Signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation
By using the MS-Dynp inflection point detection algorithm and AW-GGS signal segmentation method in mechanical fault diagnosis, the problem of difficulty in identifying the fault occurrence point and low signal segmentation efficiency is solved, and more efficient fault diagnosis is achieved.
Patent Information
- Application Number
- CN202411540896.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-31
- Publication Date
- 2025-02-07
AI Technical Summary
The fault points of mechanical components are difficult to identify, and the fault signals are variable and the timing length is too large, which affects the efficiency of signal feature extraction and signal segmentation in the later stage.
Multi-statistic dynamic programming inflection point detection algorithm (MS-Dynp) is used to accurately identify the fault occurrence point, and the fault signal is efficiently segmented through the adaptive window greedy Gaussian segmentation (AW-GGS) algorithm.
It improves the accuracy of identification of fault occurrence points and the efficiency of signal segmentation, and enhances the accuracy and reliability of subsequent fault diagnosis.
Smart Images
Figure CN119397196A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of signal processing, and in particular relates to a signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation. Background Art
[0002] Fault diagnosis of key components (such as bearings, gears, etc.) in mechanical equipment can predict mechanical failures and extend the service life of the machinery. Usually, the vibration signal of the mechanical component throughout its life cycle is collected by a directional sensor, and this signal is presented in a continuous time series format. On the one hand, subsequent fault diagnosis needs to determine the fault occurrence point and divide the fault signal. However, the vibration signal is messy and has many disturbances, making it difficult to identify the fault occurrence point. On the other hand, the fault signal is changeable and the time series length is too large, which will seriously affect the efficiency of the use of technologies such as signal feature extraction in the later stage, and the fault signal needs to be reasonably segmented. Therefore, the present invention proposes a signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation to solve the problem of difficulty in identifying the fault occurrence point of mechanical components and signal segmentation. The dynamic programming inflection point detection algorithm (Dynamic Programming for Change Points Detection, Dynp) is used to detect the signal inflection point. However, when processing complex mode changes, the Dynp algorithm has deviations. Therefore, the present invention proposes a multi-statistic dynamic programming inflection point detection algorithm (MS-Dynp) to accurately identify the fault occurrence point. In addition, Greedy Gaussian Segmentation (GGS) segments signals according to signal distribution, but GGS needs to calculate the parameters of multivariate Gaussian distribution and evaluate the fitting effect at each segmentation, which results in limited performance of GGS when processing large-scale signals. Therefore, the present invention proposes an adaptive window greedy Gaussian segmentation (AW-GGS) algorithm to reduce the computational complexity of the original GGS algorithm and achieve efficient signal segmentation. The present invention can accurately and efficiently process signals, and further provide theoretical support and decision-making basis for mechanical fault diagnosis. Summary of the invention
[0003] The purpose of the present invention is to provide a signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation, which is helpful to improve the quality of fault data sets and provide a theoretical basis and technical guidance for mechanical fault diagnosis.
[0004] The technical solution adopted by the present invention is a signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation, and its specific steps are as follows:
[0005] Step 1: Select a mechanical component (such as a bearing, a gear, etc.) that needs to be diagnosed for a fault, and use one or more directional sensors (such as a horizontal sensor, a vertical sensor) to record the vibration signal of the selected component throughout its life cycle;
[0006] Step 2, calculating the multivariate statistical sequence of the vibration signal collected by each sensor in step 1, with the goal of minimizing the statistical sequence error, calculating the inflection point of the statistical sequence, and determining the failure point of the selected component;
[0007] Step 3: According to the fault occurrence point determined in step 2, the fault signal of the selected component is divided. According to the signal fluctuation, an adaptive window is set to preliminarily divide the fault signal to obtain β preliminary division windows;
[0008] Step 4: Based on the signal distribution, maximize the objective function ξ r As the criterion, the preliminary segmentation window obtained in step 3 is re-segmented to obtain θ final segmentation windows, and the processing of the vibration signal of the selected component is completed, including detecting the fault occurrence point and segmenting the fault signal.
