Rotating machine fault feature extraction method and system

Abstract

The invention discloses a rotating machinery fault feature extraction method and system, and relates to the technical field of mechanical equipment fault diagnosis, and the method comprises the steps: obtaining an original time sequence signal, determining the number of segmented subsequences of the original time sequence signal, and converting the original time sequence signal into coarse grain sequences under different numbers of subsequences according to a preset rule; and for the coarse grain sequence under each subsequence number, calculating a corresponding Boltzmann-Shannon interaction entropy, carrying out average value calculation on entropy values of the Boltzmann-Shannon interaction entropies, and taking the average value as a final entropy value. According to the method, the sequence corresponding to each scale factor is further refined, the mode that only one coarse grain sequence is generated is changed into the mode that multiple coarse grain sequences are generated, and then the MBSIE mean value of the multiple sequences is calculated to serve as the final entropy value, so that the obtained fault information is more detailed.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical equipment fault diagnosis, and more specifically, to a method and system for extracting fault characteristics of rotating machinery. Background Art

[0002] With the development of modern industry, science and technology, and the further improvement of automation, machinery is developing towards the direction of large-scale, high-speed, continuous, centralized, and automated. Rotating mechanical equipment is in a high-speed operation state for a long time (generally more than 3,000 revolutions per minute or even up to tens of thousands of revolutions), and due to the influence of various factors, some faults will inevitably occur, and these faults often cause huge economic losses or even catastrophic consequences. Therefore, it is very important to diagnose the faults of rotating mechanical equipment.

[0003] Vibration detection and diagnosis method is the most commonly used diagnosis method for the fault diagnosis technology of rotating mechanical equipment nowadays. Vibration signals contain rich mechanical operation state information. Therefore, the vibration detection and diagnosis method can accurately diagnose most fault types in mechanical equipment. Currently, the vibration analysis methods adopted in engineering applications mainly include time-domain analysis method and frequency-domain analysis method. The time-domain analysis method mainly extracts the time-domain characteristics of faults in the time domain of the signal and then diagnoses the faults of mechanical equipment according to the changes of the time-domain characteristics. The frequency-domain analysis method is a technology that transforms the vibration signal to the frequency domain through Fourier transform for analysis, extracts the frequency-domain characteristics of the signal, and then diagnoses the faults of mechanical equipment according to the changes of the frequency-domain characteristics. The time-domain analysis method has simple and fast algorithms, but it can only judge whether there are faults in mechanical equipment and the severity of the faults, and cannot judge the fault location and fault type. The frequency-domain analysis method can not only determine whether there are faults but also determine the fault location and type. Therefore, the frequency-domain analysis method is mostly used for fine diagnosis. Currently, there are mainly two factors affecting the diagnosis effect of the frequency-domain analysis method: one is the non-stationarity of vibration signals caused by the non-uniform rotation speed of rotating mechanical equipment (such as the acceleration, deceleration, and parking of trains, and the rotation speed of wind turbines changing with the wind speed, etc.). If Fourier analysis is directly performed on these non-stationary signals, serious frequency ambiguity phenomena will occur, thus affecting the judgment of faults; the other is that due to the influence of environmental vibration noise and the noise generated by the vibration of other components of the equipment during the operation of rotating mechanical equipment, the collected vibration signals often contain vibration noise not caused by faults in addition to the vibration signals caused by the faults themselves. And these noises submerge the fault characteristic signals, especially the characteristic signals in the early stage of faults, making the faults unable to be detected in time and false alarms occur, delaying the best maintenance time and even causing huge economic losses and even catastrophic consequences. To solve the first and second problems, the prior art ("Gear Crack Fault Diagnosis Based on Angular Domain Averaging", Journal of North University of China (Natural Science Edition), Vol. 28, No. 1, 2007) adopts a cyclostationary signal with angular displacement (rotation angle) as the independent variable, first reconstructs the angular domain signal through order tracking technology, and then segments the signal in the angular domain and averages the segmented signals to achieve the method of suppressing noise. This method can suppress the influence of random noise, improve the signal-to-noise ratio, and highlight the fault characteristics. However, since this method suppresses noise by segmenting the data and then adding and averaging the data of different segments, it destroys the continuity of the data and cannot realize continuous state monitoring; in addition, since the long data is divided into short data, this method also reduces the frequency-domain resolution of the subsequent Fourier transform.

