A method for analyzing acoustic emission signal characteristics for rotating machinery condition monitoring

Abstract

The present invention relates to the technical field of acoustic emission signal feature analysis, and aims to solve the problem of feature extraction and analysis of high-noise emission signals. The present invention provides an acoustic emission signal feature analysis method for rotating machinery state monitoring, comprising: using data of rotating equipment after maintenance as benchmark data, calculating features of the benchmark data; calculating a threshold value of the benchmark feature, calculating features of normal operating data, obtaining a mean and a standard deviation of the feature, and using a threshold value exceeding K times the mean as a feature abnormality; during real-time monitoring, adopting a periodic triggering acquisition method, each time acquiring acoustic emission data of N seconds in length, and calculating corresponding feature indicators, while simultaneously storing S*N seconds of cached data in a cache queue; comparing the real-time feature indicator with the benchmark feature indicator threshold value, and if the feature indicator exceeds the threshold value, further clarifying the reason why the acoustic emission signal exceeds the threshold value, and simultaneously storing the cached original data in the form of a file; and if the feature indicator does not exceed the threshold value, only storing the real-time feature in a database.

Description

Technical Field

[0001] The present invention relates to the technical field of acoustic emission signal characteristic analysis, and in particular to an acoustic emission signal characteristic analysis method for rotating machinery state monitoring. Background Art

[0002] Existing vibration monitoring often has difficulty detecting early-stage faults in rotating machinery. Sliding bearings, thrust bearings, mechanical seals, shaft seals, and other key and vulnerable components of rotating machinery are often prone to various types of faults during operation. For early-stage faults, abnormal stress wave signals are released when moving and static components rub against each other. These signals are microscopic and difficult to identify through traditional vibration monitoring. However, acoustic emission monitoring technology can effectively identify early-stage faults in bearings and seals commonly used in rotating machinery. Acoustic emission monitoring technology has good sensitivity and a high sampling rate, and can effectively identify stress wave signals released by early-stage faults. However, due to the high sensitivity and high sampling rate of the signal, the signal is easily interfered with by noise. Currently, when using acoustic emission monitoring technology for early-stage fault monitoring and diagnosis of rotating machinery, there is a lack of effective signal analysis methods and means. Summary of the Invention

[0003] The purpose of the present invention is to provide an acoustic emission signal feature analysis method for rotating machinery state monitoring, which solves the problem of feature extraction and analysis of high-noise emission signals and realizes early fault monitoring and diagnosis of rotating machinery based on acoustic emission signals.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] A method for analyzing acoustic emission signal characteristics for rotating machinery condition monitoring, comprising:

[0006] Step 1: Use the data of the rotating equipment maintenance completed as the baseline data and calculate the characteristics of the baseline data;

[0007] Step 2: Calculate the threshold of the baseline feature. Calculate the features of the normal operating data, obtain the mean and standard deviation of the features, and use the value that exceeds K times the mean as the threshold for feature abnormality.

[0008] Step 3: During real-time monitoring, a periodic trigger acquisition method is used to collect acoustic emission data for N seconds each time, and the corresponding characteristic indicators are calculated. At the same time, the cached data of S*N seconds is stored in the cache queue;

[0009] Step 4: Compare the real-time feature index with the baseline feature index threshold. If the feature index exceeds the threshold, further clarify the reason why the acoustic emission signal exceeds the threshold, and store the cached raw data in the form of a file. If the feature index does not exceed the threshold, only the real-time feature is stored in the database.

[0010] Furthermore, baseline data features include relative effective value, relative noise floor effective value, relative high-frequency energy, relative low-frequency energy, and relative number of impact events. Because acoustic emission signals are susceptible to factors such as assembly and disassembly, feature calculation and analysis require calculating the difference in value from the baseline data to mitigate the effects of these factors on absolute values.

[0011] Further, through Calculate the effective value of the acoustic emission signal, where RMS represents the effective value of the acoustic emission signal, X i It represents the acoustic emission signal sequence Signal=[X1,X2,…X N ], N represents the length of the signal sequence;

[0012] On this basis, the relative effective value of acoustic emission compared with the benchmark data is calculated using the following formula:

[0013] RMS_Relative=(RMS-RMS_Base) / RMS_Base

[0014] Where RMS_Relative represents the effective value of the relative acoustic emission signal, RMS_Base represents the effective value of the reference acoustic emission signal, and RMS represents the effective value of the real-time monitoring acoustic emission signal.

[0015] Furthermore, the calculation steps of the acoustic emission signal noise floor characteristic value (including effective value) are as follows:

[0016] Set an empty eigenvalue sequence and set a sliding window with a length of N to slide in the acoustic emission signal sequence according to the step size STEP until the last data of the acoustic emission signal is included in the window;

[0017] Calculate the eigenvalues ​​(including effective values) of the signal within the window, and add the eigenvalues ​​to the defined eigenvalue (including effective value) sequence, and repeat the above process until the eigenvalues ​​(including effective values) within all windows are calculated;

[0018] Draw a distribution histogram of the eigenvalue (including effective value) sequence and obtain the abscissa value at the first peak in the distribution histogram, which is the eigenvalue of the acoustic emission signal background noise.

[0019] Furthermore, the specific steps for calculating the effective value of the acoustic emission signal noise floor are as follows:

[0020] (1) Set a window W of length N and set the sliding window overlap rate to ε;

[0021] (2) Set up an empty eigenvalue list and continuously add eigenvalues ​​to the list through calculation;

[0022] (3) Starting from the first data of the acoustic emission signal, the first N values ​​in the acoustic emission signal sequence are included in the sliding window W to form a signal sequence in the sliding window, and the characteristic value (including the effective value) of the signal sequence in the sliding window is calculated;

[0023] (4) According to the sliding window overlap rate ε, the calculation window W is slid backward along the acoustic emission signal sequence. The sliding distance STEP is calculated by the following formula:

[0024] STEP=max(Floor((1-ε)×N),1)

[0025] In the formula, STEP represents the sliding distance, and Floor represents the integer rounded down. If the calculated STEP is 0, STEP is 1.

[0026] (5) Sliding the window according to the sliding distance STEP to obtain a new sliding window data sequence, and recalculating the eigenvalues ​​(including valid values) of the sliding window data sequence and adding them to the eigenvalue (including valid values) list;

[0027] (6) Repeat the above steps (4)-(5) until the last element of the acoustic emission signal sequence enters the sliding window, the sliding window is terminated, the corresponding eigenvalue is calculated, and it is added to the eigenvalue list, and finally the eigenvalue list is obtained;

[0028] (7) Draw a histogram of the feature list distribution. The calculation formula for the eigenvalue calculation accuracy γ is as follows:

[0029] γ=(max(Charater)-min(Charater)) / M

[0030] Among them, γ represents the eigenvalue calculation accuracy, max represents the maximum value, min represents the minimum value, Charater represents the eigenvalue list, and M represents the number of histogram horizontal coordinate grids;

[0031] (8) Obtain the characteristic value of the acoustic emission noise background signal according to the horizontal coordinate of the first highest point of the feature list distribution histogram;

[0032] (9) Select the length N of the window W;

[0033] (10) Select the sliding window overlap ratio ε;

[0034] (11) After the rotating equipment is installed, various characteristic values ​​of the background noise signal, including the effective value, can be calculated based on the selected sliding window length N and sliding window overlap rate ε;

[0035] (12) The relative acoustic emission signal noise floor effective value is calculated by the following formula:

[0036] RMSNoise_Relative=(RMSNoise-RMSNoise_Base) / RMSNoise_Base

[0037] RMSNoise_Relative represents the relative noise floor RMS value of the acoustic emission signal, RMSNoise_Base represents the reference signal noise floor RMS value, and RMSNoise represents the real-time noise floor RMS value of the acoustic emission signal.