[0009] The present invention is also characterized in that:
[0010] Step 1 is as follows:
[0011] Select the mechanical parts that need to be diagnosed (such as bearings, gears, etc.), fix one or more direction sensors (such as horizontal and vertical directions) on the selected parts, record the number of sensors as m, and the fixed sampling time as T. The sensor collects the vibration signal of the selected parts in batches according to the sampling time. Suppose the signal amount collected each time is p, and the number of sampling times is q, then the total length of the vibration signal is n = pq. The matrix representation of the vibration signal is:
[0012]
[0013] Among them, X i ∈R m , x ji (j=1,2,...,m,i=1,2,...,n) represents the sampling value of the jth sensor at time i.
[0014] Step 2 is as follows:
[0015] Step 2.1: Calculate the multivariate statistics of the vibration signal collected in step 1, including the variance RMS Peak-to-Peak And the square root amplitude in, It represents the signal sampled by the jth sensor for the tth time. The calculation formulas of various statistics are as follows:
[0016]
[0017] Step 2.2: The multivariate statistics of the vibration signal are obtained from step 2.1. The matrix expression of the multivariate statistics sequence of the vibration signal corresponding to the jth sensor is:
[0018]
[0019] in, represents the variance sequence, represents the RMS sequence, represents the peak-to-peak sequence, represents the square root amplitude sequence;
[0020] Step 2.3: Calculate the minimum mean square error The turning point The turning point The turning point The turning point The calculation formulas are as follows:
[0021]
[0022] Combination Determine whether the jth sensor is performing the The selected component fails during sampling. in, represents the set of multivariate statistical sequences of the jth sensor;
[0023] Step 2.4: In order to reflect early signs of failure and prevent potential major damage, the failure occurrence point of the selected component is d = (bkp-1)p+1, where
[0024] Step 3 is as follows:
[0025] Step 3.1, divide the fault signal into the signal after the fault occurrence point d (including d), and the matrix of the fault signal is expressed as:
[0026]
[0027] Among them, Y v ∈R m ,y jv (j=1,2,..,m,v=1,2,...,u) represents the value of the vth fault signal collected by the jth sensor, u=n-(d-1)p, represents the length of the fault signal;
[0028] Step 3.2: Use the average standard deviation σ of the fault signal as the threshold for judging the signal fluctuation. The calculation formula of σ is as follows:
[0029]
[0030] Design an adaptive window to segment fault data. The adaptive window includes an initial window w0, a large window w1, and a small window w2, w2<w0<w1, where w0 is used to evaluate the fluctuation size of the signal to be segmented, w1 is used to capture more signal features, and w2 is used to capture the detailed features of the signal;
[0031] Step 3.3: Get the fault occurrence point of the selected component from step 2.4, and start segmenting the fault signal from the fault occurrence point:
[0032] First, set up the first window The window value α1=w0, calculate The standard deviation of the mean Compare σ1 with σ. If σ1 < σ, update α1 = w1. Otherwise, update α1 = w2.
[0033] Then, the next window The window value α2 = w0, calculate The standard deviation of the mean And compare it with σ to update α2;
[0034] And so on, until the entire fault signal is traversed.
[0035] Divide the fault signal into β segments of different sizes of initial segmentation windows The initial segmentation point set is α = {α1, .., α r ,....α β-1}.
[0036] Step 4 is as follows:
[0037] Step 4.1: The preliminary segmentation window of the β segment is obtained from step 3.3. The matrix of the rth preliminary segmentation window is expressed as:
[0038]
[0039] in, represents the e-th fault signal collected by the j-th sensor in the r-th preliminary segmentation window;
[0040] Step 4.2 can be divided into K+1 segments by K segmentation points. Let the segmentation point set make represent The f r Segments, Each timestamp in All follow Gaussian distribution By constructing the objective function ξ r , optimize the split point ξ r The calculation formula is:
[0041]
[0042] Where, Cr = -(mα r / 2)(log(2π)+1), represents a constant, represents the empirical covariance of segment i, δ is the regularization parameter and δ ≥ 0, and we have:
[0043] Step 4.3: By maximizing the objective function ξ r , for the window Re-segmentation.
