Multi-bearing fault positioning diagnosis method

Abstract

The invention relates to the technical field of bearing fault diagnosis, in particular to a multi-bearing fault positioning and diagnosing method, which comprises the following steps of: acquiring bearing vibration signals of a plurality of sensing points, extracting peak time information to complete synchronous calibration, extracting high-frequency band energy and constructing a feature vector, screening deviation features and clustering and grouping, and identifying an energy abnormal channel, positioning a bearing position, extracting time domain and frequency domain features, comparing fault modes, and outputting a fault type diagnosis result. According to the method, the consistency of multi-channel data synchronization is realized by extracting the peak point of the vibration signal and calibrating the time offset, the high-frequency energy change rate is utilized to match the weight coefficient, the detection sensitivity of tiny fault features is enhanced, and the separability of a composite fault state is improved by combining feature relative deviation rate screening and density trend clustering. A frequency section energy proportion change recognition channel is adopted, accurate positioning of a fault bearing is achieved, and accurate judgment of a fault type is achieved in combination with a multi-feature comparison mode.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bearing fault diagnosis, in particular to a multi-bearing fault positioning diagnosis method. BACKGROUND

[0002] The technical field of bearing fault diagnosis contains the technical system of detecting and evaluating the state of bearings in various rotating mechanical equipment. This field mainly focuses on identifying wear, cracks, fatigue, and various failure modes of bearings during operation through vibration signal analysis, acoustic emission detection, and temperature monitoring. The core content includes feature extraction based on bearing dynamic response signals, establishment of diagnostic models to determine bearing operating states, identification of fault types and locations, and application in various industrial equipment, aerospace, railway locomotives, and new energy vehicles. With the increasing demand for high-speed and high-load operation of electric vehicle power systems, the health detection of key components such as motor bearings and hub bearings has become an important part of ensuring vehicle reliability.

[0003] Among them, a multi-bearing fault positioning diagnosis method refers to a method for synchronous fault detection and positioning analysis of multiple key bearings. The patent subject mainly covers synchronous acquisition of bearing vibration signals, extraction of frequency band energy features from each bearing vibration signal, realization of multi-bearing fault feature data standardization through feature vector construction and normalization processing, identification and positioning of different bearing fault modes using a classification model, and determination of specific fault bearings based on actual bearing installation location information. The method is based on vibration sensor arrangement and synchronous data acquisition module, combined with the integrated application of wavelet packet energy analysis and support vector machine classification model, to complete the fault positioning diagnosis process in a multi-bearing system.

[0004] Traditional multi-bearing fault positioning diagnosis technology adopts a unified collection of vibration signals and directly extracts overall envelope or frequency domain features. Since the sampling starting points between channels are not fine-calibrated, there is a shift on the time axis, causing misalignment of high-frequency transient responses and affecting the accurate determination of subsequent frequency energy distribution. The total energy within a fixed frequency band is calculated, ignoring the energy change rate and its uneven distribution over time, resulting in weak sensitivity to local transient fault responses. In the feature selection stage, the full-quantity feature input classification model is used, which lacks evaluation of feature value dispersion distribution, leading to model clustering result dimension redundancy and boundary ambiguity, making it difficult to accurately divide the state boundaries of multiple bearings. In fault type identification, the model relies on experience classifiers, lacks logical attribution judgment of kurtosis, skewness, and main frequency combination distribution, resulting in limited type differentiation accuracy. When hub bearings and motor bearings have faults at the same time, fault feature overlap affects the accuracy of judgment, leading to fault misdiagnosis, misjudgment, and increased false alarm rate and missed alarm rate of the diagnosis system. SUMMARY

[0005] To solve the technical problems existing in the prior art, the embodiment of the present application provides a multi-bearing fault positioning diagnosis method.

[0006] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme, a multi-bearing fault positioning diagnosis method, comprising the following steps:

[0007] S1: Obtain the bearing vibration signal sequence collected by multiple sensing points of the vehicle body, extract the peak points and corresponding time stamps in the continuous period of each sequence, call the peak time sequence of each channel, identify the time offset between adjacent points and calibrate, and generate a synchronous time sequence;

[0008] S2: Based on the synchronous time sequence, extract the energy value of each high-frequency band, analyze the energy change rate of each high-frequency band in multiple time windows, detect the abnormal change band and match the weight coefficient according to the change amplitude, and construct a high-frequency energy feature vector;

[0009] S3: Based on the high-frequency energy feature vector, extract the statistical feature index of each feature item under multiple fault conditions, calculate the relative dispersion rate, screen the feature items whose relative dispersion rate exceeds the screening threshold, and according to the change trend of feature density in sample distribution, divide the clustering boundary, and generate a clustering grouping result;

[0010] S4: Based on the clustering grouping result, divide the frequency section according to the fixed frequency width and calculate the energy proportion, extract the energy proportion change amplitude peak section and identify the corresponding channel, obtain the abnormal response channel, combine the sensor point position, identify the abnormal response area and locate the bearing position, and generate a fault positioning result.

[0011] As a further scheme of the present application, the synchronous time sequence is specifically a peak time alignment list, a time offset correction vector, and a channel time reference unified identifier, the high-frequency energy feature vector is specifically a local high-frequency energy distribution data, a high-frequency energy weight coefficient sequence, and a high-frequency abnormal frequency band identification mark, and the clustering grouping result includes a feature dispersion screening subset, a feature density change boundary, and a fault sample preliminary classification cluster, and the fault positioning result specifically refers to an abnormal energy peak frequency band, a corresponding response measurement point number, and bearing position information.

[0012] As a further scheme of the present application, the step of obtaining the bearing vibration signal sequence collected by multiple sensing points of the vehicle body, extracting the peak points and corresponding time stamps in each sequence, calling the peak time sequence of each channel, identifying the time offset between adjacent points and calibrating, and generating a synchronous time sequence is specifically:

[0013] S101: Obtain bearing vibration signal sequences collected by multiple sensing points of a vehicle body, extract peak points and corresponding time stamps in a continuous period in each signal sequence, establish a peak time mapping table of each channel, and generate a peak time mapping dataset;

[0014] S102: Based on the peak time mapping dataset, identify time interval differences of adjacent peak points in each channel peak time sequence, calculate time offset values between each pair of adjacent peak points, and obtain channel time offset information;

[0015] S103: According to the channel time offset information, adjust the step of the channel peak time sequence with offset, calibrate the starting point and interval length of each channel time axis, and generate a synchronous time sequence.