[0038] Furthermore, the calculation steps of high-frequency signal energy are as follows:

[0039] The real-time acoustic emission signal sequence and the reference acoustic emission signal sequence are band-pass filtered by using a band-pass filter method, and the frequency retention range is 100K-350K. The real-time acoustic emission signal sequence and the reference acoustic emission signal sequence after filtering;

[0040] The following formula is used to calculate the effective value of the high-frequency energy of the reference acoustic emission signal and the effective value of the high-frequency energy of the real-time acoustic emission signal, and the relative change in the effective value of the high-frequency energy is calculated:

[0041] RMSHigh_Relative=(RMSHigh-RMSHigh_Base) / RMSHigh_Base

[0042] RMSHigh_Relative represents the relative high-frequency energy of the acoustic emission signal, RMSHigh_base represents the high-frequency energy of the reference signal, and RMSHigh represents the high-frequency energy of the real-time acoustic emission signal.

[0043] Furthermore, the calculation formula for the relative change in the effective value of the low-frequency energy signal is as follows:

[0044] RMSlow_Relative=(RMSlow-RMSlow_Base) / RMSlow_Base

[0045] RMSlow_Relative represents the relative low-frequency energy of the acoustic emission signal, RMSlow_base represents the low-frequency energy of the reference signal, and RMSlow represents the low-frequency energy of the real-time acoustic emission signal.

[0046] Furthermore, the steps for extracting the shock event signal sequence are as follows:

[0047] (1) Set an empty comparison noise sequence with an initial length of L, and add the first value X1 in the signal sequence to the comparison noise sequence L to form a new comparison noise sequence L1, and calculate the average value L of the comparison noise sequence L1 according to the following formula 1_ave and standard deviation L 1_delta :

[0048]

[0049] Where y i are the elements in L1. At this time, except for y1, all other elements are 0;

[0050] (2) Analyze the second value X2 in the signal sequence and use the following formula to determine whether X2 exceeds K times the standard deviation of the mean of the comparison background noise sequence:

[0051] |X2-L1_ave|>K*L1_delta

[0052] K is a constant that represents the degree to which X2 deviates from the mean of the comparison noise sequence. According to the Gaussian distribution principle, K is usually 3 or greater.

[0053] (3) If the formula in (2) is valid, do not add X2 to L1, continue to calculate the third value X3 in the signal sequence, and repeat the above steps;

[0054] (4) If the formula in (2) does not hold, add X2 to L1 to form a new comparison noise sequence L2, and calculate the average value L of the new noise sequence L2 according to the formula in step (1) 2_ave and standard deviation L 2_delta , continue to calculate the third value X3 in the signal sequence and repeat the above steps;

[0055] (5) Repeat steps (1)-(4) until the background noise sequence L is completely filled to form L N On this basis, a sliding window with a window length of WL is set and slid on the noise sequence L with a step size of 1 to obtain the data within the window. The variance of the data within the window is calculated and added to the comparison background noise variance list. The above process is repeated until the background noise sequence L is traversed to form the final comparison background noise variance list. The mean and standard deviation of the list are calculated to complete the initialization of the comparison background noise sequence.

[0056] (6) Set the temporary data list TMP with a window length of WW, and initialize the flag Flag of whether the impact event occurs to False. Slide the temporary data list TMP along the acoustic emission signal sequence. Take the Kth slide as an example to obtain TMP=[X 1+K ,X 2+K ,…,X WW+K ], calculate the variance of the data in the window, and determine whether the variance of the data in the window deviates from the SS times standard deviation of the mean of the comparison background noise variance list, as follows:

[0057] |TMP_Std-Std_Mean|>SS*Std_Delta

[0058] Where SS is a constant that indicates the degree to which the variance of the data within the window deviates from the mean variance of the background noise. According to the Gaussian distribution principle, SS is usually 3 or greater.

[0059] (7) If |TMP_Std-Std_Mean|>SS*Std_Delta holds, add the last data X in the window. WW Add an empty impact event data list Pulse = [X WW ], on this basis, continue to slide the temporary data list TMP backward for the K+1th time to obtain a new window data list TMP = [X 2+K ,X 3+K ,…,X WW+K+1 ], calculate the variance of the new window data, and determine whether the variance of the window data deviates from the SS times standard deviation of the mean of the variance list of the comparison background noise signal; if |TMP_Std-Std_Mean|>SS*Std_Delta again in the K+1th sliding, then X WW+K+1 Continue to add the impact event data list Pulse = [X WW , X WW+K+1 ], and continue to slide the temporary data list TMP backward; if the formula |TMP_Std-Std_Mean|>SS*Std_Delta does not hold in the K+1th slide, reset the flag Flag indicating whether the impact event occurs to False, and calculate the characteristics of the impact event signal in the impact event data list Pulse (impact energy Energy, number of crossings P and impact event length Q), and add them to the impact signal characteristic list, and then reset the impact event data list Pulse to empty; re-enter step (6) and continue to slide the temporary data list TMP backward;

[0060] (8) If |TMP_Std-Std_Mean|>SS*Std_Delta does not hold, set X WW+K Add to the end of the comparison noise sequence L, and at the same time pop out the first data in the original comparison noise sequence L, complete the update of the comparison noise list, and calculate the mean variance and standard deviation of the variance of the new comparison noise sequence L according to the formula in step (1);

[0061] (9) Repeat steps (6)-(8) until the temporary data list TMP slides to the end of the signal list and returns to the impact signal feature list.

[0062] Furthermore, for an impulse event signal sequence Pulse=[P1, P2, ... P S ], the calculation formula of the impulse signal energy is as follows:

[0063]

[0064] Where, Energy represents the energy value of a single impact event, P i Represents an element in the shock event sequence, and S represents the length of the shock event sequence.

[0065] Furthermore, in step 2, the threshold of the benchmark feature is calculated as follows:

[0066] Step 2.1: Calculate the characteristics of the reference acoustic emission signal;

[0067] Step 2.2: Use the data list with a window length of WD to generate the reference acoustic emission signal Signal_Base = [Y1, Y2, ..., Y i ,…Y N ] list, calculate the 8 features of the data in the window, and form an 8-feature list of the reference acoustic emission signal by sliding the window;

[0068] Step 2.3: Calculate the relative change of each feature list element compared to the baseline feature using the following formula:

[0069] Character_Relative K =(Character K -Character_Base) / Character_Base

[0070] Among them, Character_Relative K Represents the relative eigenvalue of the Kth element in each feature list, Character K Represents the Kth element in the feature list, and Character_Base represents the baseline feature value of the feature;

[0071] Step 2.4: Calculate the mean and standard deviation of the eight feature relative change lists and determine the relative feature change threshold range according to the following formula:

[0072] Threshold_K=[Mean_K-C*Delta_K,Mean_K+C*Delta_K]

[0073] Threshold_K represents the normal range of relative feature variation, Mean represents the mean, Delta represents the standard deviation, K represents the Kth feature, and C represents the abnormal coefficient, which indicates the degree of deviation from the feature mean. According to Gaussian distribution theory, C is usually 3 or greater.

[0074] Compared with the prior art, the acoustic emission signal characteristic analysis method for rotating machinery condition monitoring provided by the present invention has the following beneficial effects:

[0075] The present invention effectively identifies the early contact and lubrication status of key components of rotating machinery such as sliding bearings, thrust bearings, and mechanical seals, and can solve the problem of early fault monitoring and identification of key vulnerable components of rotating machinery.