[0044] First, initialize Traversal Each sampling point in in Display Window The sampling starting point, Display Window The sampling end point is selected according to formula (15) so that Maximum sampling point make Will Add to collection λ r ,renew
[0045] Then, traverse the interval and Each sampling point calculate and Select so that Maximum sampling point Make Maximum sampling point like make On the contrary, Will Add to collection λ r ,renew
[0046] Finally, when When the The redivision.
[0047] The above process results in θ final segmentation windows, θ = (K + 1) * β. The segmentation point set obtained in step 4.3 is
[0048] Combined with step 3.3, the set of all segmentation points is:
[0049]
[0050] Finally, the processing of the vibration signals of the selected components is completed, including the detection of the fault occurrence point and the segmentation of the fault signal, which further improves the accuracy and reliability of subsequent fault diagnosis work.
[0051] The beneficial effect of the present invention is that a signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation is proposed. In order to improve the accuracy of mechanical component fault diagnosis, the present invention combines a multivariate statistical sequence on the basis of the Dynp algorithm to reduce the error of the Dynp algorithm and accurately identify the occurrence point of mechanical component faults. In order to obtain a fault signal of appropriate length, the present invention sets an adaptive window on the basis of GGS to perform preliminary segmentation of the fault signal, solves the problem of limited performance of GGS when processing large-scale signals, and realizes efficient signal segmentation. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] Figure 1 It is an overall flow chart of the signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation of the present invention;
[0053] Figure 2 It is an example of the signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation of the present invention. The mechanical bearing vibration signal collected by the horizontal direction sensor;
[0054] Figure 3 It is a mechanical bearing vibration signal collected by a vertical direction sensor in an example of a signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation of the present invention;
[0055] Figure 4 is the MS-Dynp inflection point detection result of an example mechanical bearing of the signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation of the present invention;
[0056] Figure 5 It is the preliminary segmentation result of the mechanical bearing fault signal by the adaptive window of the signal processing method embodiment based on MS-Dynp inflection point detection and AW-GGS segmentation of the present invention;
[0057] Figure 6 This is an example of the signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation of the present invention, and the result of AW-GGS re-segmentation of the mechanical bearing fault signal. DETAILED DESCRIPTION
[0058] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0059] In the process of fault diagnosis of mechanical components, vibration signals of the whole life cycle of mechanical components are usually collected by different directional sensors. Such signals are presented in the form of continuous time series. In order to ensure the accuracy of fault diagnosis, the fault point of the mechanical component needs to be correctly located. In addition, since the fault signal is changeable and the time series length is too long, the fault signal should be reasonably segmented. The two-step operation of determining the fault point and segmenting the fault signal is classified as signal processing before fault diagnosis. The signal processing process is as follows: First, the mechanical component (such as bearings, gears, etc.) that needs to be fault diagnosed is selected, and one or more directional sensors (such as horizontal sensors, vertical sensors) record the vibration signal of the selected component throughout its life cycle; secondly, the multivariate statistical sequence of the vibration signal collected by each sensor in step 1 is calculated, and the inflection point of the statistical sequence is calculated with the goal of minimizing the statistical sequence error, and the fault point of the selected component is determined; then, according to the fault point determined in step 2, the fault signal of the selected component is divided. According to the signal fluctuation, an adaptive window is set to preliminarily segment the fault signal to obtain β preliminary segmentation windows; finally, according to the signal distribution, the objective function ξ is maximized. r As the criterion, the preliminary segmentation window obtained in step 3 is re-segmented to obtain θ final segmentation windows, and the processing of the vibration signal of the selected component is completed, including detecting the fault occurrence point and segmenting the fault signal.
[0060] The present invention is based on the signal processing method of MS-Dynp inflection point detection and AW-GGS segmentation, combined with Figure 1 , follow the steps below:
[0061] Step 1: Select a mechanical component (such as a bearing, a gear, etc.) that needs to be diagnosed for a fault, and use one or more directional sensors (such as a horizontal sensor, a vertical sensor) to record the vibration signal of the selected component throughout its life cycle;
[0062] Step 1 is as follows:
[0063] Select the mechanical parts that need to be diagnosed (such as bearings, gears, etc.), fix one or more direction sensors (such as horizontal and vertical directions) on the selected parts, record the number of sensors as m, and the fixed sampling time as T. The sensor collects the vibration signal of the selected parts in the whole life cycle in batches according to the sampling time. Suppose the signal amount collected each time is p, and the number of sampling times is q, then the total length of the vibration signal is n=pq, and the matrix of the vibration signal is expressed as:
[0064]
[0065] Among them, X i∈R m , x ji represents the sampling value of the jth sensor at time i, j=1,2,...,m,i=1,2,...,n.