[0004] Therefore, how to accurately extract the fault characteristics of rotating mechanical equipment is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for extracting fault characteristics of rotating machinery. By further refining the sequence corresponding to each scale factor, it is improved from generating only one coarse-grained sequence to generating multiple coarse-grained sequences, and then calculating the mean value of MBSIE of multiple sequences as the final entropy value, so that the obtained fault information is more detailed.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] A method for extracting fault characteristics of rotating machinery, comprising:

[0008] Obtain the original time series signal, determine the number of segmented subsequences of the original time series signal, and then convert the original time series signal into coarse-grained sequences under different numbers of segmented subsequences according to a preset rule;

[0009] For the coarse-grained sequences under each number of segmented subsequences, calculate the corresponding Boltzmann-Shannon interaction entropy, calculate the average value of the entropy values of each Boltzmann-Shannon interaction entropy, and use the average value as the final entropy value.

[0010] Preferably, the converting the original time series signal into coarse-grained sequences under different numbers of segmented subsequences according to a preset rule specifically includes: for the time series signal in the bounded region given by the original time series signal X is a time series signal with a length of N, the length size ranges from 1 to N, and belongs to [a, b], where [a, b] represents the bounded region, and the coarse-grained sequence is defined The calculation method is as follows:

[0011]

[0012] Among them, 1 ≤ j ≤ N / τ, 1 ≤ k ≤ τ; τ represents the number of segmented subsequences, and the coarse-graining process changes the original time series signal into coarse-grained sequences with a length of N / τ under different scales τ, and k is the number of entropy values.

[0013] Preferably, the calculating the average value specifically includes:

[0014]

[0015] Among them, IMBSIE represents the average entropy value of the time series signal in the bounded region a and b respectively represent the upper and lower bounds of the region; p is the histogram probability distribution, q is the probability distribution based on the geometric quantile, m = (p + q) / 2; D (p||m) and D KL (p||m) and D KL (q||m) respectively represent the standard asymmetric KL divergence between two distributions at p and q.

[0016] A rotating machinery fault feature extraction system, comprising:

[0017] A signal conversion module, which acquires the original time series signal, determines the number of segmented subsequences of the original time series signal, and then converts the original time series signal into coarse-grained sequences under different numbers of segmented subsequences according to a preset rule;

[0018] A signal calculation and recognition module, which calculates the corresponding Boltzmann-Shannon interaction entropy for the coarse-grained sequences under each number of segmented subsequences, calculates the average value of the entropy values of the Boltzmann-Shannon interaction entropies, and takes the average value as the final entropy value.

[0019] Preferably, the signal conversion module specifically includes: for the time series signal in the bounded region given by the original time series signal X is a time series signal with a length of N, the length size ranges from 1 to N, and belonging to [a, b] represents a bounded region, and the coarse-grained sequence is defined The calculation method is as follows:

[0020]

[0021] Among them, 1 ≤ j ≤ N / τ, 1 ≤ k ≤ τ; τ represents the number of segmented subsequences, and the coarse-graining process changes the original time series signal into a coarse-grained sequence with a length of N / τ at different scales τ, and k is the number of entropy values.

[0022] Preferably, the average value calculation in the signal calculation and recognition module specifically includes:

[0023]

[0024] Among them, IMBSIE represents the average entropy value of the time series signal in the bounded region of, a and b respectively represent the upper and lower bounds of the region; p is the histogram probability distribution, q is the probability distribution based on the geometric quantile, m = (p + q) / 2; D KL (p||m) and D KL (q||m) respectively represent the standard asymmetric KL divergence between two distributions between p and q.