[0016] As a further scheme of the present application, based on the synchronous time sequence, the energy value of each high-frequency band is extracted, the energy change rate of each high-frequency band in multiple time windows is analyzed, the abnormal change band is detected, and the weight coefficient is matched according to the change amplitude, and the steps of constructing the high-frequency energy feature vector are as follows:

[0017] S201: Based on the synchronous time sequence, perform frequency domain division processing on each vibration signal channel, call the signal response strength in the high-frequency band range, extract the instantaneous energy value of each high-frequency band in the continuous sampling period, establish the corresponding set of frequency band number and energy value, and generate a frequency band energy extraction data table;

[0018] S202: According to the frequency band energy extraction data table, compare the energy values of each frequency band in multiple time windows, extract the energy change amplitude value sequence in adjacent windows, calculate the deviation of the energy change amplitude value and the energy change reference value, identify the frequency band set whose deviation value exceeds the energy change threshold, and obtain the high-frequency abnormal frequency band set;

[0019] S203: Call the high-frequency abnormal frequency band set, extract the energy change amplitude value of each frequency band, match the weight coefficient with the change strength range corresponding to each frequency band, perform data fusion on the energy value of each frequency band, and generate a high-frequency energy feature vector.

[0020] As a further scheme of the present application, based on the high-frequency energy feature vector, the statistical feature index of each feature item under multiple fault conditions is extracted, the relative dispersion rate is calculated, the feature items whose relative dispersion rate exceeds the screening threshold are screened, and the clustering boundary is divided according to the change trend of the feature density in the sample distribution, and the steps of generating a clustering grouping result are as follows:

[0021] S301: Based on the high-frequency energy feature vector, the energy value sequence of each feature item in each channel under multiple working conditions is extracted, the statistical feature indexes of each feature item under multiple fault working conditions are obtained, including mean value, fluctuation value and dispersion degree, and feature item statistical index information is generated;

[0022] S302: The fluctuation value and dispersion degree value of each feature item under each working condition are extracted by calling the feature item statistical index information, the relative deviation rate is calculated, and compared with the set relative deviation rate screening threshold, the feature item set whose deviation proportion exceeds the threshold is screened, and the deviation rate feature screening result is obtained;

[0023] S303: According to the deviation rate feature screening result, the distribution frequency of each feature item in the sample feature space is extracted, the feature item density change is identified, the clustering boundary is divided, and the clustering grouping result is generated.

[0024] As a further scheme of the application, the specific formula for identifying the feature item density change is:

[0025]

[0026] The feature item density change index is calculated;

[0027] Wherein, D i represents the density change index of the i-th feature item, f ij represents the distribution frequency value of the i-th feature item in the j-th sample, represents the distribution frequency mean value of the i-th feature item in all samples, N represents the total number of samples, d ik represents the density value of the i-th feature item in the k-th local neighborhood, represents the average density value of the i-th feature item in all local neighborhoods, M represents the number of local neighborhoods, i represents the number index of the feature item, j represents the sample number index in the global sample sequence, and k represents the sample number index in the local neighborhood range.

[0028] As a further scheme of the application, based on the clustering grouping result, the frequency section is divided according to the fixed frequency width, the energy proportion change amplitude peak section is extracted and the corresponding channel is identified, the abnormal response channel is obtained, the abnormal response area is identified and the bearing position is located in combination with the sensor point position, and the fault positioning result is generated. The steps are as follows:

[0029] S401: Based on the clustering grouping result, the vibration energy value of each channel in the full frequency range is extracted, the frequency axis is divided into multiple continuous frequency sections according to the set fixed frequency width parameter, the energy cumulative value of each channel in each frequency section is obtained, the energy proportion of each frequency section is calculated, and the energy proportion data is generated;

[0030] S402: Call the energy proportion data, identify the frequency section of the energy proportion change amplitude peak in each channel, extract the corresponding channel number, and obtain an abnormal response channel set;

[0031] S403: According to the abnormal response channel set, match the sensor arrangement coordinates corresponding to each channel, identify the signal overlap area boundary of the sensor concentrated area, analyze the energy distribution center point and locate the bearing position, and generate a fault positioning result.

[0032] As a further scheme of the present application, the method further comprises:

[0033] S5: Based on the fault positioning result, call the synchronous vibration signal data of the abnormal response channel, extract the impact amplitude feature, vibration kurtosis feature, vibration skewness feature and frequency main peak feature of the channel signal in the time domain, and compare them with the data features of a plurality of known fault modes, identify the fault type of the target bearing, and output a fault type diagnosis result;

[0034] The fault type diagnosis result is specifically a channel feature extraction result, a feature comparison result, and a fault mode attribution type.

[0035] As a further scheme of the present application, based on the fault positioning result, the synchronous vibration signal data of the abnormal response channel is called, the impact amplitude feature, vibration kurtosis feature, vibration skewness feature and frequency main peak feature of the channel signal in the time domain are extracted, and they are compared with the data features of a plurality of known fault modes, the fault type of the target bearing is identified, and a fault type diagnosis result is output. The step specifically comprises:

[0036] S501: Based on the fault positioning result, the synchronous vibration signal data corresponding to the abnormal response channel is called, the instantaneous amplitude sequence of each signal in the continuous sampling interval in the time domain is extracted, the impact amplitude feature value, vibration kurtosis feature value and vibration skewness feature value are obtained, and a time domain feature index set is generated;

[0037] S502: Call the time domain feature index set, obtain the frequency distribution of each channel signal in a fixed time window, extract the peak amplitude corresponding to the frequency value in the frequency amplitude sequence, and record the main peak frequency position and amplitude change range of each channel. The correspondence between the channel and the main frequency feature is established, and the frequency main peak feature information is obtained;

[0038] S503: According to the frequency main peak feature information, by comparing with the data features of a plurality of known fault modes, the fault type of the target bearing is identified, and a fault type diagnosis result is output.

[0039] As a further scheme of the present application, the specific formula for comparison with the data features of a plurality of known fault modes is:

[0040]

[0041] calculating a matching similarity index of the fault;

[0042] S m representing a matching similarity index of the mth fault mode, F mn representing a frequency feature value of the to-be-identified channel on the nth feature item, F' mn representing a reference frequency feature value of the mth fault mode on the nth feature item, P representing the number of feature items, and m representing the fault mode number and n representing the feature item number.