[0076] The present invention proposes various index calculation methods for continuous acoustic emission signals, supplements the analysis and feature extraction methods of noisy continuous acoustic emission signals, and further expands the application of acoustic emission signals in the field of rotating machinery condition monitoring.

[0077] Based on the relative parameter changes, the lack of unified analysis and evaluation of rotating machinery acoustic emission signals under high sensitivity is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for the technical description.

[0079] Figure 1 This is a schematic diagram of the typical acoustic emission signal morphology collected directly by the acquisition equipment;

[0080] Figure 2 The acoustic emission signal analysis flow chart provided by the present invention;

[0081] Figure 3 This is a flowchart of the noise floor effective value calculation provided by the present invention;

[0082] Figure 4 This is a schematic diagram of calculating the effective value of the acoustic emission signal noise floor provided by the present invention;

[0083] Figure 5 A schematic diagram of the noise signal length and the impulse signal length provided by the present invention;

[0084] Figure 6 The flow chart of the impact event signal extraction provided by the present invention;

[0085] Figure 7 This is the effect diagram of the effective value extraction of the background noise signal provided by the present invention;

[0086] Figure 8 This is the impact event sequence recognition effect diagram provided by the present invention;

[0087] Figure 9 This is a diagram showing the impact signal characteristic distribution provided by the present invention;

[0088] Figure 10 A schematic diagram comparing the number of impact events provided by the present invention;

[0089] Figure 11 Schematic diagram of effective value and spectrum provided by the present invention;

[0090] Figure 12 Schematic diagram of severe friction downsampling and envelope spectrum provided by the present invention. DETAILED DESCRIPTION

[0091] The following is further explained in detail through specific implementation methods.

[0092] When using acoustic emission monitoring technology for early fault monitoring and diagnosis of rotating machinery, the typical acoustic emission signal form is as follows: Figure 1 shown.

[0093] When carrying out early fault monitoring and diagnosis of key components such as bearings and seals of rotating machinery based on acoustic emission signals, the main analyzed features include effective value, background noise effective value, high-frequency signal energy, low-frequency signal energy, impact event energy, impact event crossing times, impact event length, and impact event times. Early fault monitoring and diagnosis of rotating machinery based on acoustic emission is carried out in the form of online continuous monitoring and regular inspections. Taking online continuous monitoring as an example, the acoustic emission signal analysis process for rotating machinery status monitoring and fault diagnosis proposed in the present invention is as follows: Figure 2 As shown in the figure, the offline inspection method is basically the same.

[0094] Due to the high sensitivity of acoustic emission signals and the influence of factors such as installation, there is a certain error in directly using the absolute value of the acoustic emission signal. Therefore, when using acoustic emission to carry out rotating equipment condition monitoring, the main focus is on monitoring the relative changes in characteristic indicators. In specific applications, the monitoring data of the rotating machinery when it is newly installed and in good operating condition should be used as the baseline data. This data is used as the benchmark for evaluating and analyzing the data collected regularly offline and online.

[0095] Therefore, the present invention provides a method for analyzing acoustic emission signal characteristics for rotating machinery state monitoring, which specifically includes the following steps:

[0096] Step 1: Calculate the characteristics of the baseline data, using data from rotating equipment after maintenance is complete and in good operating condition. The determination of good condition is primarily based on expert opinion after the equipment is installed, commissioned, and put into operation.

[0097] Benchmark data features include relative effective value, relative noise floor effective value, relative high-frequency energy, relative low-frequency energy, and relative number of impact events. The meanings of each feature are as follows:

[0098] (1) The relative effective value reflects the energy of the acoustic emission signal. During stable operation, the energy of the acoustic emission signal should remain basically unchanged. When the energy of the acoustic emission signal increases significantly, it indicates that the interaction between the relative moving parts has intensified.

[0099] (2) The relative background noise effective value reflects the operating state of the friction pair. When the background noise increases slightly, it indicates that the state of the friction pair and the oil film are deteriorating. When the background noise increases significantly, it indicates that the contact state of the friction pair is poor, and even dry friction occurs.

[0100] (3) The relative high-frequency energy reflects the slight interaction between the friction pairs and the relative action of the fluid and the friction pairs. If the high-frequency energy increases, it indicates that the contact state between the sealing friction pairs is deteriorated and the seal has certain wear and tear problems.

[0101] (4) The energy in the relatively low-frequency band reflects the macroscopic interaction between the dynamic and static friction pairs. The significant increase in the energy in the low-frequency band indicates that there may be problems such as dynamic and static friction.

[0102] (5) The relative number of impact events reflects the mutual contact between the friction pairs. During normal operation, the relative movement between the friction pairs should be fully lubricated. When the lubrication is insufficient and the degree of direct contact between the friction pairs is aggravated, the number of impact events per unit time increases.

[0103] Step 2: Calculate the threshold of the baseline feature. Use existing sliding windows and statistical principles to calculate the features of the normal operation data, obtain the mean and variance of the features, and use the value exceeding K times the standard deviation of the mean as the threshold for feature abnormality, where K represents the degree to which the feature deviates from the baseline feature. According to the 3σ distribution principle, K is usually 3 or greater.

[0104] Step 3: During real-time monitoring, to avoid data explosion caused by high sampling rates, a periodic triggering acquisition method is used. Each time, acoustic emission data is collected for N seconds, and the corresponding characteristic indicators are calculated. At the same time, a buffer queue is stored for S*N seconds of buffered data. Here, S represents the number of buffered segments, and N represents the length of the buffered data.

[0105] Step 4: Compare the real-time characteristic index with the threshold of the baseline characteristic index. If the characteristic index exceeds the threshold, further clarify the reason why the acoustic emission signal exceeds the threshold from the perspectives of the acoustic emission signal waveform and impact characteristics, acoustic emission signal spectrum, envelope spectrum, etc. At the same time, the cached raw data is stored in the form of a file on the computer hard disk; if the characteristic index does not exceed the threshold, only the real-time characteristics are stored in the database. It is worth noting that high-frequency data has a single and continuous format. Directly storing it in the database will make data retrieval difficult. Storing it in the form of a file and indexing it in the database can greatly speed up the storage and retrieval of high-frequency data.

[0106] For a real-time acquired acoustic emission signal sequence Signal=[X1,X2,…,X i ,…X N ], its reference acoustic emission signal sequence Signal_Base=[Y1,Y2,…,Y i,…Y N ], the calculation process of typical acoustic emission characteristics is described below.

[0107] In step 1, the calculation steps of the effective value of the acoustic emission signal are as follows:

[0108] The calculation formula of the effective value of the acoustic emission signal is shown in formula (1-1):

[0109]

[0110] In the formula, RMS represents effective value, X i Represents an element in the acoustic emission signal sequence, and N represents the length of the acoustic emission signal sequence.

[0111] The effective value of the reference acoustic emission signal is RMS_Base, and the calculation formula of the relative acoustic emission effective value RMS_Relative is shown in formula (1-2):

[0112] RMS_Relative=(RMS-RMS_Base) / RMS_Base (1-2)

[0113] RMS represents the effective value of the real-time monitoring acoustic emission signal, RMS_Relative represents the relative effective value of the acoustic emission, and RMS_Base represents the effective value of the reference acoustic emission signal.

[0114] For the acoustic emission signal of normal operation, the signal is used as the reference signal, and the reference acoustic emission signal effective value is calculated as the reference effective value.