[0025] Through the above technical solutions, compared with the prior art, the present invention discloses a rotating machinery fault feature extraction method and system. By further refining the sequences corresponding to each scale factor, it is improved from only generating one coarse-grained sequence to generating multiple coarse-grained sequences, and then calculating the MBSIE mean value of multiple sequences as the final entropy value, so that the obtained fault information is more detailed. Description of the Drawings

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the provided drawings.

[0027] Figure 1 Schematic diagram of the method steps provided by the present invention;

[0028] Figure 2 Schematic diagram of the improved multi-scale entropy coarse-graining provided by the present invention;

[0029] Figure 3(a) is the RN-MBSIE distribution diagram provided by the present invention;

[0030] Figure 3(b) is the RN-IMBSIE distribution diagram provided by the present invention;

[0031] Figure 3(c) is the VN-MBSIE distribution diagram provided by the present invention;

[0032] Figure 3(d) is the VN-IMBSIE distribution diagram provided by the present invention;

[0033] Figure 4 Noise signal diagram with different signal-to-noise ratios provided by the present invention;

[0034] Figure 5(a) is the anti-noise performance diagram of IMBSIE provided by the present invention;

[0035] Figure 5(b) is the anti-noise performance diagram of MBSIE provided by the present invention;

[0036] Figure 6(a) is the error bar diagram of IMBSIE provided by the present invention;

[0037] Figure 6(b) is the mean value diagram of IMBSIE provided by the present invention;

[0038] Figure 6(c) is the error bar diagram of MBSIE provided by the present invention;

[0039] Figure 6(d) is the mean value diagram of MBSIE provided by the present invention;

[0040] Figure 7(a) is the normal state vibration signal diagram provided by the present invention;

[0041] Figure 7(b) is the rubbing state vibration signal diagram provided by the present invention;

[0042] Figure 7(c) is the unbalance state vibration signal diagram provided by the present invention;

[0043] Figure 7(d) is the misalignment state vibration signal diagram provided by the present invention;

[0044] Figure 8(a) is a comparison diagram of the separation effect of IMBSIE provided by the present invention;

[0045] Figure 8(b) is a comparison diagram of the separation effect of MBSIE provided by the present invention;

[0046] Figure 8(c) is a comparison diagram of the separation effect of IMAE provided by the present invention;

[0047] Figure 8(d) is a comparison diagram of the separation effect of MAE provided by the present invention. Detailed implementation manners

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0049] The embodiment of the present invention discloses a method for extracting fault characteristics of rotating machinery, as Figure 1 shown, including:

[0050] Obtain the original time series signal, determine the number of segmented subsequences of the original time series signal, and then convert the original time series signal into coarse-grained sequences under different numbers of segmented subsequences according to a preset rule;

[0051] For each coarse-grained sequence under different numbers of segmented subsequences, calculate the corresponding Boltzmann-Shannon interaction entropy, calculate the average value of the entropy values of each Boltzmann-Shannon interaction entropy, and use the average value as the final entropy value.

[0052] In a specific embodiment, converting the original time series signal into coarse-grained sequences at different scales according to a preset rule specifically includes: for the time series signal in the bounded region given by the original time series signal X is a time series signal with a length of N, and the length size ranges from 1 to N and belongs to [a, b], where [a, b] represents the bounded region, and the coarse-grained sequence is defined The calculation method is as follows:

[0053]

[0054] where 1 ≤ j ≤ N / τ, 1 ≤ k ≤ τ; τ represents the number of segmented subsequences, and the coarse-graining process changes the original time series signal into a coarse-grained sequence with a length of N / τ at different scales τ, and k is the number of entropy values.

