Fault feature extraction method based on entropy theory

Abstract

The invention belongs to the technical field of rotating mechanical equipment fault detection, and particularly relates to a fault feature extraction method based on an entropy theory, which comprises the following steps: collecting vibration signals of a bearing in different health states, and segmenting the vibration signals into data samples with equal length; a fine time shifting multi-scale coarse graining method is adopted to generate a sequence under each scale, and the integrity of a signal structure is kept; extracting a local maximum value and a local minimum value in each section of signal, constructing a key mode set and recording the position of the key mode set, calculating probability distribution of key mode intervals under each scale, and thus obtaining a corresponding attention entropy value; and finally, fusing the attention entropies under the scales to form a feature set representing multiple scales. According to the method, a strategy of combining fine time shifting and coarse graining is introduced, the problem of instability of traditional multi-scale entropy in a high-scale space is solved, and the consistency and integrity of feature extraction results under all scales are effectively guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of rotating mechanical equipment fault detection, and in particular relates to a fault feature extraction method based on entropy theory. Background Art

[0002] Rolling bearings are key components in high-speed train axleboxes, gearboxes, and motor systems, and their operating status is directly related to the safety and reliability of the entire vehicle. Bearings are prone to various faults during operation due to factors such as material fatigue, lubrication failure, and assembly deviations. Fault signatures from bearing vibration signals are difficult to extract, and due to the complex operating environment of key components, they are easily obscured by strong background noise, posing a significant challenge to bearing fault feature extraction and diagnosis.

[0003] In existing research, the methods for extracting rolling bearing fault features mainly include:

[0004] Time-domain analysis methods, such as mean, variance, kurtosis, and impulse factor, can simply and intuitively reflect the amplitude characteristics and impact conditions of vibration signals over time and are widely used in fault diagnosis. Frequency-domain analysis methods reveal fault characteristic frequencies through Fourier transforms. However, traditional time-domain and frequency-domain statistics are easily affected by changes in speed and load. They are generally applicable to steady-state bearing conditions and are not suitable for precise quantitative analysis. They are prone to misjudgments or omissions, affecting diagnostic accuracy. Time-frequency domain analysis methods, such as short-time Fourier transform (STFT), continuous wavelet transform (CWT), and Wigner-Ville distribution (WVD), can capture instantaneous fault characteristics in non-stationary signals by simultaneously considering both time and frequency characteristics. However, these methods are easily interfered with by strong background noise or complex operating conditions, resulting in leakage or distortion of fault characteristics, which affects diagnostic effectiveness.

[0005] In recent years, entropy theory has been introduced to the field of mechanical fault feature extraction as an effective tool for measuring time series complexity. Methods such as sample entropy (SE), permutation entropy (PE), fuzzy entropy (FE), and dispersion entropy (DE) have improved the robustness of feature extraction to a certain extent. SE and FE assess complexity by measuring the similarity between data samples, but require setting a similarity tolerance parameter and are highly sensitive to these parameters. FE, by encoding relative size order, can quickly quantify local feature changes, but its ability to identify strong local impact signals is limited. DE measures complexity from the perspective of distribution patterns, but is highly sensitive to the sequence length of data samples and exhibits significant volatility when processing short-duration signals. Furthermore, the traditional multiscale entropy (MSE) method employs a coarse-graining strategy, resulting in a sharp decrease in the number of sequence samples as the scale factor increases. This results in increased volatility in high-scale entropy values ​​and loss of feature information, severely impacting fault signal extraction.

[0006] In summary, the existing feature extraction methods based on entropy theory still have problems such as strong parameter dependence and poor multi-scale feature stability when processing fault signals of key components such as rolling bearings of high-speed train bogies. Summary of the Invention

[0007] In order to solve the technical problems that the fault features of the above-mentioned key components and equipment such as rolling bearings are difficult to extract and have strong parameter dependence under strong background noise and non-stationary environments, the present invention provides a fault feature extraction method based on entropy theory to provide reliable support for fault diagnosis under complex working conditions.