[0066] Step 2, calculating the multivariate statistical sequence of the vibration signal collected by each sensor in step 1, with the goal of minimizing the statistical sequence error, calculating the inflection point of the statistical sequence, and determining the failure point of the selected component;
[0067] Step 2 is as follows:
[0068] Step 2.1: Calculate the multivariate statistics of the vibration signal collected in step 1, including the variance RMS Peak-to-Peak And the square root amplitude in, represents the signal sampled by the j-th sensor at the t-th time, j = 1, 2, ..., m, t = 1, 2, ..., q, and the calculation formulas of each statistic are as follows:
[0069]
[0070]
[0071] Step 2.2: The multivariate statistics of the vibration signal are obtained from step 2.1. The matrix expression of the multivariate statistics sequence of the vibration signal corresponding to the jth sensor is:
[0072]
[0073] in, represents the variance sequence, represents the RMS sequence, represents the peak-to-peak sequence, represents the square root amplitude sequence;
[0074] Step 2.3: Calculate the minimum mean square error The turning point The turning point The turning point The turning point The calculation formulas are as follows:
[0075]
[0076] Combination Determine whether the jth sensor is performing the The selected component fails during sampling. in, represents the set of multivariate statistical sequences of the jth sensor;
[0077] Step 2.4: In order to reflect early signs of failure and prevent potential major damage, the failure occurrence point of the selected component is d = (bkp-1)p+1, where
[0078] Step 3: According to the fault occurrence point determined in step 2, the fault signal of the selected component is divided, and according to the signal fluctuation, an adaptive window is set to preliminarily divide the fault signal to obtain β preliminary division windows;
[0079] Step 3 is as follows:
[0080] Step 3.1, divide the fault signal into the signal after the fault occurrence point d (including d), and the matrix of the fault signal is expressed as:
[0081]
[0082] Among them, Y v ∈R m ,y jv (j=1,2,..,m,v=1,2,...,u) represents the value of the vth fault signal collected by the jth sensor, u=n-(d-1)p, represents the length of the fault signal;
[0083] Step 3.2: Use the average standard deviation σ of the fault signal as the threshold for judging the signal fluctuation. The calculation formula of σ is as follows:
[0084]
[0085] Design an adaptive window to segment fault data. The adaptive window includes an initial window w0, a large window w1, and a small window w2, w2<w0<w1, where w0 is used to evaluate the fluctuation size of the signal to be segmented, w1 is used to capture more signal features, and w2 is used to capture the detailed features of the signal;
[0086] Step 3.3: Get the fault occurrence point of the selected component from step 2.4, and start segmenting the fault signal from the fault occurrence point:
[0087] First, set up the first window The window value α1=w0, calculate The standard deviation of the mean Compare σ1 with σ. If σ1 < σ, update α1 = w1. Otherwise, update α1 = w2.
[0088] Then, the next window The window value α2 = w0, calculate The standard deviation of the mean And compare it with σ to update α2;
[0089] And so on, until the entire fault signal is traversed;
[0090] Divide the fault signal into β segments of different sizes of initial segmentation windows The initial segmentation point set is α = {α1, .., α r ,....α β-1}.
[0091] Step 4: Based on the signal distribution, maximize the objective function ξ r As the criterion, the preliminary segmentation window obtained in step 3 is re-segmented to obtain θ final segmentation windows, and the processing of the vibration signal of the selected component is completed, including detecting the fault occurrence point and segmenting the fault signal.