[0115] In step 1, the effective value of the acoustic emission signal noise floor is calculated as follows:

[0116] The impact signal in the acoustic emission signal is difficult to filter out directly, and the traditional method of filtering out the impact signal and then calculating the effective value of the background noise is difficult to implement. There are two main forms in the acoustic emission waveform, namely the impact signal and the background noise, of which the background noise accounts for a larger proportion. Therefore, by calculating the characteristic value of the acoustic emission signal in a certain width window, through sliding window and statistical principles, the background noise characteristics of the acoustic emission signal can be obtained by using the characteristic distribution. The calculation process of the effective value of the background noise RMSNoise of the acoustic emission signal is as follows: Figure 3 As shown, the details are as follows:

[0117] First, set an empty eigenvalue sequence, and set a sliding window with a length of N to slide in the acoustic emission signal sequence according to the step size STEP until the last data of the acoustic emission signal sequence is contained in the window; then, calculate the eigenvalue (including the effective value) of the signal in the window, and add the eigenvalue (including the effective value) to the defined eigenvalue (including the effective value) sequence, and repeat the above process until the eigenvalue (including the effective value) in all windows is calculated; finally, draw the distribution histogram of the eigenvalue (including the effective value) sequence, and obtain the horizontal coordinate value at the first peak in the distribution histogram, that is, the eigenvalue (including the effective value) of the acoustic emission signal background noise.

[0118] It should be noted that this method can also be used for other characteristic indicators of background noise, such as variance, peak-to-peak value, kurtosis and other common characteristic indicators. Since the background noise accounts for a larger proportion, the position where the first peak appears in the distribution diagram is the background noise characteristic value (including effective value). The calculation method of the effective value of the background noise of the acoustic emission signal is as follows: Figure 4 shown.

[0119] The specific calculation steps are as follows:

[0120] (1) Set a sliding window W with a length of N and set the sliding window overlap rate to ε.

[0121] (2) Set an empty eigenvalue list Charater = [], and continuously add eigenvalues ​​(including valid values) to the list through the calculations in steps (3) to (6) below.

[0122] (3) Starting from the first data of the acoustic emission signal, the acoustic emission signal sequence Signal=[X1,X2,…,X i ,…] are put into the sliding window W to form the signal sequence Signal_W1=[X1,X2,…,X i ,…X N ], calculate the eigenvalue (effective value) of the signal sequence Signal_W1 in the sliding window, and obtain Charater = [RMS1].

[0123] (4) According to the overlapping rate ε of the sliding window, the acoustic emission signal sequence Signal=[X1,X2,…,X i ,…] Slide the calculation window W backward, and the sliding distance STEP is calculated by formula (1-3):

[0124] STEP=max(Floor((1-ε)×N),1) (1-3)

[0125] Where ε represents the sliding overlap ratio, which can default to 0.9, N represents the sliding window length, and Floor represents rounding down to an integer. If the calculated STEP is 0, STEP is set to 1.

[0126] (5) Slide the window W according to the sliding distance STEP to obtain a new sliding window data sequence Data_W2 = [X 1+STEP ,X 2+STEP ,…,X N+STEP ], and recalculate the eigenvalues ​​(including valid values) of the sliding window data sequence Signal_W2, and add them to the eigenvalue (including valid values) list to obtain Charater = [RMS1, RMS2].

[0127] (6) Repeat the above steps (4) and (5) until the last K-th sliding window data sequence Signal_WK=[X 1+(K-1)*STEP ,X 2+(K -1)*STEP,…,X N+(K-1)*STEP The last element X in ] N+(K-1)*STEP Exceeded the acoustic emission signal sequence Signal=[X1,X2,…,X i ,…], the sliding window ends, and the remaining values ​​of the acoustic emission signal sequence are added to the K-th sliding window data sequence Signal_WK, and the corresponding eigenvalues ​​(including effective values) are calculated and added to the eigenvalue (including effective values) list, and finally the eigenvalue (including effective values) list Charater = [RMS1, RMS2,…, RMS K ].

[0128] (7) Divide the eigenvalues ​​(including effective values) into M equal parts from the minimum to the maximum, and draw a histogram of the distribution of the feature (including effective values) list Charater = [RMS1, RMS2, ..., RMSK]. The horizontal axis represents the eigenvalue. The calculation formula of the calculation accuracy γ of the eigenvalue is shown in (1-4):

[0129] γ=(max(Charater)-min(Charater)) / M (1-4)

[0130] Among them, γ represents the eigenvalue calculation accuracy, max represents the maximum value, min represents the minimum value, Charater represents the eigenvalue list, and M represents the number of histogram horizontal coordinate grids.

[0131] (8) According to the horizontal coordinate of the first highest point of the feature list distribution histogram, the characteristic value (including effective value) RMSNoise of the acoustic emission signal background noise can be obtained.

[0132] (9) Selection of the length N of the window W. N should not be too large or too small. A typical acoustic emission signal includes the noise length S and the impulse signal length L, such as Figure 5 shown.

[0133] The selection principle of sliding window W is as follows:

[0134] If the window W is too long, the background noise and the impulse signal will appear in the window at the same time. As a result, the features calculated when the window slides are the combined features of the background noise and the impulse signal. Therefore, the maximum length of the window cannot exceed the length S of the background noise signal between two impulse signals. In order to ensure that the window slides, in most cases, it only contains the noise signal.

[0135] If the window length is too short, the calculated eigenvalues ​​within the window will be unstable, and the calculated eigenvalues ​​may not truly reflect the eigenvalues ​​of the noise floor signal. When the window W slides in the noise floor, if the calculated characteristics of the noise floor within the window fluctuate too much, it indicates that the window length N is too short. The length N of the window W can be determined according to the following principles:

[0136] a. Initialize the length N of the window W to the minimum value of the impulse signal length L and 0.25 times S, that is, N = min(L, 0.25*S);

[0137] b. Make the window function slide in the background noise that does not contain the impact signal, calculate the background noise eigenvalue within the sliding window, and form the background noise eigenvalue sequence L = [C1, C2, ..., C M ], calculate the variance of the background noise feature sequence. If the variance exceeds the set value Ω, it means that the eigenvalue fluctuates too much and the window length is too short. The window length can be increased by 10%, and the above process can be repeated until the variance does not exceed the set value Ω.

[0138] (10) Selection of the sliding window overlap ratio ε. In theory, the larger the ε, the better. However, too large a value will lead to a greater amount of computation. During initialization, ε can be set to 0.99 by default. If the drawn histogram is not smooth enough, increase the sliding window overlap ratio ε. If the histogram is very smooth but the computation time is long, reduce the sliding window overlap ratio ε appropriately.

[0139] (11) After the rotating equipment is installed, the shape of the acoustic emission signal is basically determined. After selecting the appropriate sliding window length N and sliding window overlap rate ε according to the above steps (9) and (10), the above process can be used to obtain various characteristic values ​​of the acoustic emission signal background noise including the effective value.

[0140] (12) The effective value of the base noise floor of the reference signal obtained by the above method is RMSNoise_Base, and the effective value of the base noise floor of the real-time acoustic emission signal is RMSNoise. The calculation formula of the effective value of the base noise floor of the relative acoustic emission signal RMSNoise_Relative is shown in formula (1-5).

[0141] RMSNoise_Relative=(RMSNoise-RMSNoise_Base) / RMSNoise_Base(1-6)

[0142] RMSNoise_Relative represents the relative noise floor RMS value of the acoustic emission signal, RMSNoise_Base represents the reference signal noise floor RMS value, and RMSNoise represents the real-time noise floor RMS value of the acoustic emission signal.