[0043] The technical scheme provided by the embodiment of the present application has at least the following beneficial effects:

[0044] By extracting the peak points of the vibration signals and calibrating the time offset, the consistency of the multi-channel data synchronization is realized, the detection sensitivity of the micro-fault features is enhanced by using the high-frequency energy change rate matching weight coefficient, the separability of the compound fault state is improved by combining the feature relative dispersion rate screening and the density trend clustering, the fault bearing is accurately positioned by using the frequency section energy proportion change identification channel, and the precise identification of the fault type is realized by combining the multi-feature comparison mode. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical scheme in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0046] Figure 1 The workflow of the present application is shown in the figure. DETAILED DESCRIPTION

[0047] The technical scheme in the present application will be described below with reference to the drawings.

[0048] In the embodiments of the present application, the words such as "example", "for example" and the like are used to represent as an example, illustration or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0049] In the embodiments of the present application, "image" and "picture" can be used interchangeably, and it should be pointed out that their meanings are consistent when the distinction is not emphasized.

[0050] In the embodiments of the present application, sometimes the subscript such as W1 can be written in the form of non-subscript such as W1, and the meanings are consistent when the distinction is not emphasized.

[0051] In order to make the technical problems, technical solutions and advantages of the present application clearer, specific embodiments will be described in detail below with reference to the drawings.

[0052] Please refer to Figure 1 The present application provides a technical solution, a multi-bearing fault positioning and diagnosis method, comprising the following steps:

[0053] S1: Obtain the bearing vibration signal sequence collected by multiple sensing points of the vehicle body, extract the peak points and corresponding time stamps in the continuous period of each sequence, call the peak time sequence of each channel, identify the time offset between adjacent points and calibrate, and generate a synchronous time sequence;

[0054] S2: Based on the synchronous time sequence, extract the energy value of each high-frequency band, analyze the energy change rate of each high-frequency band in multiple time windows, detect the abnormal change band, and match the weight coefficient according to the change amplitude, and construct a high-frequency energy feature vector;

[0055] S3: Based on the high-frequency energy feature vector, extract the statistical feature index of each feature item under multiple fault conditions, calculate the relative dispersion rate, select the feature items whose relative dispersion rate exceeds the screening threshold, and according to the change trend of the feature density in the sample distribution, divide the clustering boundary, and generate a clustering grouping result;

[0056] S4: Based on the clustering grouping result, divide the frequency section according to the fixed frequency width and calculate the energy proportion, extract the energy proportion change amplitude peak section and identify the corresponding channel, obtain the abnormal response channel, combine the sensor point position, identify the abnormal response area and locate the bearing position, and generate a fault positioning result;

[0057] S5: Based on the fault positioning result, call the synchronous vibration signal data of the abnormal response channel, extract the impact amplitude feature, vibration kurtosis feature, vibration skewness feature and frequency main peak feature of the channel signal in the time domain, and compare with the data features of the known multiple fault modes, identify the fault type of the target bearing, and output the fault type diagnosis result.

[0058] The synchronization time sequence is specifically a peak time alignment list, a time offset correction vector, and a unified identification of channel time reference. The high-frequency energy feature vector is specifically local high-frequency energy distribution data, a high-frequency energy weight coefficient sequence, and a high-frequency abnormal frequency band identification mark. The clustering grouping result includes a feature dispersion screening subset, a feature density change boundary, and a preliminary classification cluster of fault samples. The fault positioning result is specifically an abnormal energy peak frequency band, a corresponding response measurement point number, and bearing position information. The fault type diagnosis result is specifically a channel feature extraction result, a feature comparison result, and a fault mode attribution type.

[0059] The bearing vibration signal sequences collected by the multiple sensing points of the vehicle body are obtained. The peak points and corresponding time stamps in the continuous periods of each sequence are extracted. Each channel peak time sequence is called. The time offset between adjacent points is identified and calibrated. The steps of generating the synchronization time sequence are specifically as follows:

[0060] S101: Obtain the bearing vibration signal sequences collected by the multiple sensing points of the vehicle body. Extract the peak points and corresponding time stamps in the continuous periods of each signal sequence. Establish the peak time mapping table of each channel. Generate the peak time mapping dataset.

[0061] The multiple acceleration sensors are fixedly installed at typical structural parts. The vibration response data of multiple channels during the operation of the bearing are collected. The vibration signals of each channel are periodically divided. The number of sampling points in each period is set to 1024 points. The maximum amplitude point and its occurrence time in each period are extracted. The peak amplitude and the corresponding time stamp are paired to form a data pair. For example, the peak value extracted in a certain period in channel T1 is 2.86 m / s 2 , and the corresponding time stamp is 841 ms. The record T1-2.86-841 is formed. According to the time sequence, the three-element data is combined to form the peak time sequence. The above operation is repeated. The time-amplitude mapping structure is established for each channel to form the channel peak time mapping table. For example, the peak time extracted in a certain period in channel T2 is 843 ms and 847 ms, and the corresponding amplitudes are 3.12 m / s 2 and 2.74 m / s 2 , respectively. The mapping structure is {843:3.12, 847:2.74}. The mapping table of channel T5 is {844:4.21, 849:2.91}. After merging the mapping tables of all channels according to the number, they are uniformly named as M1-M10. The peak time mapping dataset in a unified format is constructed. The fields are channel number, peak amplitude, and peak time stamp. The sorting basis is ascending time stamp. The data units are: none, m / s 2 , and ms.

[0062] S102: Based on the peak time mapping dataset, identify the time interval difference between adjacent peak points in each channel peak time sequence, calculate the time offset value between each pair of adjacent peak points, and obtain channel time offset information;

[0063] Traverse the peak timestamp sequence of each channel, and calculate the time interval between adjacent two points in turn. Assuming that the timestamp of a channel is T1, T2, T3, the time interval vector is: Δt i = T i+1 -T i , wherein Δt i is the i-th time interval, T i and T i+1 are the timestamps of adjacent peak points, and the unit is ms. Assuming that the T1 timestamp sequence of a channel is 841, 845, and 849 ms, the corresponding Δt is 4 and 4 ms. Assuming that the set time interval reference value is Δt 基准 = 5 ms, the offset calculation uses: Δd = |Δt i - Δt 基准 |, wherein Δd is the offset value, Δt i is the actual interval value, and Δt 基准 is the set reference value. For example, when Δt i = 10 ms, Δd = |10-5| = 5 ms. Set the judgment threshold to ε = 2 ms. When Δd > ε, record the offset event and form the offset record table. Further define the offset intensity level coefficient as: wherein β is the offset level coefficient. The division criteria are: β < 0.2 is low offset, 0.2 ≤ β < 0.5 is medium offset, and β ≥ 0.5 is high offset. If β = 1.0, it belongs to the high offset type. Collect the offset records of all channels to form a channel time offset information set, including: channel number, peak sequence number, actual interval value, offset value, and offset level.