[0055] In a specific embodiment, calculate the BSIE of each coarse-grained sequence, and then average the k entropy values to obtain the IMBSIE under this scale factor. The calculation method is as follows:

[0056]

[0057] Among them, IMBSIE represents the time series signal of the bounded region of the average entropy value, a and b respectively represent the upper and lower bounds of the region; p is the probability distribution of the histogram, q is the probability distribution based on the geometric quantile, and m = (p + q) / 2; D KL (p||m) and D KL (q||m) respectively represent the standard asymmetric KL divergence between the two distributions at p and q. For the sake of understanding, as Figure 2 Figure 15 shows the improved multi-scale entropy coarse-graining schematic diagram when the scale τ = 3 is given.

[0058] A rotating machinery fault feature extraction system, comprising:

[0059] A signal conversion module, which acquires the original time series signal, determines the number of segmented subsequences of the original time series signal, and then converts the original time series signal into coarse-grained sequences under different numbers of segmented subsequences according to a preset rule;

[0060] A signal calculation and recognition module, which calculates the corresponding Boltzmann-Shannon interaction entropy for the coarse-grained sequences under each number of segmented subsequences, calculates the average value of the entropy values of each Boltzmann-Shannon interaction entropy, and takes the average value as the final entropy value.

[0061] In a specific embodiment, the signal conversion module specifically includes: for the time series signal of the bounded region given by the original time series signal X is a time series signal with a length of N, the length size ranges from 1 to N, and belongs to [a, b] representing the bounded region. Define the coarse-grained sequence The calculation method is as follows:

[0062]

[0063] Among them, 1 ≤ j ≤ N / τ, 1 ≤ k ≤ τ; τ represents the number of segmented subsequences, and the coarse-graining process changes the original time series signal into a coarse-grained sequence with a length of N / τ at different scales τ, and k is the number of entropy values.

[0064] In a specific embodiment, the average value calculation in the signal calculation and recognition module specifically includes:

[0065]

[0066] Among them, IMBSIE represents the time series signal of the bounded region of the average entropy value, a and b respectively represent the upper and lower bounds of the region; p is the probability distribution of the histogram, q is the probability distribution based on the geometric quantile, and m = (p + q) / 2; D KL (p||m) and D KL (q||m) respectively represent the standard asymmetric KL divergences between the two distributions between p and q.

[0067] To verify the performance of the proposed algorithm, the present invention uses different noise categories to respectively verify the timing length stability, noise resistance, and signal recognition ability of IMBSIE. To verify the effectiveness and rationality of IMBSIE and prevent the influence of randomness, the noise Red Noise (RN) and f 2 the distribution of the multi-scale entropy of IMBSIE and MBSIE under different timing lengths (N = 1000, N = 2000, N = 3000, N = 4000, and N = 5000) of the noise Violet Noise (VN) is as Figures 3(a)-3(d) .

[0068] Obviously, with the increase of the scale factor under the two kinds of noise, the distribution differences between IMBSIE and MBSIE are relatively large. Especially, MBSIE shows relatively strong fluctuations, while IMBSIE is more stable overall. Specifically, taking the signal with length N = 2000 as an example, under the noise RN, the maximum and minimum values of IMBSIE and MBSIE are 0.573, 0.489 and 0.609, 0.515 respectively, with differences of 0.084 and 0.094 respectively. It can be seen that the stability of IMBSIE under RN is better than that of MBSIE. Under the f 2 noise VN, the maximum and minimum values of IMBSIE and MBSIE are 0.471, 0.322 and 0.505, 0.324 respectively, with differences of 0.149 and 0.181 respectively. It can be seen that the stability of IMBSIE under VN is better than that of MBSIE. In summary, the stability of IMBSIE is higher than that of MBSIE.

[0069] To verify the noise resistance of IMBSIE, the present invention analyzes the entropy value stability of IMBSIE and MBSIE for RN and White Gaussian Noise (WGN). To avoid the influence of randomness, 30 groups of RN and WGN with a timing length of 2000 are taken, and the entropy values under their different scale factors are calculated and their means are taken. The distribution of the three kinds of multi-scale entropy, namely IMFAE, IMAE, and MAE, is as Figure 4It can be seen that IMFAE is more stable and has a smaller mean value, and its performance is still better than that of IMAE and MAE. Specifically, the maximum values of RN and WGN of IMFAE, IMAE, and MAE are 0.125 and 0.033, 0.155 and 0.031, 0.178 and 0.063 respectively, and the differences are 0.092, 0.124, and 0.115 respectively. It can be seen that IMFAE still has stability for different types of noise and has good noise resistance performance.

