation state and adaptively adjust the periodic synchronous sampling data volume N per rotation of the shaft. While ensuring the reliability of bearing detection, the amount of data collection is significantly reduced, thereby reducing data transmission costs, processing costs and storage costs. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 is a flow chart of the method of the present invention;
[0040] Figure 2 There is no fault signal for the bearing;
[0041] Figure 3 It is the bearing fault signal;
[0042] Figure 4 This is the test result when the bearing is fault-free;
[0043] Figure 5 It is the amount of synchronous sampling data within each rotation of the shaft when the bearing is fault-free;
[0044] Figure 6 The detection result of the bearing failure;
[0045] Figure 7 It is the amount of synchronous sampling data within each rotation of the shaft when the bearing fails. DETAILED DESCRIPTION
[0046] An adaptive detection method for early bearing faults based on sequential probability ratio test, such as Figure 1 As shown, the specific steps include:
[0047] Step 1: Determine prior knowledge;
[0048] Prior knowledge includes: the installation positions of the vibration sensor and the speed sensor, and the sticking positions of the sampling reflective strips that match the speed sensor; the minimum amount N1, maximum amount N2 and initial amount N0 of the periodic synchronous sampling data within each rotation of the shaft, the sampling number k of the shaft within the time window, the probability density functions f (std) and g (std) of the fault characteristics when the bearing is fault-free and faulty, and two thresholds T1 and T2.
[0049] The vibration sensor is used to obtain the bearing vibration signal and needs to be installed near the bearing to be tested. The speed sensor is used to measure the shaft speed signal and needs to be installed near the shaft. The reflective strip used in conjunction with the speed sensor needs to be pasted on the shaft. The minimum amount N1 and maximum amount N2 of synchronous sampling data within each rotation of the shaft are determined according to the form of bearing failure. The recommended selection range for the minimum amount N1 is [1,10], and the selection range for the maximum amount N2 is [50,100]. The initial amount N0 is determined according to the health status of the bearing to be tested. If the bearing to be tested is a brand new bearing, the initial amount N0 can be set to the minimum amount N1; if the bearing to be tested is a bearing close to failure, the initial amount N0 can be set to the maximum amount N2; if the health status of the bearing to be tested is unknown, Among them, the function round() is a rounding function.
[0050] The sampling shaft speed k in the time window determines the time window length of fault detection. The probability density functions f(std) and g(std) of the fault characteristics when the bearing is fault-free and faulty need to be obtained based on statistical data.
[0051] The two thresholds T1 and T2 are determined according to the first risk level α and the second risk level β. The first risk level α and the second risk level β determine the reliability of bearing fault detection. In the process of bearing fault detection, there are two detection risks. The first risk is that a fault is detected when the bearing is not faulty. The probability of this risk is called the first risk level α (or the false alarm rate). The second risk is that a fault is not detected when the bearing has already failed. The probability of this risk is called the second risk level β (or the missed alarm rate). The two thresholds T1 and T2 are shown in the following formula:
[0052]
[0053] Step 2: Obtain the periodic synchronous sampling data X of the bearing
[0054] Use the speed sensor and vibration sensor to obtain the periodic synchronous sampling data X of the bearing. The amount of periodic synchronous sampling data per rotation of the shaft is N, and the number of sampling rotations is k. Therefore, the length of the periodic synchronous sampling data X is kN. The periodic synchronous sampling data X can be expressed as [x1, x2, x3, ..., x kN ].
[0055] Step 3: Construct N groups of periodically synchronized single-point sampling data.
[0056] The periodic synchronous single-point sampling data refers to the data obtained by synchronously collecting a single data within one rotation period of the shaft. During the sampling process, N data are synchronously collected within each rotation period of the shaft. Therefore, the periodic synchronous sampling data X obtained in step 2 can be constructed into N groups of periodic synchronous single-point sampling data [X1, X2, X3, …, X i ,…,X N ]. The i-th group of periodic synchronous single-point sampling data X i (i=1,2,3,…,N) can be expressed as [x i ,x (i+N) ,x (i+2N) ,...,x (i+(k-1)N) ].
[0057] Step 4: Calculate the standard deviation Std = [Std1, Std2, Std3, ..., Std i ,…,Std N ] and use it as a fault feature.
[0058] For the i-th group of periodic synchronous single-point sampling data X i , its standard deviation is
[0059]
[0060] Among them, μ iRepresents the i-th group of periodic synchronous single-point sampling data X i The mean of .
[0061] Step 5: Calculate the log-likelihood ratio of the standard deviation of N groups of periodic synchronous single-point sampling data L = [L1, L2, L3, L, L i ,L,L N ];
[0062] According to the probability density distribution functions f(std) and g(std) of the pre-set fault characteristics when the bearing is fault-free and faulty, the logarithmic likelihood ratios of the standard deviations of N groups of periodic synchronous single-point sampling data are calculated respectively. i ,L,L N ],in
[0063]
[0064] Step 6: Accumulate the log-likelihood ratio of the standard deviation of the synchronous single-point sampling data of N periods to obtain the cumulative sum S;
[0065]
[0066] Based on the sequential probability ratio test theory, the log-likelihood ratio of the standard deviation of N-period synchronous single-point sampling data is accumulated to obtain the cumulative sum S, which is used as the fault indicator.