[0064] S103: According to the channel time offset information, adjust the step of the channel peak time sequence with offset, calibrate the starting point and interval length of each channel time axis, and generate a synchronous time sequence;

[0065] According to the channel number and time index position recorded as medium or high offset, correct the starting time and point interval of the corresponding time axis. Assuming that the original timestamp is T 原 , the offset value is Δd, and the offset coefficient is β, the corrected timestamp is: T * = T 原 - β·Δd, wherein T * is the calibrated timestamp, T 原 is the original timestamp, β is the offset level coefficient, and Δd is the offset value. For example, if T 原= 850ms, β = 1.0, Δd = 5ms, then T * = 845ms. The corrected timestamp of each channel needs to be aligned to the earliest peak point, and the points are unified, for example, the earliest time in multiple channels is 839ms, and all channels are synchronized and aligned to 839ms. For interval adjustment, the relative proportion is kept consistent, that is, the linear redistribution is performed according to the original interval proportion in the corrected time sequence. The time sequence after all corrections is uniformly stored as a synchronous time sequence, the structure is: channel number + corrected timestamp vector, for example, {T1: [839, 843, 847], T2: [839, 846, 856]}, which is used as the input time reference for subsequent frequency domain processing.

[0066] Based on the synchronous time sequence, the energy value of each high-frequency band is extracted, the abnormal change band is detected by analyzing the energy change rate of each high-frequency band in multiple time windows, and the weight coefficient is matched according to the change amplitude, and the steps of constructing the high-frequency energy feature vector are as follows:

[0067] S201: Based on the synchronous time sequence, the frequency domain division processing is performed on each vibration signal channel, the signal response strength in the high-frequency band range is called, the instantaneous energy value of each high-frequency band in the continuous sampling period is extracted, the corresponding set of band number and energy value is established, and the frequency band energy extraction data table is generated;

[0068] Taking the synchronized signal of each channel as the input source, the fast Fourier transform processing is performed on the signal, the time domain signal is converted into frequency energy distribution form, the upper limit of frequency sampling is set to 12800Hz, and when the frequency band is divided, each 1000Hz is numbered as a group, the numbering format is F1 to F12, in the frequency band range corresponding to each number, the frequency response amplitude value is called, and the energy value of each frequency band in the continuous period is selected as the instantaneous energy input, wherein the energy value is defined as the average of the sum of the amplitude values of all frequency points in the frequency band, and the energy of each frequency band is constructed in the data table according to the three-dimensional structure of channel number, frequency band number and period number. For example, in channel T1, the number F4 corresponds to the frequency band of 3000-4000Hz, and in the 5th sampling period, all frequency point amplitudes are {2.3, 1.8, 2.7, 2.9}m / s 2 , then the frequency band energy is E = (2.3 2 + 1.8 2 + 2.7 2 + 2.9 2 ) / 4 = 7.015m 2 / s 4, record as T1–F4–5–7.015, the above structure is constructed in a dictionary table manner, the fields are channel number, frequency band number, cycle number, energy value, wherein the channel number is derived from the sensor identification corresponding to the synchronization time sequence, the frequency band number is determined according to the frequency interval, the cycle number is generated by taking the interval of the sampling frame number as an integer, the energy value is a real floating point number, and the data table is finally output as a set of energy values of all channels and all frequency bands in multiple cycles.

[0069] S202: According to the frequency band energy extraction data table, the energy values of each frequency band in multiple time windows are compared, the energy change amplitude value sequence in adjacent windows is extracted, the deviation value of the energy change amplitude value and the energy change reference value is calculated, the frequency band set whose deviation value exceeds the energy change threshold is identified, and a high-frequency abnormal frequency band set is obtained;

[0070] Taking the frequency band number as an index, the energy value sequence in the continuous time window is extracted, the energy change amplitude between each adjacent window is calculated, the time window length is set to 10 sampling cycles, the energy of frequency band F4 in window W1 and W2 is E1 and E2 respectively, the change amplitude value ΔE = |E2-E1| is calculated, the calculation is performed for each pair of adjacent windows, the change amplitude value sequence is formed, and further deviation identification is performed on the change amplitude value. The energy change reference value is set to μ = 5.0m 2 / s 4 , the energy change threshold is ε = 2.5m 2 / s4, the deviation calculation result is δ = |ΔE-μ|, when δ>ε, it is determined that the frequency band is an abnormal change frequency band, and the corresponding frequency band number, channel number and time window number are stored in the abnormal frequency band temporary table. For example, the energy values of frequency band F7 in channel T2 between W1 and W2 are 12.6 and 6.0, then ΔE = 6.6, δ = 1.6, which does not meet the condition; while the energy values of F9 in T4 between W3 and W4 are 4.5 and 9.8, then ΔE = 5.3, δ = 0.3, which meets the condition and is included in the abnormal set. The abnormal frequency band set field includes channel number, frequency band number, window number, change amplitude value and deviation value, and finally a high-frequency abnormal frequency band set is generated.

[0071] S203: Call the high-frequency abnormal frequency band set, extract the energy change amplitude value of each frequency band, match the weight coefficient by using the change strength range corresponding to each frequency band, perform data fusion on the energy value of each frequency band, and generate a high-frequency energy feature vector;

[0072] Take each abnormal frequency band record as an index, take out its corresponding change amplitude ΔE value, according to the historical energy change range of the frequency band, the intensity is graded, and the weight coefficient is matched, the weight coefficient setting reference rule is: if ΔE≤2.5, assign 0.2; if 2.5<ΔE≤5.0, assign 0.5; if ΔE>5.0, assign 0.8. Take the frequency band F8 of channel T5 as an example, the change amplitude is 6.2, then the weight is 0.8, then the energy value of the frequency band in all cycles is executed weighted operation, assuming that the energy value in a certain cycle is 9.3, then the weighted energy value is E'=9.3x0.8=7.44, the process is repeated for all abnormal frequency bands, and finally the weighted energy value sequence of each frequency band on each channel is generated, the sequence is arranged in ascending order according to the frequency band number, and integrated into a vector format, for example, the output of channel T1 is [7.44, 5.28, 3.96,…], a unified high-frequency energy feature vector data structure is constructed, the fields include channel number, frequency band number, weighted energy value, weight coefficient, and the output is a standard structure data set for the next stage feature selection processing module to input and use.