[0070] Feature extraction of rotating machinery fault signals is vulnerable to noise interference, which requires IMBSIE, as a feature extraction tool, to have good noise resistance. In this subsection, the distribution of IMBSIE and MBSIE for simulation signals and signals with different signal-to-noise ratio noises added is analyzed, and their noise resistance is evaluated and analyzed. The noisy signal of simulated rotating machinery true low-frequency eddy current is obtained by the following formula, and the signal diagram is as Figure 4 .

[0071]

[0072] In the formula, Φ(t) represents the vibration signal of the rotating machinery containing noise; φ(t) is the interference signal with different signal-to-noise ratios. In this section, 10, 20, 30, and 40 dB are taken respectively.

[0073] The entropy value distribution of IMBSIE and MBSIE under different noise conditions is shown in Figures 5(a)-5(b) . It can be seen that the curve of IMBSIE is relatively smooth and has little fluctuation under several different signal-to-noise ratio conditions; the curve of MBSIE fluctuates violently as a whole and has obvious fluctuations under various signal-to-noise ratios, and is greatly affected by noise. Obviously, the noise resistance of IMBSIE is significantly better than that of MBSIE.

[0074] As a feature extraction method, IMBSIE needs to have good signal feature recognition ability. To compare the recognition performance of IMBSIE and MBSIE, RN and Gaussian white noise signal (WGN) are introduced in this invention to test the two feature extraction methods. The specific experimental results are as Figures 6(a)-6(d) , it can be seen that the means of the two entropies for RN and WGN feature extraction do not cross, which proves that both methods can recognize noise signals. Further, the error bars of the two noise signals of MBSIE cross at many scales, while the error bars of IMBSIE do not cross at any scale, which proves that the recognition performance of CMAE is better. The specific standard deviations and means are shown in Table 1. It can be seen that the mean and standard deviation of IMBSIE are both smaller than those of MBSIE under the two kinds of noises, further verifying that IMBSIE has better stability.

[0075] Table 1 Mean and standard deviation of different types of entropy for different signals

[0076]

[0077] In the present invention, a rotor test bench is used to simulate the normal state of the unit and three common faults of the unit, namely rubbing, unbalance, and misalignment. Among them, the set speed of the unit is 1200 r / min, the sampling frequency is 2048 HZ, 100 groups of data are collected for each of the four states of the unit, and each group of data contains 2048 vibration signals, as Figures 7(a)-7(d) shown.

[0078] To clarify the superiority of the feature extraction method proposed in the present invention, the multi-scale approximate entropy (MAE) and the improved multi-scale approximate entropy (IMAE) methods are introduced for comparison with the method of the present invention, and the extracted features are visually analyzed by means of t-SNE. The specific visualization results are as Figures 8(a)-8(d) . It can be seen that IMBSIE effectively extracts the signal features of different unit states, making the scatter points corresponding to the same state closely aggregated together and the four states well separated from each other, and there is no aliasing of the features of different states; the visualization result of MBSIE can roughly separate the four states and the feature points are relatively close, but there is stacking in individual fault states; although there is no aliasing phenomenon in IMAE, the feature distribution of the rubbing state is relatively scattered; the rubbing and unbalance states of MAE are not separated and there is an aliasing phenomenon. In summary, IMBSIE has better feature extraction ability.