[0067] Step 7: Based on the cumulative sum S calculated in step 6 and the two thresholds T1 and T2 pre-set in step 1, determine whether the bearing is faulty, and adaptively determine the amount of synchronous sampling data N per rotation of the shaft for fault detection in the next time window.
[0068] When the cumulative sum S>T2, it indicates that the bearing has an early fault, and the bearing early fault detection process is stopped;
[0069] When the cumulative sum S is less than T1, it indicates that the bearing is in a healthy state, and the amount of synchronous sampling data N per rotation of the shaft for the next time window fault detection is adjusted to the minimum value N1, and go to step 2 to continue the bearing early fault detection process.
[0070] When the cumulative sum T1≤S≤T2, it means that it is temporarily impossible to determine whether the bearing is faulty. At this time, the synchronous sampling data volume N of each rotation of the shaft for the next time window fault detection is adjusted to Among them, the function ceil() is a rounding up function, and goes to step 2 to continue the bearing fault detection process.
[0071] According to existing literature, the vibration signals of the bearing without fault and fault are simulated, such as Figure 2 and 3As shown, the signal is used to verify the feasibility of the present invention.
[0072] In this embodiment, the vibration sensor selected is an acceleration sensor, and the initial amount N0=1, the minimum amount N1=1 and the maximum amount N2=100 of the periodic synchronous sampling data of the rotating machinery are set, and the number of rotations of the sampling shaft in the time window is k=10. Two thresholds T1 and T2 are determined according to the first risk level α and the second risk level β. In this embodiment, the first risk level α and the second risk level β are both 1%, so the two thresholds T1=-4.6 and T2=4.6 are determined. The initial amount N0=1 indicates that the measured bearing is a brand new bearing.
[0073] According to the initial amount N0=1 of synchronous sampling data within each rotation of the shaft, the acceleration synchronization signal X is synchronously collected using the laser speed sensor and the acceleration sensor. The collected synchronous vibration signal X is regarded as a one-dimensional vector X i (i=1,2,3,…,10). Calculate the standard deviation of the one-dimensional vector and the corresponding log-likelihood ratio, and then accumulate the log-likelihood ratio into a cumulative sum S. Since only one set of periodic synchronous sampling signals is obtained in the time window, a log-likelihood ratio is obtained. At this time, the cumulative sum S is equal to the log-likelihood ratio.
[0074] The cumulative sum S is used as a fault indicator and compared with two pre-set thresholds. When the cumulative sum S>T2, it indicates that the bearing has an early fault, and the bearing fault detection process is stopped; when the cumulative sum S<T1, it indicates that the bearing is in a healthy state, and the amount of synchronous sampling data N per cycle of the shaft for the next fault detection is adjusted to the minimum value N1, and go to step 2 to continue the bearing early fault detection process. When the cumulative sum T1≤S≤T2, it means that it is temporarily impossible to determine whether the bearing has an early fault. At this time, the amount of synchronous sampling data N per cycle of the shaft for the next time window is adjusted to Among them, the function ceil() is a rounding up function, and goes to step 2 to continue the bearing fault detection process.
[0075] Figure 4 and Figure 5 The detection results when the bearing is fault-free and the amount of periodic synchronous sampling data per rotation of the shaft are shown respectively. In 9 fault detections, all cumulative sums S are less than the threshold value T2, indicating that no bearing fault is detected. In all the above detections, the amount of periodic synchronous sampling data N per rotation of the shaft for each fault detection is less than the preset maximum value of 100, and there are four detections where the amount of periodic synchronous sampling data N is 1. The method of the present invention reduces the amount of periodic synchronous sampling data per rotation of the shaft when the bearing is fault-free, thereby reducing the cost of data transmission, processing and storage.
[0076] Figure 6 and Figure 7The detection results of bearing failure and the amount of synchronous sampling data per rotation of the shaft are shown respectively. In the 9 bearing fault detections, although the cumulative sum S of some time windows is less than the threshold value T2, there are two detections where the cumulative sum S is greater than the threshold value T2, indicating that the bearing has failed. Due to the influence of noise in the signal, the cumulative sum S of some time windows is less than the threshold value T2. Under the influence of noise, the method of the present invention can still adaptively detect bearing faults. The amount of synchronous sampling data N per rotation of the shaft for the two fault detections is a maximum value of 100, which increases the amount of data collected per rotation of the shaft, improves the ability to obtain fault information, and thus improves the reliability of fault detection.