[0073] Based on the high-frequency energy feature vector, statistical feature indexes of each feature item under multiple fault working conditions are extracted, the relative dispersion rate is calculated, the feature items with a relative dispersion rate exceeding a screening threshold are screened, and according to the change trend of the feature density in the sample distribution, the clustering boundary is divided, and the clustering grouping result is generated. The steps are as follows:

[0074] S301: Based on the high-frequency energy feature vector, the energy value sequence of each feature item in each channel under multiple working conditions is extracted, the statistical feature indexes of each feature item under multiple fault working conditions are obtained, including mean value, fluctuation value and dispersion degree, and feature item statistical index information is generated;

[0075] The high-frequency energy feature vector corresponding to each channel is extracted from the synchronous vibration signal data collected under different working conditions (such as normal, slight peeling, inner ring crack, outer ring damage, etc. typical bearing fault working conditions), and the multi-condition sample sequence of each frequency band energy value is extracted. For example, the weighted energy value sequence of channel T1 frequency band F4 in working condition A is [6.2, 6.5, 6.1], and in working condition B it is [8.9, 9.1, 8.7]. The energy values under the above different working conditions are arranged and combined to form a multi-condition energy value matrix of the feature item. Then, the statistical indicators are calculated for each feature item under each working condition. The mean value is calculated by dividing the sum of all sample values by the sample size. The fluctuation value is calculated by subtracting the maximum value from the minimum value. The dispersion degree is calculated by calculating the standard deviation of the sample. For example, under channel T2 frequency band F5, the energy sample of working condition C is [5.2, 5.8, 6.0], the mean value is 5.67, the fluctuation value is 0.8, and the dispersion degree standard deviation is about 0.4. All the calculated mean value, fluctuation value and dispersion degree indicators are arranged in table form, with each row recording the channel number, frequency band number, working condition type, mean value, fluctuation value and dispersion degree. Finally, the feature item statistical indicator information is generated, providing standardized basic data for subsequent deviation rate screening and clustering grouping.

[0076] S302: Call the feature item statistical indicator information, extract the fluctuation value and dispersion degree value of each feature item under each working condition, calculate the relative deviation rate, and compare it with the set relative deviation rate screening threshold value. The feature item set with deviation proportion exceeding the threshold value is screened out to obtain the deviation rate feature screening result;

[0077] Take the feature item as the index, extract the corresponding fluctuation value sequence and dispersion degree value sequence from the statistical indicator information table. For each feature item, take the fluctuation value sequence and dispersion degree value sequence under each working condition, and calculate the deviation rate with the feature item as the unit. The relative deviation rate is defined as the ratio of the fluctuation value to the mean value multiplied by 100%, i.e. the deviation rate equals (fluctuation value / mean value) x 100%. For example, the mean value of channel T3 frequency band F6 under working condition A is 5.0, and the fluctuation value is 1.5, so the deviation rate is 30%. After the deviation rate of each feature item under each working condition is calculated, it is compared with the set deviation rate screening threshold value, which is set to 25%. When the deviation rate exceeds 25%, the feature item is included in the candidate set to form the deviation rate feature screening result. The screening result consists of channel number, frequency band number, working condition type and deviation rate, and records all feature items that meet the screening conditions for subsequent feature clustering operation.

[0078] S303: According to the deviation rate feature screening result, extract the distribution frequency of each feature item in the sample feature space, identify the feature item density change, divide the clustering boundary, and generate the clustering grouping result;

[0079] The specific formula for identifying the feature item density change is:

[0080]

[0081] Calculate the feature item density change index;

[0082] Where D i represents the density change index of the i-th feature item, f ij represents the distribution frequency value of the i-th feature item in the j-th sample, represents the average distribution frequency of the i-th feature item in all samples, N represents the total number of samples, d ik represents the density value of the i-th feature item within the k-th local neighborhood, represents the average density value of the i-th feature item in all local neighborhoods, M represents the number of local neighborhoods, i represents the index of the feature item, j represents the index of the sample in the global sample sequence, and k represents the index of the sample within the local neighborhood range.

[0083] Formula:

[0084]

[0085] Formula details and formula calculation derivation process:

[0086] The formula is used to calculate the density change index of the feature item in the sample feature space, and the result is used to divide the feature item clustering boundary;

[0087] Parameter meaning and setting value:

[0088] f ij is the distribution frequency value of the i-th feature item in the j-th global sample, and f 11 = 0.32, f 12 = 0.28, f 13 = 0.35, f 14 = 0.30, f 15 = 0.33;

[0089] is the average distribution frequency of the i-th feature item in all samples, and is calculated by summing all f ij and dividing by the sample number N, where N is the global sample number, and N = 5;

[0090] d ik is the density value of the i-th feature item within the k-th local neighborhood, and the density value is obtained by normalizing the number of neighboring points of the local sample point in the feature space, and d i1 = 0.12, d i2 = 0.15, d i3 = 0.11;

[0091] is the average density value of the ith feature in all local neighborhoods, and is the density value of the ith feature in the jth local neighborhood. ik After summation, divide by the number of local neighborhoods M, M is the number of local neighborhoods, and let M = 3.

[0092] Substitute the parameters into the formula for calculation:

[0093]

[0094] 0.000016 + 0.001296 + 0.001156 + 0.000256 + 0.000196 = 0.00292;

[0095]

[0096] -0.0067 + 0.0233 - 0.0167 = 0.000;

[0097] D i = 0.000584 + 0 = 0.000584;

[0098] The result 0.000584 indicates that the density of the ith feature changes very little in the global sample space and in the local neighborhood, and this result is used to determine the clustering attribution of the feature. The feature with small density change tends to belong to the same clustering cluster.

[0099] Based on the clustering grouping result, the frequency band is divided according to the fixed frequency width, and the energy proportion is calculated. The peak section of the energy proportion change amplitude is extracted, and the corresponding channel is identified. The abnormal response channel is obtained. In combination with the sensor point position, the abnormal response area is identified and the bearing position is located. The steps of generating the fault positioning result are as follows:

[0100] S401: Based on the clustering grouping result, the vibration energy value of each channel in the full frequency range is extracted. According to the set fixed frequency width parameter, the frequency axis is divided into multiple continuous frequency sections. The energy cumulative value of each channel in each frequency section is obtained. The energy proportion of each frequency section is calculated, and the energy proportion data is generated.