[0008] The present invention provides a fault feature extraction method based on entropy theory, comprising the following steps:

[0009] Step S1: Signal acquisition and segmentation preprocessing:

[0010] First, the vibration signals of key components are collected in real time under different health conditions. The collected original vibration signals are then divided into several non-overlapping data samples of a fixed length, and the length of each sample is a pre-set fixed value.

[0011] Step S2, fine time shift coarse graining process:

[0012] For time series , x i is the signal data at the i-th time point, and the scale factor is , the signal is segmented by gradually moving the window, which is multi-scale coarse-graining;

[0013] Furthermore, the specific process is expressed as:

[0014]

[0015] in, is the scale factor, For scale The coarse-grained sequence, N is the length of the original time series;

[0016] Step S3, attention entropy calculation:

[0017] First, define the key pattern of the signal, determine the extreme points of the signal, extract the local maximum and minimum values ​​in each sample, and record their positions; use the extreme points to form a key pattern set, record the key pattern position, and use this as the basis for attention entropy calculation. Suppose the signal is , extract all local extreme values ​​and form a key pattern set as , where each is the position index of the local extreme value in the original sequence.

[0018] For each coarse-grained sequence, the intervals between key patterns are calculated, and the frequencies of occurrence of each interval are counted to obtain the corresponding probability distribution; based on the statistical laws of key pattern intervals, the probability distribution of each coarse-grained sequence is calculated. Attention entropy , The scale factor is After fine time shift and coarse grain processing sequence, get The entropy value of the coarse-grained sequence is calculated, and the function of attention entropy and scale factor is formed as follows:

[0019]

[0020] Among them, AE （τ） For scale The attention entropy value under For scale Next The probability of occurrence of an interval, τ max is the maximum scale factor;

[0021] Based on the above formula, the fine time-shifted multi-scale attention entropy is defined as RTSMAE(x,τ), and the formula is:

[0022]

[0023] Where, The scale factor is After fine time shift and coarse grain processing A sequence, is the maximum scale factor, is the original time series signal;

[0024] Step S4: Multi-scale entropy feature fusion:

[0025] The attention entropy obtained at different scales is combined until Reaching the maximum scale factor ,get Collection As follows:

[0026]

[0027] The above-mentioned fine time-shifted multi-scale attention entropy method is used for calculation to extract feature information reflecting the complexity of signals at different scales.

[0028] Beneficial effects of the present invention:

[0029] The entropy-based fault feature extraction method described in this paper collects vibration signals under different operating conditions and processes them in segments. A multi-scale coarse-graining strategy is then introduced to generate a coarse-grained sequence. Local extreme points in each signal segment are extracted to form a set of key patterns. Key pattern intervals are then counted, a probability distribution is constructed, and attention entropy is calculated. The attention entropy values ​​at each scale are then integrated to form a feature vector, comprehensively reflecting the structural changes in the signal. This method demonstrates significant advantages in feature extraction and is effective for vibration signal processing and analysis tasks under complex operating conditions and in high-noise environments. Compared with traditional feature extraction methods, this method significantly enhances the robustness and reliability of fault diagnosis results, making it particularly suitable for complex industrial environments such as high-speed trains. Furthermore, the entire feature extraction process requires no hyperparameters, avoiding repeated parameter adjustments due to equipment differences or operating conditions, enhancing the method's versatility and ease of use. Regarding multi-scale analysis, this method combines fine time shifting with coarse-graining to overcome the instability of traditional multi-scale entropy methods at high scales, effectively ensuring the consistency and integrity of feature extraction results at all scales. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is the overall process of the fault feature extraction method based on entropy theory described in the present invention;

[0031] Figure 2 The white noise and Noise time domain curve;

[0032] Figure 3 The white noise and RTSMAE, CMAE, and MAE entropy curves of noise at different signal lengths;

[0033] Figure 4 The different entropy methods described in this invention are used to analyze white noise and Error bar graph of noise;