Claims
1. A method for adaptive detection of early bearing faults based on sequential probability ratio test, characterized in that: The following steps are involved: S1. Determine prior knowledge; Including: the installation position of the vibration sensor and the speed sensor, and the sticking position of the sampling reflective strip that matches the speed sensor; the minimum amount N1, maximum amount N2 and initial amount N0 of the periodic synchronous sampling data within each rotation of the shaft, the number of rotations k of the shaft sampling within the time window, the probability density function f (std) and g (std) of the fault characteristics when the bearing is fault-free and faulty, and the threshold T1 and threshold T2; S2, obtaining periodic synchronous sampling data X of the bearing; S3, constructing N groups of periodic synchronous single-point sampling data based on the periodic synchronous sampling data amount N within each rotation of the shaft; S4, calculating the standard deviation Std of N groups of periodic synchronous single-point sampling data, and using it as the fault feature; S5, calculating the log-likelihood ratio L of the standard deviation of N groups of periodically synchronized single-point sampling data; S6, accumulating the log-likelihood ratio of the standard deviation of N-period synchronous single-point sampling data to obtain the cumulative sum S, and using it as a fault indicator; S7. Determine whether the bearing is faulty based on the threshold value, and adaptively determine the amount of synchronous sampling data N per rotation of the shaft for fault detection in the next time window.
2. The method for adaptive detection of early bearing faults based on sequential probability ratio test according to claim 1 is characterized in that: In S1: The vibration sensor is used to obtain the bearing vibration signal and is installed near the bearing to be tested; the speed sensor is used to measure the shaft speed signal and is installed near the shaft, and the sampling reflective strip used in conjunction with the speed sensor is pasted on the shaft; The minimum amount N1 and maximum amount N2 of synchronous sampling data within each rotation of the shaft are determined according to the bearing fault form. The selection range of the minimum amount N1 is [1,10], and the selection range of the maximum amount N2 is [50,100]; The initial quantity N0 is determined according to the health status of the bearing being tested. If the bearing being tested is a brand new bearing, the initial quantity N0 is set to the minimum quantity N1; if the bearing being tested is close to failure, the initial quantity N0 is set to the maximum quantity N2; if the health status of the bearing being tested is unknown, the initial quantity N0 is set to Among them, the function round() is a rounding function; The sampling shaft speed k in the time window determines the time window length of fault detection. The probability density functions f(std) and g(std) of the fault characteristics when the bearing is fault-free and faulty are obtained based on statistical data. The threshold T1 and the threshold T2 are determined according to the first risk level α and the second risk level β. The first risk level α and the second risk level β determine the reliability of bearing fault detection. In the process of bearing fault detection, there are two detection risks. The first risk is that the bearing is faulty but a fault is detected. The probability of this risk is called the first risk level α. The second risk is that the bearing is faulty but a fault is not detected. The probability of this risk is called the second risk level β. The two thresholds T1 and T2 are shown in the following formulas:
3. The method for adaptive detection of early bearing faults based on sequential probability ratio test according to claim 1 is characterized in that: The amount of periodic synchronous sampling data per rotation of the rotating shaft in S2 is N, the number of rotations of the sampling rotating shaft is k, the length of the periodic synchronous sampling data X is kN, and the periodic synchronous sampling data X is expressed as [x1,x2,x3,...,x kN ]; In S3, the periodic synchronous sampling data X obtained in S2 is constructed into N groups of periodic synchronous single-point sampling data [X1, X2, X3, ..., X i ,…,X N ], the i-th group of periodic synchronous single-point sampling data X i (i=1,2,3,…,N) is expressed as [x i ,x (i+N) ,x (i+2N) ,...,x (i+(k-1)N) ].
4. The method for adaptive detection of early bearing faults based on sequential probability ratio test according to claim 1 is characterized in that: In S4, the standard deviation Std=[Std1, Std2, Std3,…, Std i ,…,Std N ] and use it as a fault feature; For the i-th group of periodic synchronous single-point sampling data X i , its standard deviation can be obtained as Among them, μ i Represents the i-th group of periodic synchronous single-point sampling data X i The mean of .
5. The method for adaptive detection of early bearing faults based on sequential probability ratio test according to claim 1 is characterized in that: In S5: According to the probability density distribution functions f(std) and g(std) of the pre-set fault characteristics when the bearing is fault-free and faulty, the logarithmic likelihood ratios of the standard deviations of N groups of periodic synchronous single-point sampling data are calculated respectively. i ,L,L N ],in 6. The method for adaptive detection of early bearing faults based on sequential probability ratio test according to claim 1 is characterized in that: In S6: Based on the sequential probability ratio test theory, the log-likelihood ratio L of the standard deviation of N-period synchronous single-point sampling data is accumulated to obtain the cumulative sum S, which is used as the fault indicator.
7. The method for adaptive detection of early bearing faults based on sequential probability ratio test according to claim 1 is characterized in that: In S7: When the cumulative sum S>T2, it indicates that the bearing has an early fault, and the bearing early fault detection process is stopped; When the cumulative sum S < T1, it indicates that the bearing is in a healthy state, then the synchronous sampling data amount N of each rotation of the shaft for the next time window fault detection is adjusted to the minimum value N1, and the process goes to S2 to continue the early bearing fault detection process; When the cumulative sum T1≤S≤T2, it means that it is temporarily impossible to determine whether the bearing is faulty. At this time, the synchronous sampling data volume N per rotation of the shaft for the next time window fault detection is adjusted to Among them, the function ceil() is a rounding up function, and goes to S2 to continue the bearing fault detection process.