[0101] For the vibration signal data with divided cluster labels, the frequency energy spectrum curve corresponding to each channel is called, all frequency points in the range of 0 Hz to the upper limit of the sampling frequency are traversed, the upper limit of the sampling is set to 12800 Hz, the frequency axis is divided into multiple continuous frequency sections according to the set fixed frequency width parameter, the fixed frequency width is set to 500 Hz, that is, F1 is 0-500 Hz, F2 is 500-1000 Hz, and so on. 25 frequency sections are divided, the energy accumulation of each channel in each section is performed, the accumulation method is to sum the amplitude square at all frequency points in the section and then take the average, for example, the amplitudes of channel T1 in section F3 are 2.1, 2.4, 2.3 and 2.5 m / s 2 , then the energy accumulation value is calculated as (2.1 2 +2.4 2 +2.3 2 +2.5 2 ) / 4=5.3225m 2 / s 4 , then the energy accumulation value of each frequency section of each channel is calculated, the energy proportion of the section is the ratio of the energy value of the single section to the total energy value of the frequency section multiplied by 100%, for example, the total energy of T1 channel is 150 m 2 / s 4 , the energy of a frequency section is 7.5 m 2 / s 4 , then the energy proportion of the section is 5%, the energy proportion values of all frequency sections are stored as a list, and the channel number and frequency section number are associated, and finally the energy proportion data table is sorted, the fields are channel number, frequency section number and energy proportion, which is used for subsequent identification of abnormal frequency sections.

[0102] S402: Call the energy proportion data, identify the frequency section with the peak value of energy proportion change amplitude in each channel, and extract the corresponding channel number to obtain a set of abnormal response channels;

[0103] Based on the energy proportion data table, the energy proportion change amount of each channel is traversed in the continuous frequency section, the change amplitude is defined as the absolute value of the difference between the energy proportions of adjacent frequency sections, if the energy proportion of the frequency section Fi is Pi, the section Fi+1 is Pi+1, then the change amplitude is |Pi+1-Pi|, the change sequence is formed by extracting all change amplitude values, the peak value identification reference value is set to the average change amplitude plus a set reference, for example, the average change amplitude is 3%, the reference addition value is 2%, the peak threshold is 5%, when the change amplitude value is greater than 5%, the corresponding frequency section number and channel number are recorded, all frequency sections that meet the peak value condition are extracted as abnormal energy response sections, and the channel number corresponding to the frequency section is extracted, an abnormal response channel set is established, and the abnormal response channel set format is abnormal channel number+corresponding frequency section number set, the fields include channel number, abnormal section number, and peak change amplitude.

[0104] S403: According to the abnormal response channel set, the sensor arrangement coordinates corresponding to each channel are matched, the signal overlapping area boundary of the sensor concentrated area is identified, the energy distribution center point is analyzed and the bearing position is located, and a fault positioning result is generated;

[0105] The spatial coordinate data corresponding to each channel number in the sensor installation position table is called, the spatial coordinate data format is three-dimensional coordinates (x, y, z), the coordinate set corresponding to the abnormal response channel is extracted, the signal density in the local area of all abnormal channels is counted based on the coordinate points of all abnormal channels, the local radius is set to 30 cm, the spatial distances between all sensor points are calculated according to the Euclidean distance, if the distance between two sensors is less than 30 cm, they are recorded as a pair of adjacent relationship, the number of adjacent sensors of each sensor point is counted, and the sensor region with an adjacent number greater than or equal to 3 is identified as a signal overlapping region. Further analyze the average energy proportion of each channel frequency section in the overlapping region, set the center point of the sensor region with the highest average energy proportion as the energy distribution center point, and locate the sensor number corresponding to the center point as the potential fault source sensor. Combined with the actual assembly relationship table of the sensor and the bearing, the corresponding bearing number is mapped, and the fault positioning result is output. The fault positioning result field is the bearing number, the corresponding sensor number, and the energy peak frequency section.

[0106] Based on the fault positioning result, the synchronous vibration signal data of the abnormal response channel is called, the impact amplitude feature, the vibration kurtosis feature, the vibration skewness feature and the frequency main peak feature of the channel signal in the time domain are extracted, and compared with the data features of a plurality of known fault modes, the fault type of the target bearing is identified, and the steps of outputting the fault type diagnosis result are as follows:

[0107] S501: Based on the fault location result, the synchronous vibration signal data corresponding to the abnormal response channel is called, the instantaneous amplitude sequence of each signal in the continuous sampling interval in the time domain is extracted, the impact amplitude characteristic value, the vibration kurtosis characteristic value and the vibration skewness characteristic value are obtained, and the time domain characteristic index set is generated;

[0108] With the identified abnormal response channel number as the index, the vibration signal data segment of the corresponding channel is extracted from the synchronous time sequence, each data segment is divided according to a fixed sampling period, for example, 1024 points are taken as a continuous sampling interval, the instantaneous amplitude sequence of each signal is extracted, the instantaneous amplitude extraction is the absolute value of the signal, then the impact amplitude characteristic value of each signal is calculated, the impact amplitude characteristic is defined as the difference between the maximum amplitude and the average amplitude in the interval, assuming that the maximum amplitude of the first section is 5.2 m / s 2 , the average amplitude is 1.8 m / s 2 , then the impact amplitude characteristic is 3.4 m / s 2 , the vibration kurtosis characteristic of each signal is extracted, the kurtosis is defined as the ratio of the fourth order central moment to the square of the second order central moment, the vibration skewness characteristic is extracted as the ratio of the third order central moment to the cube of the standard deviation, in the specific operation, if the standard deviation of a section signal is 0.9 and the third order central moment is 0.5, then the skewness characteristic value is about 0.685, after the above characteristic extraction is completed, the impact amplitude, vibration kurtosis and vibration skewness three indexes extracted in each sampling interval of each channel are arranged into a vector in time sequence, the channel number, interval number and characteristic value three tuples are recorded, and finally the time domain characteristic index set is generated.

[0109] S502: Call the time domain characteristic index set, obtain the frequency distribution of each channel signal in a fixed time window, extract the frequency value corresponding to the peak amplitude in the frequency amplitude sequence, and record the main peak frequency position and amplitude change range of each channel, establish the correspondence between the channel and the main frequency characteristic, and obtain the frequency main peak characteristic information;

[0110] The short-time Fourier transform processing is performed on the signal data in each continuous sampling interval with the channel number as the index, the frequency-amplitude pairs in the corresponding time window are extracted, the peak amplitude in the frequency amplitude sequence is extracted and the corresponding frequency value is recorded, the peak extraction rule is set as selecting the local maximum amplitude and the amplitude value is higher than 1.5 times of the average amplitude of the frequency band, if the frequency amplitude sequence in the interval is {1.5, 2.0, 5.2, 2.4, 1.8}, the average amplitude is 2.58, then 5.2 is the main peak amplitude meeting the condition, the corresponding frequency is 3600 Hz, the frequency position and the amplitude variation range are recorded, the amplitude variation range is defined as the difference between the main peak amplitude and the secondary peak amplitude divided by the main peak amplitude multiplied by 100%, if the secondary peak amplitude is 2.4, the variation range is 53.8%, all the extraction results are arranged according to the channel number, the corresponding relationship table of the main peak frequency position and the amplitude variation range of each channel is generated, and the frequency main peak characteristic information is output.