[0092] Step 4 is as follows:
[0093] Step 4.1: The preliminary segmentation window of the β segment is obtained from step 3.3. The matrix of the rth preliminary segmentation window is expressed as:
[0094]
[0095] in, represents the e-th fault signal collected by the j-th sensor in the r-th preliminary segmentation window;
[0096] Step 4.2 can be divided into K+1 segments by K segmentation points. Let the segmentation point set make represent The f r Segments, Each timestamp in All follow Gaussian distribution By constructing the objective function ξ r , optimize the split point ξ r The calculation formula is:
[0097]
[0098] Among them, C r =-(mα r / 2)(log(2π)+1), represents a constant, represents the empirical covariance of segment i, δ is the regularization parameter and δ ≥ 0, and we have:
[0099] Step 4.3: By maximizing the objective function ξ r , for the window Re-segmentation is performed, r = 1, 2, ..., β.
[0100] Step 4.3 is implemented as follows:
[0101] First, initialize Traversal Each sampling point in in Display Window The sampling starting point, Display Window The sampling end point is selected according to formula (15) so that Maximum sampling point make Will Add to collection λ r ,renew
[0102] Then, traverse the interval and Each sampling point calculate and Select so that Maximum sampling point Make Maximum sampling point like make On the contrary, Will Add to collection λ r ,renew
[0103] Finally, when When the The redivision.
[0104] The above process results in θ final segmentation windows, θ = (K + 1) * β. The segmentation point set obtained in step 4.3 is
[0105] Combined with step 3.3, the set of all segmentation points is:
[0106]
[0107] Finally, the processing of the vibration signals of the selected components is completed, including the detection of the fault occurrence point and the segmentation of the fault signal, which further improves the accuracy and reliability of subsequent fault diagnosis work.
[0108] like Figure 4As shown, compared with the traditional method of a single indicator, the present invention combines multivariate statistics with the Dynp inflection point detection method to more accurately identify the failure point of the bearing component. Figure 5 and Figure 6 As shown, the present invention can reasonably segment the signal according to the signal distribution, and this segmentation method improves the accuracy and reliability of subsequent diagnosis work. In addition, when the deep learning method is used to enhance the fault signal sample, the training sample obtained by the present invention can better capture the signal characteristics, thereby improving the quality of the generated sample.
[0109] Example 1
[0110] The present invention is based on the signal processing method of MS-Dynp inflection point detection and AW-GGS segmentation, combined with Figure 1 , follow the steps below:
[0111] Step 1: Select mechanical parts such as bearings and gears that need to be diagnosed, and use one or more directional sensors such as horizontal sensors and vertical sensors to record the vibration signals of the selected parts throughout their life cycle;
[0112] Step 2, calculating the multivariate statistical sequence of the vibration signal collected by each sensor in step 1, with the goal of minimizing the statistical sequence error, calculating the inflection point of the statistical sequence, and determining the failure point of the selected component;
[0113] Step 3: According to the fault occurrence point determined in step 2, the fault signal of the selected component is divided, and according to the signal fluctuation, an adaptive window is set to preliminarily divide the fault signal to obtain β preliminary division windows;
[0114] Step 4: Based on the signal distribution, maximize the objective function ξ r As the criterion, the preliminary segmentation window obtained in step 3 is re-segmented to obtain θ final segmentation windows, and the processing of the vibration signal of the selected component is completed, including detecting the fault occurrence point and segmenting the fault signal.