[0034] Figure 5 This is the signal waveform diagram of different fault types of the CWRU bearing data set described in the present invention;

[0035] Figure 6 These are the visualization results of different feature extraction methods described in the present invention. DETAILED DESCRIPTION

[0036] The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings. The embodiments described are only some of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0037] Example 1

[0038] This embodiment provides a fault feature extraction method based on entropy theory, including the following steps:

[0039] Step S1: Signal acquisition and segmentation preprocessing:

[0040] First, the bearing vibration signal is collected in real time under different health conditions. The collected original vibration signal is then divided into several non-overlapping data samples of a fixed length, and the length of each sample is a pre-set fixed value.

[0041] Step S2, fine time shift coarse graining process:

[0042] For time series , x i is the signal data at the i-th time point, and the scale factor is , the signal is segmented by gradually moving the window, that is, multi-scale coarse-graining, and the fine time-shift coarse-graining method is used to replace the traditional non-overlapping windowing method. The process is expressed as:

[0043]

[0044] in, is the scale factor, For scale The coarse-grained sequence, N is the length of the original time series;

[0045] Time series generated by fine time shift and coarse graining method In scale The length of In the traditional multi-scale analysis method, the length of the coarse-grained sequence is (Indicates not greater than (maximum positive integer), the data length is much shorter than that of the RTSMAE method. This demonstrates that the proposed method considers the connection between the data before and after the breakpoint, thereby reducing errors and avoiding the loss of key information. The proposed method obtains coarse-grained sequences at different scales, preserving signal continuity and important information to the greatest extent possible.

[0046] Step S3, attention entropy calculation:

[0047] First, define the key pattern of the signal and determine the extreme points of the signal. Extract the local maximum and minimum values ​​in each sample segment and record their locations. Extract all local extreme values ​​and obtain: , where each is the position index of the local extreme value in the original sequence;

[0048] For each coarse-grained sequence, calculate the intervals between key patterns and the probability distribution of key pattern intervals; based on the statistical laws of key pattern intervals, calculate the probability distribution of each coarse-grained sequence. Attention entropy , The scale factor is After fine time shift and coarse grain processing sequence, get The entropy value of the coarse-grained sequence is calculated, and the function of attention entropy and scale factor is formed as follows:

[0049]

[0050] Among them, AE （τ） For scale The attention entropy value under For scale Next The probability of occurrence of an interval, τ max is the maximum scale factor;

[0051] Based on the above formula, the fine time-shifted multi-scale attention entropy is defined as RTSMAE(x,τ), and the formula is:

[0052]

[0053] Where, The scale factor is After fine time shift and coarse grain processing A sequence, is the maximum scale factor, is the original time series signal;

[0054] Step S4: Multi-scale entropy feature fusion:

[0055] Repeat steps S2 and S3 until Reaching the maximum scale factor ,get Collection As follows:

[0056]

[0057] The obtained multi-scale vector is .

[0058] Example 2

[0059] On the basis of Example 1, in order to further illustrate the effectiveness of the method of the present invention, experimental verification is carried out using a simulated signal and a real bearing vibration signal.

[0060] 1. Verification and analysis based on simulation signals:

[0061] In this experiment, the commonly used white noise and Noise is used to evaluate the feature extraction performance of the RTSMAE proposed in this invention, and its signal waveform is as follows Figure 2 As shown. Two multi-scale entropy methods, composite multi-scale attention entropy (CMAE) and multi-scale attention entropy (MAE), are introduced to compare and analyze with RTSMAE. By analyzing the entropy value distribution under different time series signal lengths, their feature extraction capabilities are obtained. The white noise and The entropy values ​​of the two simulation signals are calculated by three methods, and the entropy value curves under different signal lengths are obtained as follows: Figure 3 shown.