[0111] S503: According to the frequency main peak characteristic information, the fault type of the target bearing is identified by comparing with the data characteristics of a plurality of known fault modes, and a fault type diagnosis result is output.

[0112] The specific formula for comparison with the data characteristics of a plurality of known fault modes is:

[0113]

[0114] The fault matching similarity index is calculated.

[0115] Wherein, S m represents the matching similarity index of the mth fault mode, F mn represents the frequency characteristic value of the to-be-identified channel on the nth characteristic item, F' mn represents the reference frequency characteristic value of the mth fault mode on the nth characteristic item, P represents the number of characteristic items, m represents the fault mode number, and n represents the characteristic item number.

[0116] Formula:

[0117]

[0118] Formula details and formula calculation derivation process:

[0119] The formula is used to calculate the matching similarity index between the target channel characteristic vector and the known fault mode characteristic vector, and the result is used for fault type diagnosis.

[0120] Parameter meaning and setting value:

[0121] F mn is the frequency characteristic value of the to-be-identified channel on the nth characteristic item, which is detected and obtained by wavelet packet decomposition of the vibration signal, and the unit is m / s2 The frequency amplitude of a certain test sample in five typical frequency band energy items is respectively F m1 = 2.3, F m2 = 1.8, F m3 = 2.9, F m4 = 2.1, F m5 = 1.7;

[0122] F' mn The reference frequency characteristic value of the mth failure mode on the nth characteristic item, derived from the same bearing failure record in the historical failure database, has the same unit as m / s 2 The reference value is F' m1 = 2.5, F' m2 = 1.9, F' m3 = 3.0, F' m4 = 2.2, F' m5 = 1.6;

[0123] P is the number of characteristic items, and the number of frequency energy characteristics is 5, that is, P = 5;

[0124] The parameters are substituted into the formula for calculation:

[0125]

[0126] 4.923 * 5.124 = 25.23;

[0127]

[0128] The result 0.9992 indicates that the characteristic vector extracted by the detection channel is similar to the target reference failure mode, and the target bearing can be preliminarily judged as the identified failure type.

[0129] The above-described embodiments can be implemented in whole or in part by software, hardware (e.g., circuitry), firmware, or any combination thereof. When implemented in software, the above-described embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center through wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid state disk.

[0130] It should be understood that the term "and / or" herein merely describes an association relationship of associated objects, which means that there can be three relationships, for example, A and / or B can represent three cases of A alone, A and B together, and B alone, where A and B can be singular or plural. In addition, the character " / " herein generally represents an "or" relationship between the front and rear associated objects, but it can also represent an "and / or" relationship, which can be understood in the context before and after.

[0131] In the present application, "at least one" means one or more, and "multiple" means two or more. "At least one of the following" or the like means any combination of the items, including any combination of single or multiple items. For example, at least one of a, b, or c can represent a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be single or multiple.

[0132] It should be understood that in various embodiments of the present application, the size of the sequence number of the above-described processes does not mean the order of execution, and the execution order of the processes should be determined by their functions and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0133] Those skilled in the art can clearly understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0134] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the devices, apparatuses and units described above can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0135] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0136] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0137] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically independently, or two or more units can be integrated into one unit.

[0138] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the present application that essentially contribute to the prior art or the parts of the technical solutions can be embodied in the form of software products. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0139] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A multi-bearing fault location diagnostic method, characterized by, The method comprises: S1: acquiring bearing vibration signal sequences collected by multiple sensing points of a vehicle body, extracting peak points and corresponding time stamps in a continuous period in each sequence, calling each channel peak time sequence, identifying time offset between adjacent points and calibrating, and generating a synchronous time sequence; S2: based on the synchronous time sequence, extracting the energy value of each high-frequency band, analyzing the energy change rate of each high-frequency band in multiple time windows, detecting abnormal change bands and matching weight coefficients according to the change amplitude, and constructing a high-frequency energy feature vector; S3: based on the high-frequency energy feature vector, extracting statistical feature indexes of each feature item under multiple fault conditions, calculating the relative dispersion rate, screening feature items with a relative dispersion rate exceeding a screening threshold, and dividing the clustering boundary according to the change trend of feature density in sample distribution, and generating a clustering grouping result; S4: based on the clustering grouping result, dividing frequency sections according to a fixed frequency width and calculating energy proportion, extracting energy proportion change amplitude peak sections and identifying corresponding channels, acquiring abnormal response channels, identifying abnormal response areas and positioning bearing positions in combination with sensor point positions, and generating a fault positioning result.

2. The multi-bearing fault location diagnostic method of claim 1, wherein, The synchronous time sequence is specifically a peak time alignment list, a time offset correction vector, and a channel time reference unified identifier. The high-frequency energy feature vector is specifically local high-frequency energy distribution data, a high-frequency energy weight coefficient sequence, and a high-frequency abnormal frequency band identification mark. The clustering grouping result includes a feature dispersion screening subset, a feature density change boundary, and a preliminary classification cluster of fault samples. The fault positioning result specifically refers to an abnormal energy peak frequency band, a corresponding response measurement point number, and bearing position information.

3. The multi-bearing fault location diagnostic method of claim 1, wherein, The steps of acquiring bearing vibration signal sequences collected by multiple sensing points of a vehicle body, extracting peak points and corresponding time stamps in a continuous period in each sequence, calling each channel peak time sequence, identifying time offset between adjacent points and calibrating, and generating a synchronous time sequence are specifically: S101: acquiring bearing vibration signal sequences collected by multiple sensing points of a vehicle body, extracting peak points and corresponding time stamps in a continuous period in each signal sequence, establishing a peak time mapping table for each channel, and generating a peak time mapping dataset; S102: based on the peak time mapping dataset, identifying time interval differences between adjacent peak points in each channel peak time sequence, calculating time offset values between each pair of adjacent peak points, and acquiring channel time offset information; S103: based on the channel time offset information, adjusting the step length of the channel peak time sequence with offset, calibrating the start point and interval length of each channel time axis, and generating a synchronous time sequence.