[0143] In step 1, the high-frequency energy of the acoustic emission signal is calculated as follows:

[0144] The high-frequency energy of the acoustic emission signal can reflect the slight (microscopic) interaction between the moving and static components, and can identify problems such as seal wear and lubrication degradation. The high-frequency energy of the acoustic emission signal includes the energy range of 100K-350K. The calculation method of the high-frequency signal energy is as follows:

[0145] (1) Using bandpass filtering, the real-time acquired acoustic emission signal sequence Signal = [X1, X2, ..., X i ,…X N ] and the reference acoustic emission signal sequence Signal_Base=[Y1,Y2,…,Y i ,…Y N ] is band-pass filtered, and the frequency retention range is 100K-350K. The real-time acoustic emission signal sequence after filtering is Signal_HFilet=[Xx1,Xx2,…,Xx i ,…Xx N ] and the reference acoustic emission signal sequence Signal_Base_HFilt=[Yy1,Yy2,…,Yy i ,…Yy N ]

[0146] (2) Use formula (1-7) to calculate the effective value RMSHigh_Base of the high-frequency energy of the baseline acoustic emission signal and the effective value RMSHigh of the high-frequency energy of the real-time acoustic emission signal, and use formula (1-7) to calculate the relative change RMSHigh_Relative of the effective value of the high-frequency energy.

[0147] RMSHigh_Relative=(RMSHigh-RMSHigh_Base) / RMSHigh_Base(1-7)

[0148] RMSHigh_Relative represents the relative high-frequency energy of the acoustic emission signal, RMSHigh_base represents the high-frequency energy of the baseline signal, and RMSHigh represents the high-frequency energy of the real-time acoustic emission signal. In step 1, the low-frequency energy of the acoustic emission signal is calculated as follows:

[0149] The low-frequency energy of the acoustic emission signal can reflect the relatively serious (macroscopic) interaction between the moving and static components, and can identify abnormal friction, rubbing and other faults. The low-frequency n energy signal of the acoustic emission signal mainly refers to the signal within 20K. The calculation method of the low-frequency energy signal is basically the same as the calculation method of the high-frequency energy signal mentioned above. The effective value RMSlow_Base of the low-frequency energy of the baseline acoustic emission signal and the effective value RMSlow of the low-frequency energy of the real-time acoustic emission signal are calculated, and the relative change RMSlow_Relative of the effective value of the low-frequency energy signal is calculated using formula (1-8).

[0150] RMSlow_Relative=(RMSlow-RMSlow_Base) / RMSlow_Base (1-8)

[0151] RMSlow_Relative represents the relative low-frequency energy of the acoustic emission signal, RMSlow_base represents the low-frequency energy of the reference signal, and RMSlow represents the low-frequency energy of the real-time acoustic emission signal.

[0152] When the low-frequency energy component in the acoustic emission signal increases significantly, downsampling and envelope demodulation analysis are performed to further clarify the reason for the increase in low-frequency signal energy. The process of acoustic emission signal downsampling and envelope demodulation analysis is as follows:

[0153] (1) For the real-time collected acoustic emission signal Signal = [X1, X2, ..., X i ,…X N ], by downsampling the signal by 1 / D times (D is usually 100, 200, 500, 1000), the acoustic emission signal is downsampled to Signal_Resample = [xx1, xx2, ..., xx i ,…xx N / D ]

[0154] (2) Each value xx in the downsampled acoustic emission signal sequence i Perform Hilbert transform to obtain the transformed complex number yy i , find yy i Model|yy i |, and recompose the transformed acoustic emission signal sequence Signal_Trans=[|yy1|,|yy2|,…,|yy i |,…|yy N / D |]

[0155] (3) The transformed acoustic emission signal sequence Signal_Trans = [|yy1|,|yy2|,…,|yy i |,…|yy N / D|]By performing spectrum analysis, the envelope demodulation spectrum of the acoustic emission signal can be obtained. Combined with the rotation frequency f of the rotating machinery, it is observed whether the rotation frequency f and its multiple n*f are obviously present in the frequency components, so as to further determine whether serious dynamic and static friction has occurred.

[0156] In step 1, the acoustic emission signal impact characteristics are calculated as follows:

[0157] The impact characteristics of the acoustic emission signal include the number of impact events, impact event energy, impact event length, and impact event crossing times. By analyzing the impact event characteristics, we can determine whether the source of the impact signal has changed, and whether the contact state of the moving and static components has changed. The calculation of the impact characteristics of the acoustic emission signal first requires the extraction of the impact event. The extraction of the impact event adopts Figure 6 Extraction was performed using the procedures and methods shown.

[0158] The extraction of shock event signals is mainly based on comparing the fluctuations of the background noise signal with the fluctuations of the shock signal. Therefore, the change in variance can be used to distinguish the background noise and the shock signal (the effective value of the background noise signal and the effective value of the fluctuation signal are also significantly different, and other features such as the effective value can also be used for calculation and extraction). The specific implementation process is as follows:

[0159] (1) Set an empty comparison noise sequence L0 = [k1, k2, ..., k i ,…,k L ], and the signal sequence Signal=[X1,X2,…,X i ,…X N ] is added to the comparison noise sequence L to form a new comparison noise sequence L1=[X1,k2,…,k i ,…,k N ], and calculate the average value L of the noise sequence L1 according to formula (1-9) 1_ave and standard deviation L 1_delta :

[0160]

[0161] Among them, yi is the element in L1. At this time, except y1, the rest of the elements are 0.

[0162] (2) Analyze the signal sequence Signal=[X1,X2,…,X i ,…], and use formula (1-10) to determine whether X2 exceeds K times the standard deviation of the mean of the comparison background noise sequence (K should be at least greater than or equal to 4, and can be 4, 5, etc. In addition, a box plot can be used to determine whether X2 exceeds the normal range of the comparison background noise list. If it exceeds the normal range, it should be eliminated):

[0163] |X2-L 1_ave |>K*L 1_delta (1-10)

[0164] K is a constant that represents the degree to which X2 deviates from the mean of the comparison noise sequence. According to the Gaussian distribution principle, K is usually 3 or greater.

[0165] (3) If (1-10) is true, it means that X2 deviates too much from the normal value of the existing comparison background noise list. In order to avoid affecting the mean, X2 is not added to L1, and the signal sequence Signal = [X1, X2, ..., X i ,…] and repeat the above steps;

[0166] (4) If (1-10) does not hold, it means that X2 is within a reasonable range. Add X2 to L1 to form a new comparison noise sequence L2 = [X1, X2, ..., k i ,…,k N ], and calculate the average value L of the new background noise sequence L2 according to formula (1-9) 2_ave and standard deviation L 2_delta , continue to calculate the signal sequence Signal=[X1,X2,…,X i ,…] and repeat the above steps;

[0167] (5) Repeat the above steps (1) to (4) until the background noise sequence L is completely filled to form L N =[X1,X2,…,X t ,…,X L ], on this basis, set the sliding window with a window length of WL, slide on the noise sequence L with a step size of 1, and obtain the data in the window Data1_WL=[X1,X2,…,X WL ], calculate the data variance Std1 in the window, and add it to the comparison background noise variance list STD1 = [Std1], and slide one data backward to obtain the new window data Data2_WL = [X2, X3, ..., X WL+1 ], calculate the data variance Std2 in the window, and add it to the comparison background noise variance list STD2 = [Std1, Std2], repeat the above process until the background noise sequence L is traversed, and the final comparison background noise variance list STD = [Std1, Std2, ... Std (L-WL) ], calculate the mean Std_Mean and standard deviation Std_Delta of the list, and complete the initialization of the background noise sequence.

[0168] Regarding the selection of the length L of the background noise sequence, the variance of the background noise signal is used as the comparison value of the variance of the impulse signal. In order to ensure that the boundary of the variance of the background noise signal can be obtained smoothly, the length of L cannot be too short. If it is too short, the variance of the background noise signal used for reference will fluctuate too much, and the recognition of the impulse signal will be unstable and easy to miss. However, L should not be too long. If it is too long, the variance of the reference background noise signal will be too stable and easy to misidentify. According to the length of the noise signal in the acoustic emission signal waveform is S, the selection range of L should be [0.4S, 0.8S], and L should not be less than 40 data points. The minimum value of the window length WL used for calculating the reference background noise variance threshold is set to 10, and the maximum value is L-30.