[0062] Depend on Figure 3 It can be seen that among the results of the three multi-scale attention entropy, when the signal length is 500, white noise and The RTSMAE, CMAE and MAE curves of noise all show violent fluctuations, indicating that the entropy value is unstable. When the signal length is greater than 500, RTSMAE shows better stability than CMAE and MAE. The entropy curve trends of signals of different lengths are basically the same, and the entropy value volatility is small. This is mainly because RTSMAE adopts a new coarse-grained processing method, which retains the key patterns in the signal to the greatest extent and performs averaging processing, which can reduce the error of entropy values ​​under different signal lengths. The other two entropy methods show irregular and large fluctuations with the changes in scale factor and data length. The main reason for this phenomenon is that the increase in scale factor makes the length of coarse-grained sequence gradually smaller, and AE shows instability in short time series. When the signal length is greater than 500, the curve of RTSMAE shows a slow upward trend with the change of scale factor, but the numerical fluctuation range is small, considering the computational cost and entropy value characteristics; in this experiment, the data length is selected as 3000 and the scale factor is 200. The maximum value is 20. It is worth noting that When the scale factor is large, the sequence still maintains a length of 2981 points, while the traditional coarse-graining method only has 150 points, with a length advantage of 19.9 times, which significantly improves the entropy stability when the scale factor is large.

[0063] In order to illustrate the practicability and stability of the method of the present invention, this experiment selected 50 groups of white noise with a length of 3000 and The noise signal is analyzed; and its nonlinear analysis performance and feature extraction ability are evaluated by comparing with composite multiscale attention entropy (CMAE), multiscale attention entropy (MAE), multiscale sample entropy (MSE), multiscale permutation entropy (MPE), multiscale scatter entropy (MDE) and multiscale fuzzy entropy (MFE).

[0064] The parameters of the entropy methods introduced in this experiment are set as follows: Each method includes a scaling factor set to 20. CMAE and MAE are similar to RTSMAE and do not require hyperparameters. MSE, MPE, MDE, and MFE are all set to their respective optimal parameters. The error bar plots (mean and standard deviation) are calculated using different entropy methods for two different signals. Figure 4 shown.

[0065] from Figure 4 It can be seen that the results of different entropy methods for the two groups of signals are different. Figure 4 (a) Display The RTSMAE value of the signal is greater than that of the white noise signal. Among the seven multiscale entropy methods, only the RTSMAE and MAE curves do not overlap, and the RTSMAE standard deviation is smaller. The RTSMAE difference between the two signal groups is more obvious, indicating that RTSMAE can distinguish the two noise signals with a small error and a more stable entropy value. The error bars of the CMAE, MSE, MPE, MDE, and MFE curves overlap to varying degrees, indicating that they cannot effectively distinguish the two noise signals. Therefore, the RTSMAE feature extraction method can distinguish the two signals with better stability.

[0066] 2. Verification and analysis based on real rolling bearing signals:

[0067] This experiment uses rolling bearing health and failure data from Case Western Reserve University (CWRU) in the United States. The rolling bearing model in the CWRU data set is SKF6205. The experimental sampling frequency is 12kHz and the speed is 1797r / min. A total of six bearing status data types are selected. Each data item is randomly divided into 40 groups of non-repeating samples (sample length is 3000). Detailed information about the experimental data samples is shown in Table 1. The time domain waveform of the data set is shown in the following figure. Figure 5 (a) shows the corresponding amplitude spectrum. Figure 5 (b)

[0068] Depend on Figure 5 It can be seen that when a rolling bearing vibration signal fails, the time domain waveform exhibits significant periodic impact characteristics. However, relying solely on time domain or frequency domain analysis often makes it difficult to accurately determine the health of the bearing. Therefore, the proposed RTSMAE is used to extract fault features from the vibration signal.