[0079] Finally, to further compare the effects of the above four feature extraction methods, a classifier that optimizes the parameters of the kernel extreme learning machine by means of the whale algorithm is used to classify the above feature extraction results. The results are as follows:

[0080] Table 2 Diagnostic accuracies of different methods

[0081]

[0082] As can be seen from Table 2, under this classifier, the diagnostic accuracy of IMBSIE is 100%, and all four states of the rotor vibration test bench are correctly identified, while the accuracy of MBSIE is only 97.5%, and misjudgments occur in the misalignment and unbalance states; the recognition rate of IMAE is 98.5%, and the diagnostic rate of MAE is only 96%. It can be seen that the classification effect of IMBSIE is the best, higher than the other three methods, further proving the effectiveness of IMFAE feature extraction.

[0083] In this specification, the various embodiments are described in a progressive manner. The key point of each embodiment is to describe the differences from other embodiments. For the same or similar parts between the embodiments, reference can be made to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and reference can be made to the description of the method part for related parts.

[0084] The foregoing description of the disclosed embodiments enables those skilled in the art to practice or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for extracting fault features of rotating machinery, characterized in that: include: Acquire an original time series signal, determine the number of segmentation subsequences of the original time series signal, and then convert the original time series signal into coarse-grained sequences with different numbers of segmentation subsequences according to a preset rule; For each coarse-grained sequence with each number of segmented subsequences, the corresponding Boltzmann-Shannon interaction entropy is calculated, and the entropy values ​​of the Boltzmann-Shannon interaction entropies are averaged, and the average value is used as the final entropy value.

2. A method for extracting fault features of rotating machinery according to claim 1, characterized in that: The step of converting the original time series signal into a coarse-grained sequence with different numbers of segmented subsequences according to a preset rule specifically includes: for the time series signal of the bounded area given by the original time series signal X is a time series signal of length N, the length ranges from 1 to N, belongs to [a,b], [a,b] represents a bounded region, and defines a coarse-grained sequence The calculation is as follows: Among them, 1≤j≤N / τ, 1≤k≤τ; τ represents the number of segmentation subsequences, and the coarse-grained processing converts the original time series signal into a coarse-grained sequence of length N / τ at different scales τ, and k is the number of entropy values.

3. A method for extracting fault features of rotating machinery according to claim 2, characterized in that: The average value calculation specifically includes: Among them, IMBSIE represents the timing signal of the bounded area The average entropy value of a and b represent the upper and lower bounds of the region respectively; p is the histogram probability distribution, q is the probability distribution based on geometric quantiles, m = (p + q) / 2; D KL (p||m) and D KL (q||m) represents the standard asymmetric KL divergence between two distributions p and q, respectively.

4. A rotating machinery fault feature extraction system, using a rotating machinery fault feature extraction method according to any one of claims 1 to 3, characterized in that: include: A signal conversion module, which obtains an original time series signal, determines the number of segmentation subsequences of the original time series signal, and then converts the original time series signal into coarse-grained sequences with different numbers of segmentation subsequences according to a preset rule; The signal calculation and recognition module calculates the corresponding Boltzmann-Shannon interaction entropy for each coarse-grained sequence under the number of segmented subsequences, calculates the average entropy value of each Boltzmann-Shannon interaction entropy, and uses the average value as the final entropy value.

5. A rotating machinery fault feature extraction system according to claim 4, characterized in that: The signal conversion module specifically includes: for the time series signal of the bounded area given by the original time series signal X is a time series signal of length N, the length ranges from 1 to N, belongs to [a,b] to represent a bounded region, and defines a coarse-grained sequence The calculation is as follows: Among them, 1≤j≤N / τ, 1≤k≤τ; τ represents the number of segmentation subsequences, and the coarse-grained processing converts the original time series signal into a coarse-grained sequence of length N / τ at different scales τ, and k is the number of entropy values.

6. A rotating machinery fault feature extraction system according to claim 5, characterized in that: The signal calculation and identification module performs average value calculation specifically including: Among them, IMBSIE represents the timing signal of the bounded area The average entropy value of a and b represent the upper and lower bounds of the region respectively; p is the histogram probability distribution, q is the probability distribution based on geometric quantiles, m = (p + q) / 2; D KL (p||m) and D KL (q||m) represents the standard asymmetric KL divergence between two distributions p and q, respectively.

Images