[0169] (6) Set the temporary data list TMP with a window length of WW, and initialize the flag Flag indicating whether the impact event occurs to False, and store the temporary data list TMP along the acoustic emission signal Signal = [X1, X2, ..., X i ,…] performs the Kth sliding, and obtains TMP=[X 1+K ,X 2+K ,…,X WW+K ], calculate the variance TMP_Std of the data in the window, and determine whether the variance of the data in the window deviates from SS times the standard deviation of the mean of the variance list of the comparison background noise signal, that is, whether formula (1-11) is valid.

[0170] |TMP_Std-Std_Mean|>SS*Std_Delta (1-11)

[0171] Where SS is a constant that indicates the degree to which the variance of the data within the window deviates from the mean variance of the background noise. According to the Gaussian distribution principle, SS is usually 3 or greater.

[0172] (7) If |TMP_Std-Std_Mean|>SS*Std_Delta holds, add the last data X in the window. WW Add an empty impact event data list Pulse = [X WW ], on this basis, continue to slide the temporary data list TMP backward for the K+1th time to obtain a new window data list TMP = [X 2+K ,X 3+K ,…,X WW+K+1 ], calculate the variance of the new window data, and determine whether the variance of the window data deviates from the SS times standard deviation of the mean of the variance list of the comparison background noise signal. If |TMP_Std-Std_Mean|>SS*Std_Delta again holds true in the K+1th sliding, then X WW+K+1 Continue to add the impact event data list Pulse = [X WW , X WW+K+1], while continuing to slide the temporary data list TMP backward; if the equation |TMP_Std-Std_Mean|>SS*Std_Delta does not hold in the K+1th slide, reset the flag Flag indicating whether the impact event has occurred to False, and simultaneously calculate the characteristics of the impact event signal in the impact event data list Pulse (impact energy Energy, number of crossings P, and impact event length Q), and add them to the impact signal characteristic list, then reset the impact event data list Pulse to empty. Re-enter step (6) and continue to slide the temporary data list TMP backward;

[0173] (8) If |TMP_Std-Std_Mean|>SS*Std_Delta does not hold, set X WW+K Add to the end of the comparison noise sequence L, and pop out the first data in the original comparison noise sequence L, complete the update of the comparison noise list, and calculate the mean variance and standard deviation of the variance of the new comparison noise sequence L according to step (5);

[0174] (9) Repeat steps (6) to (8) until the temporary data list TMP slides to the end of the signal list, and returns the impact signal feature list (impact energy Energy, number of crossings P, and impact event length Q).

[0175] a. For an impulse event signal sequence Pulse=[P1,P2,…P S ], the calculation of the impact signal energy Energy is shown in formula (1-12).

[0176]

[0177] Where, Energy represents the energy value of a single impact event, P i Represents an element in the shock event sequence, and S represents the length of the shock event sequence.

[0178] From the acoustic emission signal Signal=[X1,X2,…,X i ,…] calculated in M ​​impact events (the calculation method of M is shown in the calculation method of the number of impact event characteristics below) is M_Enenrgy=[Energy1,Energy2,…,Energy M], in order to avoid the deviation of energy value calculation of a single impact event, M_Enenrgy is plotted into a histogram using the method of step (7) of the acoustic emission signal noise floor effective value meter, and the energy of the impact event with the highest distribution density is used as the energy value Enenrgy of the acoustic emission signal impact event. According to this method, the impact event energy Enenrgy_Base of the baseline acoustic emission signal is obtained, and the relative impact event energy change Enenrgy_Relative is calculated according to formula (1-13).

[0179] Energy_Relative=(Energy-Energy_Base) / Energy_Base (1-13)

[0180] RMSlow_Relative represents the relative acoustic emission signal impact event energy, Energy_Base represents the baseline acoustic emission signal impact event energy, and Energy represents the real-time acoustic emission signal impact event energy.

[0181] b. For an impulse event signal sequence Pulse=[P1,P2,…P S ], the calculation of the number of impulse signal event crossings P is shown in formula (1-14):

[0182] P=Z+X (1-14)

[0183] Among them, Z represents the local maximum value, X represents the local minimum value, and the acoustic emission signal Signal=[X1,X2,…,X i ,…], the number of crossings of the M impact events calculated is M_P=[Z1,Z2,…,Z M ], Similarly, M_P is plotted into a histogram, and the number of impact event crossings with the highest distribution density is taken as the number of impact event crossings P of the acoustic emission signal. Based on this method, the number of impact event crossings P_Base of the baseline acoustic emission signal is obtained, and the relative number of impact event crossings P_Relative is calculated according to formula (1-15).

[0184] P_Relative=(P-P_Base) / P_Base (1-15)

[0185] P_Relative represents the number of relative acoustic emission signal impact event crossings, P_Base represents the number of baseline acoustic emission signal impact event crossings, and P represents the number of real-time acoustic emission signal impact event crossings.

[0186] c. For an impulse event signal sequence Pulse=[P1,P2,…P S ], the calculation of the impulse signal event length Q is shown in formula (1-16)

[0187] Q=len(Pulse) (1-16)

[0188] Len represents the length of the calculated data list, Pulse represents the data list of an impulse signal event, and Q represents the length of an impulse event.

[0189] That is, Q represents the length of a shock event signal sequence, from the acoustic emission signal Signal=[X1,X2,…,X i ,…], the event length of the M impact events is M_Q=[S1,S2,…,S M ], similarly, M_Q is plotted as a histogram, and the length of the impact event with the highest distribution density is taken as the length Q of the impact event of the acoustic emission signal. Based on this method, the impact event length Q_Base of the baseline acoustic emission signal is obtained, and the relative impact event length Q_Relative is calculated according to formula (1-17).

[0190] Q_Relative=(Q-Q_Base) / Q_Base (1-17)

[0191] Q_Relative represents the relative acoustic emission signal impact event signal length, Q_Base represents the baseline acoustic emission signal impact event signal length, and Q represents the real-time acoustic emission signal impact event signal length.

[0192] d. From the acoustic emission signal Signal = [X1, X2, ..., X i ,…] sequence, the total number of impact events eventN=len(M_Q) (or len(M_P), len(M_Enenrgy)) can be counted, and the total number of impact events eventN_base of the baseline acoustic emission signal can be obtained, and the relative change value of the total number of impact events eventN_Relative can be calculated according to formula (1-18).

[0193] eventN_Relavive=(eventN-eventN_Base) / eventN_Base (1-18)

[0194] eventN_Relavive represents the number of relative acoustic emission signal impact events, eventN_Base represents the number of baseline acoustic emission signal impact events, and eventN represents the number of real-time acoustic emission signal impact events.

[0195] In step 2, the baseline feature threshold is calculated as follows:

[0196] (1) Using the characteristic calculation method shown above, calculate the reference acoustic emission signal Signal_Base = [Y1, Y2, ..., Y i ,…Y N ] are the effective value of the acoustic emission signal RMS_Base, the effective value of the background noise RMSNoise_Base, the effective value of the high-frequency energy RMSHigh_Base, the effective value of the low-frequency energy RMSlow_Base, the impact event energy Enenrgy_Base, the number of impact event crossings P_Base, the impact event length Q_Base, and the number of impact events eventN_base.