4. The multi-bearing fault location diagnostic method of claim 3, wherein, The steps of extracting the energy value of each high-frequency band based on the synchronous time sequence, analyzing the energy change rate of each high-frequency band in multiple time windows, detecting abnormal change bands, and matching weight coefficients according to the change amplitude to construct a high-frequency energy feature vector are specifically: S201: Based on the synchronization time sequence, the frequency domain division processing is performed on each vibration signal channel, the signal response strength in the high frequency band range is called, the instantaneous energy value of each high frequency band in the continuous sampling period is extracted, the corresponding set of band number and energy value is established, and the frequency band energy extraction data table is generated; S202: According to the frequency band energy extraction data table, the energy values of each frequency band in multiple time windows are compared, the energy change amplitude value sequence in adjacent windows is extracted, the deviation value between the energy change amplitude value and the energy change reference value is calculated, the frequency band set whose deviation value exceeds the energy change threshold is identified, and the high frequency abnormal frequency band set is obtained; S203: The high frequency abnormal frequency band set is called, the energy change amplitude value of each frequency band is extracted, the change strength range matching weight coefficient corresponding to each frequency band is utilized, the energy value of each frequency band is data fused, and the high frequency energy feature vector is generated.

5. The multi-bearing fault location diagnostic method of claim 4, wherein, Based on the high frequency energy feature vector, the statistical feature index of each feature item under multiple fault conditions is extracted, the relative dispersion rate is calculated, the feature items whose relative dispersion rate exceeds the screening threshold are screened, and the clustering boundary is divided according to the change trend of feature density in the sample distribution, and the clustering grouping result is generated. The steps are specifically as follows: S301: Based on the high frequency energy feature vector, the energy value sequence of each feature item in each channel under multiple conditions is extracted, the statistical feature index of each feature item under multiple fault conditions is obtained, including mean value, fluctuation value and dispersion degree, and the feature item statistical index information is generated; S302: The feature item statistical index information is called, the fluctuation value and dispersion degree value of each feature item under each condition are extracted, the relative dispersion rate is calculated, and compared with the set relative dispersion rate screening threshold, the feature item set whose deviation proportion exceeds the threshold is screened, and the dispersion rate feature screening result is obtained; S303: According to the dispersion rate feature screening result, the distribution frequency of each feature item in the sample feature space is extracted, the feature item density change is identified, the clustering boundary is divided, and the clustering grouping result is generated.

6. The multi-bearing fault location diagnostic method of claim 5, wherein, The specific formula for identifying the feature item density change is as follows: The feature item density change index is calculated; where D i denotes the density change indicator of the i-th feature item, f ij denotes the distribution frequency value of the i-th feature item in the j-th sample, denotes the average distribution frequency of the i-th feature item in all samples, N denotes the total number of samples, d ik denotes the density value of the i-th feature item within the k-th local neighborhood, denotes the average density value of the i-th feature item in all local neighborhoods, M denotes the number of local neighborhoods, i denotes the index of the feature item, j denotes the index of the sample in the global sample sequence, and k denotes the index of the sample within the local neighborhood range.

7. The multi-bearing fault location diagnostic method of claim 5, wherein, Based on the clustering grouping result, the frequency section is divided according to the fixed frequency width, the energy proportion is calculated, the energy proportion change amplitude peak section is extracted, the corresponding channel is identified, the abnormal response channel is obtained, the abnormal response area is identified and the bearing position is located in combination with the sensor point position, and the fault positioning result is generated. The steps are specifically as follows: S401: Based on the clustering grouping result, the vibration energy value of each channel in the full frequency band range is extracted, the frequency axis is divided into multiple continuous frequency sections according to the set fixed frequency width parameter, the energy cumulative value of each channel in each frequency section is obtained, the energy proportion of each frequency section is calculated, and the energy proportion data is generated; S402: The energy proportion data is called, the frequency section of the energy proportion change amplitude peak in each channel is identified, and the corresponding channel number is extracted, and the abnormal response channel set is obtained; S403: According to the abnormal response channel set, matching the sensor arrangement coordinates corresponding to each channel, identifying the signal overlapping area boundary of the sensor central area, analyzing the energy distribution center point and positioning the bearing position, and generating a fault positioning result.

8. The multi-bearing fault location diagnostic method of claim 1, wherein, The method further comprises: S5: Based on the fault positioning result, the synchronous vibration signal data of the abnormal response channel is called, the impact amplitude feature, the vibration kurtosis feature, the vibration skewness feature and the frequency main peak feature of the channel signal in the time domain are extracted, and compared with the data features of the known multiple fault modes, the fault type of the target bearing is identified, and a fault type diagnosis result is output; The fault type diagnosis result is specifically the channel feature extraction result, the feature comparison result and the fault mode attribution type.

9. The multi-bearing fault location diagnostic method of claim 8, wherein, Based on the fault positioning result, the synchronous vibration signal data of the abnormal response channel is called, the impact amplitude feature, the vibration kurtosis feature, the vibration skewness feature and the frequency main peak feature of the channel signal in the time domain are extracted, and compared with the data features of the known multiple fault modes, the fault type of the target bearing is identified, and a fault type diagnosis result is output, and the steps are specifically: S501: Based on the fault positioning result, the synchronous vibration signal data corresponding to the abnormal response channel is called, the instantaneous amplitude sequence in the continuous sampling interval of each signal in the time domain is extracted, the impact amplitude feature value, the vibration kurtosis feature value and the vibration skewness feature value are obtained, and a time domain feature index set is generated; S502: The time domain feature index set is called to obtain the frequency distribution of each channel signal in a fixed time window, the peak amplitude corresponding to the frequency value in the frequency amplitude sequence is extracted, and the main peak frequency position and amplitude change range of each channel are recorded, the correspondence between the channel and the main frequency feature is established, and the frequency main peak feature information is obtained; S503: According to the frequency main peak feature information, by comparing with the data features of the known multiple fault modes, the fault type of the target bearing is identified, and a fault type diagnosis result is output.

10. The multi-bearing fault location diagnostic method of claim 9, wherein, The specific formula for comparison with the data features of the known multiple fault modes is: Calculate the fault matching similarity index; where S m a matching similarity index representing the mth fault mode, F mn a frequency feature value of the to-be-identified channel on the nth feature item, F' mn a reference frequency feature value of the mth fault mode on the nth feature item, P represents the number of feature items, and m and n represent fault mode numbers and feature item numbers, respectively.

Images