[0115] Example 2
[0116] The present invention is based on the signal processing method of MS-Dynp inflection point detection and AW-GGS segmentation, combined with Figure 1 , follow the steps below:
[0117] Step 1: Select a mechanical component (such as a bearing, a gear, etc.) that needs to be diagnosed for a fault, and use one or more directional sensors (such as a horizontal sensor, a vertical sensor) to record the vibration signal of the selected component throughout its life cycle;
[0118] Step 2, calculating the multivariate statistical sequence of the vibration signal collected by each sensor in step 1, with the goal of minimizing the statistical sequence error, calculating the inflection point of the statistical sequence, and determining the failure point of the selected component;
[0119] Step 2 is as follows:
[0120] Step 2.1: Calculate the multivariate statistics of the vibration signal collected in step 1, including the variance RMS Peak-to-Peak And the square root amplitude in, represents the signal sampled by the j-th sensor at the t-th time, j = 1, 2, ..., m, t = 1, 2, ..., q, and the calculation formulas of each statistic are as follows:
[0121]
[0122] Step 2.2: The multivariate statistics of the vibration signal are obtained from step 2.1. The matrix expression of the multivariate statistics sequence of the vibration signal corresponding to the jth sensor is:
[0123]
[0124] in, represents the variance sequence, represents the RMS sequence, represents the peak-to-peak sequence, represents the square root amplitude sequence;
[0125] Step 2.3: Calculate the minimum mean square error The turning point The turning point The turning point The turning point The calculation formulas are as follows:
[0126]
[0127] Combination Determine whether the jth sensor is performing the The selected component fails during sampling. in, represents the set of multivariate statistical sequences of the jth sensor;
[0128] Step 2.4: In order to reflect early signs of failure and prevent potential major damage, the failure occurrence point of the selected component is d = (bkp-1)p+1, where
[0129] Step 3: According to the fault occurrence point determined in step 2, the fault signal of the selected component is divided, and according to the signal fluctuation, an adaptive window is set to preliminarily divide the fault signal to obtain β preliminary division windows;
[0130] Step 3 is as follows:
[0131] Step 3.1, divide the fault signal into the signal after the fault occurrence point d (including d), and the matrix of the fault signal is expressed as:
[0132]
[0133] Among them, Y v ∈R m ,y jv (j=1,2,..,m,v=1,2,...,u) represents the value of the vth fault signal collected by the jth sensor, u=n-(d-1)p, represents the length of the fault signal;
[0134] Step 3.2: Use the average standard deviation σ of the fault signal as the threshold for judging the signal fluctuation. The calculation formula of σ is as follows:
[0135]
[0136] Design an adaptive window to segment fault data. The adaptive window includes an initial window w0, a large window w1, and a small window w2, w2<w0<w1, where w0 is used to evaluate the fluctuation size of the signal to be segmented, w1 is used to capture more signal features, and w2 is used to capture the detailed features of the signal;
[0137] Step 3.3: Get the fault occurrence point of the selected component from step 2.4, and start segmenting the fault signal from the fault occurrence point:
[0138] First, set up the first window The window value α1=w0, calculate The standard deviation of the mean Compare σ1 with σ. If σ1 < σ, update α1 = w1. Otherwise, update α1 = w2.
[0139] Then, the next window The window value α2 = w0, calculate The standard deviation of the mean And compare it with σ to update α2;
[0140] And so on, until the entire fault signal is traversed;
[0141] Divide the fault signal into β segments of different sizes of initial segmentation windows The initial segmentation point set is α = {α1, .., αr ,....α β-1}.
[0142] Step 4: Based on the signal distribution, maximize the objective function ξ r As the criterion, the preliminary segmentation window obtained in step 3 is re-segmented to obtain θ final segmentation windows, and the processing of the vibration signal of the selected component is completed, including detecting the fault occurrence point and segmenting the fault signal.
[0143] Step 4 is as follows:
[0144] Step 4.1: The preliminary segmentation window of the β segment is obtained from step 3.3. The matrix of the rth preliminary segmentation window is expressed as:
[0145]
[0146] in, represents the e-th fault signal collected by the j-th sensor in the r-th preliminary segmentation window;
[0147] Step 4.2 can be divided into K+1 segments by K segmentation points. Let the segmentation point set make represent The f r Segments, Each timestamp in All follow Gaussian distribution By constructing the objective function ξ r , optimize the split point ξ r The calculation formula is:
[0148]
[0149] Among them, C r =-(mα r / 2)(log(2π)+1), represents a constant, represents the empirical covariance of segment i, δ is the regularization parameter and δ ≥ 0, and we have:
[0150] Step 4.3: By maximizing the objective function ξ r , for the window Re-segmentation is performed, r = 1, 2, ..., β.
[0151] Finally, the processing of the vibration signals of the selected components is completed, including the detection of the fault occurrence point and the segmentation of the fault signal, which further improves the accuracy and reliability of subsequent fault diagnosis work.