[0197] (2) Using the data list with a window length of WD, the reference acoustic emission signal Signal_Base = [Y1, Y2, ..., Y i ,…Y N ] list, and calculate the 8 features of the data in the window according to the above feature calculation method, and form the 8 feature lists of the reference acoustic emission signal by sliding the window, which are as follows:

[0198] RMSs=[RMS1,RMS2,…,RMS (N-WD) ] (1-19)

[0199] RMSNoises=[Noise1, Noise2,…, Noise (N-WD) ] (1-20)

[0200] RMSHighs=[RMSHigh1, RMSHigh2,…, RMSHigh (N-WD) ] (1-21)

[0201] RMSlows=[RMSlow1, RMSlow2,…, RMSlow (N-WD) ] (1-22)

[0202] Enenrgys=[Enenrgy1, Enenrgy2,…, Enenrgy (N-WD) ] (1-23)

[0203] Ps=[P1, P2,…, P (N-WD) ] (1-24)

[0204] Qs=[Q1, Q2,…, Q (N-WD) ] (1-25)

[0205] eventNumbers=[ventNumber1, ventNumber2,…, ventNumber(N-WD) ] (1-26)

[0206] It should be noted that since the length of the window WD is less than the reference data length N, when calculating the total number of impact events, the number of impact events in the window ventN must be calculated. K Converted to the number of impact events relative to the entire benchmark data length Number K , for the number of impacts ventN in each window shown in formula (1-26) K , convert using formula (1-27) and round up.

[0207]

[0208] Retrieve a list of impact event counts relative to the entire baseline data length

[0209] Numbers=[Number1, Number2,…, Number (N-WD) ] (1-28)

[0210] (3) For the above 8 feature lists, calculate each element Character K The relative change of Character_Relative with respect to the baseline feature Character in (1) K , the calculation formula is shown in (1-29):

[0211] Character_Relative K =(Character K -Character_Base) / Character_Base(1-29)

[0212] The relative change list of the above 8 feature lists can be obtained.

[0213] (4) Calculate the mean and standard deviation of the relative change list of 8 features, and determine the relative feature change threshold Threshold range according to formula (1-30). K represents the Kth feature, and there are 8 features in total. C is the abnormal coefficient. According to statistical principles, C is usually 3, but it can also take a value greater than 3.

[0214] Threshold_K=[Mean_K-C*Delta_K,Mean_K+C*Delta_K] (1-30)

[0215] Threshold_K represents the normal range of relative feature variation, Mean represents the mean, Delta represents the standard deviation, K represents the Kth feature, and C represents the abnormal coefficient, which indicates the degree of deviation from the feature mean. According to Gaussian distribution theory, C is usually 3 or greater.

[0216] The relative change threshold range of the reference acoustic emission signal characteristics can be determined through the above steps (1) to (4).

[0217] The above method is used to analyze the acoustic emission signal, and the effect of extracting the effective value of the background noise signal is shown in Figure 7 As shown, the shock event sequence is identified as Figure 8 As shown, the characteristic distribution of the impact signal is plotted as Figure 9 As shown, the number of impact events is compared with Figure 10 As shown, the effective value and spectrum are shown in Figure 11 As shown, the severe friction downsampling and envelope spectrum are shown in Figure 12 It can be seen that the method provided by the present invention can realize the extraction of acoustic emission signal features.

[0218] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention are intended to be covered by the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be based on the scope of protection of the claims.

Claims

1. A method for analyzing acoustic emission signal characteristics for rotating machinery condition monitoring, characterized in that: include: Step 1: Use the data of the rotating equipment after maintenance as the baseline data. The characteristics of the baseline data include the relative noise floor effective value and the relative number of impact events. Calculate the characteristics of the baseline data. The specific steps for calculating the effective value of the acoustic emission signal noise floor are as follows: (1) Set a window W of length N and set the sliding window overlap rate to ε; (2) Set up an empty eigenvalue list and continuously add eigenvalues ​​to the list through calculation; (3) Starting from the first data of the acoustic emission signal, the first N values ​​in the acoustic emission signal sequence are included in the sliding window W to form a signal sequence in the sliding window, and the characteristic value of the signal sequence in the sliding window is calculated; (4) According to the sliding window overlap rate ε, the calculation window W is slid backward along the acoustic emission signal sequence. The sliding step size is calculated by the following formula: STEP=max(Floor((1-ε)×N),1) In the formula, STEP represents the sliding distance, and Floor represents the integer rounded down. If the calculated STEP is 0, STEP is 1. (5) Slide the window according to the sliding step size to obtain a new sliding window data sequence, recalculate the eigenvalue of the sliding window data sequence, and add it to the eigenvalue list; (6) Repeat the above steps (4)-(5) until the last element of the acoustic emission signal sequence enters the sliding window, the sliding window is terminated, the corresponding eigenvalue is calculated, and it is added to the eigenvalue list to obtain the final eigenvalue list; (7) Draw a histogram of the feature list distribution. The calculation formula for the eigenvalue calculation accuracy γ is as follows: γ=(max(Charater)-min(Charater)) / M In the formula, γ represents the eigenvalue calculation accuracy, max represents the maximum value, min represents the minimum value, Charater represents the eigenvalue list, and M represents the number of histogram horizontal coordinate grids; (8) Obtain the characteristic value of the acoustic emission noise background signal according to the horizontal coordinate of the first highest point of the feature list distribution histogram; (9) Select the length N of the window W; (10) Select the sliding window overlap ratio ε; (11) After the rotating equipment is installed, various characteristic values ​​of the background noise signal, including the effective value, are calculated based on the selected sliding window length N and sliding window overlap ratio ε; (12) The relative acoustic emission signal noise floor effective value is calculated by the following formula: RMSNoise_Relative=(RMSNoise-RMSNoise_Base) / RMSNoise_Base RMSNoise_Relative represents the relative noise floor effective value of the acoustic emission signal, RMSNoise_Base represents the reference signal noise floor effective value, and RMSNoise represents the real-time noise floor effective value of the acoustic emission signal; The number of relative impact events reflects the mutual contact of the friction pairs. During normal operation, the relative movement between the friction pairs should be fully lubricated. When the lubrication is insufficient and the direct contact between the friction pairs is aggravated, the number of impact events per unit time increases. Step 2: Calculate the threshold of the baseline feature. Calculate the features of the normal operating data, obtain the mean and standard deviation of the features, and use the value that exceeds K times the mean as the threshold for feature abnormality. Step 3: During real-time monitoring, a periodic trigger acquisition method is used to collect acoustic emission data for N seconds each time, and the corresponding characteristic indicators are calculated. At the same time, the cached data of S*N seconds is stored in the cache queue; Step 4: Compare the real-time characteristic index with the baseline characteristic index threshold. If the characteristic index exceeds the threshold, further clarify the reason why the acoustic emission signal exceeds the threshold, and save the cached raw data in the form of a file; If the feature index does not exceed the threshold, only the real-time features are stored in the database.

2. The acoustic emission signal characteristic analysis method for rotating machinery state monitoring according to claim 1, characterized in that: The benchmark data features also include relative effective value, relative high-frequency energy, and relative low-frequency energy.

3. The acoustic emission signal characteristic analysis method for rotating machinery state monitoring according to claim 1, characterized in that: pass Calculate the effective value of the acoustic emission signal and use the following formula to calculate the relative effective value of the acoustic emission compared with the benchmark data: RMS_Relative=(RMS-RMS_Base) / RMS_Base In the formula, RMS represents the effective value of the real-time monitored acoustic emission signal, X i It represents the acoustic emission signal sequence Signal=[X1,X2,…X N ], N represents the length of the signal sequence, RMS_Relative represents the relative sound emission effective value, and RMS_Base represents the reference sound emission signal effective value.