[0069] Table 1 CWRU bearing health status dataset

[0070]

[0071] In order to verify the effectiveness of the refined time-shifted multi-scale attention entropy (RTSMAE) method, this experiment uses six common multi-scale entropy methods, including composite multi-scale attention entropy (CMAE), multi-scale attention entropy (MAE), multi-scale sample entropy (MSE), multi-scale permutation entropy (MPE), multi-scale scattered entropy (MDE), and multi-scale fuzzy entropy (MFE), as comparison methods to compare with RTSMAE. Seven multi-scale entropy methods were used to calculate the entropy values ​​of 240 samples of 6 types of bearing health status data in the CWRU bearing fault dataset. The t-SNE algorithm was used to reduce the dimensionality of the 20-scale features extracted by multi-scale entropy and the features extracted by single-scale AE, and the dataset was visualized in two dimensions. Figure 6 It should be noted that the clustering ability of the algorithm represents the feature extraction ability of the method: the smaller the intra-class distance between samples in the same cluster, and the larger the inter-class distance between different clusters, the better the feature extraction ability.

[0072] Depend on Figure 6 As shown in (a) and (g), for the six-category classification problem, RTSMAE and MFE extract the best features, effectively distinguishing between different types of sample features. Features extracted by other methods all suffer from varying degrees of aliasing. The figures also show that compared to traditional entropy methods, RTSMAE, CMAE, and MAE, which extend attention entropy to multi-scale space, generally have better feature extraction capabilities. Figure 6 This shows that RTSMAE can be used as a method for fault feature extraction and has good extraction performance.

[0073] In addition, the solution time of different feature extraction methods is analyzed in Table 2.

[0074] Table 2 Solution time of different entropy methods

[0075]

[0076] As can be seen in Table 2, RTSMAE takes 116.63 seconds to extract features from data of different fault types, while MFE requires a higher time cost, requiring 21,120 seconds, or 5.87 hours, to extract features from the dataset. Although CMAE and MAE, which are similar to RTSMAE, have higher solution efficiency, they do so at the expense of key information in the time series. While CMAE, MPE, and AE exhibit high computational efficiency, their feature extraction performance is inferior to RTSMAE. A comprehensive analysis of the visual feature extraction results and computational efficiency shows that the RTSMAE proposed in this paper is a fast and effective feature extraction tool.

Claims

1. A fault feature extraction method based on entropy theory, characterized by: The following steps are involved: Step S1: Signal acquisition and segmentation preprocessing: First, the vibration signals of key components are collected in real time under different health conditions. The collected original vibration signals are then divided into several non-overlapping data samples of a fixed length, and the length of each sample is a pre-set fixed value. Step S2, fine time shift coarse graining process: For time series , x i is the signal data at the i-th time point, and the scale factor is , the signal is segmented by gradually moving the window, which is multi-scale coarse-graining; Step S3, attention entropy calculation: First, define the key pattern of the signal, determine the extreme points of the signal, extract the local maximum and minimum values ​​in each sample, and record their positions; use the extreme points to form a key pattern set, record the key pattern position, and use this as the basis for attention entropy calculation. Suppose the signal is , extract all local extreme values ​​and form a key pattern set as , where each is the position index of the local extreme value in the original sequence; For each coarse-grained sequence, the intervals between key patterns are calculated, and the frequencies of occurrence of each interval are counted to obtain the corresponding probability distribution; based on the statistical laws of key pattern intervals, the probability distribution of each coarse-grained sequence is calculated. Attention entropy , The scale factor is After fine time shift and coarse grain processing sequence, get The entropy value of the coarse-grained sequence is calculated, and the function of attention entropy and scale factor is formed as follows: ， Among them, AE （τ） For scale The attention entropy value under For scale Next The probability of occurrence of an interval, τ max is the maximum scale factor; Based on the above formula, the fine time-shifted multi-scale attention entropy is defined as RTSMAE(x,τ), and the formula is: ， Where, The scale factor is After fine time shift and coarse grain processing A sequence, is the maximum scale factor, is the original time series signal; Step S4: Multi-scale entropy feature fusion: The attention entropy obtained at different scales is combined until Reaching the maximum scale factor ,get Collection As follows: 。 2. The fault feature extraction method based on entropy theory according to claim 1, characterized in that: The fine time shift coarse graining process in step S2 is expressed as: ， in, is the scale factor, For scale The coarse-grained sequence is , and N is the length of the original time series.

Images