4. The acoustic emission signal characteristic analysis method for rotating machinery state monitoring according to claim 1, characterized in that: Acoustic emission signal sequence Signal=[X1,X2,…X N The calculation steps of the noise floor eigenvalue are as follows: Set an empty eigenvalue sequence and set a sliding window with a length of N to slide in the acoustic emission signal sequence according to the step size STEP until the last data of the acoustic emission signal sequence is included in the window; Calculate the eigenvalue of the signal within the window and add the eigenvalue to the defined eigenvalue sequence. Repeat the above process until the eigenvalue calculation of all windows is completed. Draw a distribution histogram of the eigenvalue sequence and obtain the abscissa value at the first peak in the distribution histogram, which is the eigenvalue of the acoustic emission signal background noise.

5. The acoustic emission signal characteristic analysis method for rotating machinery state monitoring according to claim 1, characterized in that: The calculation steps of high-frequency signal energy are as follows: The real-time acoustic emission signal sequence and the reference acoustic emission signal sequence are band-pass filtered by using a band-pass filter method, and the frequency retention range is 100K-350K. The real-time acoustic emission signal sequence and the reference acoustic emission signal sequence after filtering; The following formula is used to calculate the effective value of the high-frequency energy of the reference acoustic emission signal and the effective value of the high-frequency energy of the real-time acoustic emission signal, and the relative change in the effective value of the high-frequency energy is calculated: RMSHigh_Relative=(RMSHigh-RMSHigh_Base) / RMSHigh_Base RMSHigh_Relative represents the relative high-frequency energy of the acoustic emission signal, RMSHigh_base represents the high-frequency energy of the reference signal, and RMSHigh represents the high-frequency energy of the real-time acoustic emission signal.

6. The acoustic emission signal characteristic analysis method for rotating machinery state monitoring according to claim 1, characterized in that: The calculation formula for the relative change in the effective value of the low-frequency energy signal is as follows: RMSlow_Relative=(RMSlow-RMSlow_Base) / RMSlow_Base RMSlow_Relative represents the relative low-frequency energy of the acoustic emission signal, RMSlow_base represents the low-frequency energy of the reference signal, and RMSlow represents the low-frequency energy of the real-time acoustic emission signal.

7. The acoustic emission signal characteristic analysis method for rotating machinery condition monitoring according to claim 1, characterized in that: The steps for extracting the shock event signal sequence are as follows: (1) Set an empty comparison noise sequence with an initial length of L, and add the first value X1 in the acoustic emission signal sequence to the comparison noise sequence L to form a new comparison noise sequence L1, and calculate the average value L of the comparison noise sequence L1 according to the following formula 1_ave and standard deviation L 1_delta : Where y i are the elements in L1. At this time, except for y1, all other elements are 0; (2) Analyze the second value X2 in the acoustic emission signal sequence and use the following formula to determine whether X2 exceeds K times the standard deviation of the mean of the comparison background noise sequence: |X2-L 1_ave |>K*L 1_delta Where K is a constant, which indicates the degree to which X2 exceeds the mean of the noise sequence compared with the original. (3) If the formula in (2) is valid, do not add X2 to L1, continue to calculate the third value X3 in the signal sequence, and repeat the above steps; (4) If the formula in (2) does not hold, add X2 to L1 to form a new comparison noise sequence L2, and calculate the average value L of the new noise sequence L2 according to the above formula 2_ave and standard deviation L 2_delta , continue to calculate the third value X3 in the signal sequence and repeat the above steps; (5) Repeat steps (1)-(4) until the background noise sequence L is completely filled to form L N On this basis, a sliding window with a window length of WL is set and slid on the noise sequence L with a step size of 1 to obtain the data within the window. The variance of the data within the window is calculated and added to the comparison background noise variance list. The above process is repeated until the background noise sequence L is traversed to form the final comparison background noise variance list. The mean and standard deviation of the list are calculated to complete the initialization of the comparison background noise sequence. (6) Set the temporary data list TMP with a window length of WW, and initialize the flag Flag of whether the impact event occurs to False. Slide the temporary data list TMP along the acoustic emission signal sequence. Take the Kth slide as an example to obtain TMP=[X 1+K ,X 2+K ,…,X WW+K ], calculate the variance of the data in the window, and determine whether the variance of the data in the window deviates from the SS times standard deviation of the mean of the comparison background noise variance list, as follows: |TMP_Std-Std_Mean|>SS*Std_Delta Where SS is a constant, which indicates the degree to which the variance of the data in the window deviates from the mean of the variance of the contrast background noise; (7) If |TMP_Std-Std_Mean|>SS*Std_Delta holds, add the last data X in the window. WW Add an empty impact event data list Pulse = [X WW ], on this basis, continue to slide the temporary data list TMP backward for the K+1th time to obtain a new window data list TMP = [X 2+K ,X 3+K ,…,X WW+K+1 ], calculate the variance of the new window data, and determine whether the variance of the window data deviates from the SS times standard deviation of the mean of the variance list of the comparison background noise signal. If |TMP_Std-Std_Mean|>SS*Std_Delta again holds true in the K+1th sliding, then X WW+K+1 Continue to add the impact event data list Pulse = [X WW， X WW+K+1 ], and continue to slide the temporary data list TMP backward; if the formula |TMP_Std-Std_Mean|>SS*Std_Delta does not hold in the K+1th slide, reset the flag Flag indicating whether the impact event occurs to False, and calculate the characteristics of the impact event signal in the impact event data list Pulse, and add it to the impact signal characteristic list, and then reset the impact event data list Pulse to empty; re-enter step (6) and continue to slide the temporary data list TMP backward; (8) If |TMP_Std-Std_Mean|>SS*Std_Delta does not hold, set X WW+K Add to the end of the comparison noise sequence L, and pop out the first data in the original comparison noise sequence L to complete the update of the comparison noise list, and calculate the mean variance and standard deviation of the variance of the new comparison noise sequence L according to the above formula; (9) Repeat steps (6)-(8) until the temporary data list TMP slides to the end of the signal list and returns to the impact signal feature list.

8. The acoustic emission signal characteristic analysis method for rotating machinery state monitoring according to claim 7, characterized in that: For an impulse event signal sequence Pulse=[P1,P2,…P S ], the calculation formula of the impulse signal energy is as follows: Where, Energy represents the energy value of a single impact event, P i Represents an element in the shock event sequence, and S represents the length of the shock event sequence.

9. The acoustic emission signal characteristic analysis method for rotating machinery state monitoring according to claim 1, characterized in that: In step 2, the threshold of the baseline feature is calculated as follows: Step 2.1: Calculate the characteristics of the reference acoustic emission signal; Step 2.2: Use the data list with a window length of WD to generate the reference acoustic emission signal Signal_Base = [Y1, Y2, ..., Y i ,…Y N ] list, calculate the 8 features of the data in the window, and form an 8-feature list of the reference acoustic emission signal by sliding the window; Step 2.3: Calculate the relative change of each feature list element compared to the baseline feature using the following formula: Character_Relative K =(Character K -Character_Base) / Character_Base In the formula, Character_Relative K Represents the relative eigenvalue of the Kth element in each feature list, Character K Represents the Kth element in the feature list, and Character_Base represents the baseline feature value of the feature; Step 2.4: Calculate the mean and standard deviation of the eight feature relative change lists and determine the relative feature change threshold range according to the following formula: Threshold_K=[Mean_K-C*Delta_K, Mean_K+C*Delta_K] Threshold_K represents the normal range of relative feature variation, Mean represents the mean, Delta represents the standard deviation, K represents the Kth feature, and C represents the abnormal coefficient.

Images