[0152] Example 3
[0153] In order to verify the feasibility of the present invention, the present invention is further explained in combination with the embodiments and the drawings in the specification. The vibration signal of the mechanical component selected this time is the Xi'an Jiaotong University bearing data set (XJTU-SY). XJTU-SY contains the full life cycle vibration signals of 15 rolling bearings under 3 working conditions. Two direction sensors are used to record the full life cycle vibration signals of the bearings in the horizontal and vertical directions respectively. In XJTU-SY, the failure causes of the bearings include inner ring wear, cage breakage, outer ring wear and outer ring cracking. The present invention selects bearing 2_1 as an example to explain the effectiveness of the invention. The failure cause of bearing 2_1 is inner ring wear. Its full life cycle vibration signals in the horizontal and vertical directions are as follows: Figure 2 , Figure 3 shown.
[0154] Each sampling time is T = 1.28s, the signal quantity of each sampling is p = 32768, the sampling times of bearing 2_1 is q = 491, and the total length of the collected signal is n = pq = 16089088. First, the fault occurrence point of the signal is selected, and the statistical sequence inflection point of the horizontal and vertical sensors is calculated with the goal of minimizing the statistical sequence error, as shown in Figure 4 As shown, the inflection point of the statistical sequence of the horizontal sensor is B1 = {465, 465, 468, 465}, and the inflection point of the statistical sequence of the vertical sensor is B2 = {468, 468, 471, 471}. It means that the bearing fails at the 465th sampling in the horizontal direction and at the 467th sampling in the vertical direction. Combining the signals in the two directions, it is determined that the bearing fails at the 465th sampling. The fault point of the bearing is d = (bkp-1)p+1 = 15204353. Since the fault signal sequence divided is too long, the 480th sampling signal is selected and segmented. The threshold of the selected signal is σ = 3.403, and the adaptive window is set to w0 = 6830, w1 = 5464, w2 = 8196. The fault signal is initially segmented with the goal of minimizing the statistical sequence error. The preliminary segmentation results are shown in the figure. Figure 5 As shown in the figure, the signal is divided into five initial segmentation windows, and the initial segmentation point set α = {8196, 13660, 19124, 24588}. According to the signal distribution, the initial segmentation signal is re-segmented with the maximization of the objective function as the criterion. The final segmentation result is as follows Figure 6As shown, the set of all segmentation points is fg={1901,2883,6051,7050,9119,10196,11892,12975,14590,16045,17124,18130,19920,20901,22664,23630,25767,27502,29913}. From the segmentation results, it can be seen that the present invention can effectively capture the characteristics of the signal and realize the reasonable segmentation of the fault signal, which lays a foundation for the subsequent fault diagnosis work and helps to improve the accuracy and efficiency of the diagnosis.
Claims
1. A signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation, characterized in that: Follow the steps below to implement it: Step 1: Select a mechanical component that needs to be diagnosed for a fault, and use one or more directional sensors to record the vibration signal of the selected component throughout its life cycle; Step 2, calculating the multivariate statistical sequence of the vibration signal collected by each sensor in step 1, with the goal of minimizing the statistical sequence error, calculating the inflection point of the statistical sequence, and determining the failure point of the selected component; Step 3: Divide the fault signal of the selected component, set an adaptive window according to the signal fluctuation, preliminarily segment the fault signal, and obtain β preliminary segmentation windows; Step 4: re-segment the preliminary segmentation window obtained in step 3 to obtain θ final segmentation windows, and complete the processing of the vibration signal of the selected component, including detecting the fault occurrence point and segmenting the fault signal.
2. The signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation according to claim 1, characterized in that: The step 1 is specifically as follows: Select the mechanical parts that need to be diagnosed for faults, fix one or more directional sensors on the selected parts, record the number of sensors as m, and the fixed sampling time as T. The sensors collect the vibration signals of the selected parts in the whole life cycle in batches according to the sampling time. Suppose the signal amount collected each time is p, and the number of sampling times is q, then the total length of the vibration signal is n=pq, and the matrix of the vibration signal is expressed as: Among them, X i ∈R m , x ji represents the sampling value of the jth sensor at time i, j=1,2,...,m,i=1,2,...,n.
3. The signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation according to claim 2, characterized in that: The step 2 is specifically as follows: Step 2.1: Calculate the multivariate statistics of the vibration signal collected in step 1, including the variance RMS Peak-to-Peak And the square root amplitude in, represents the signal sampled by the j-th sensor at the t-th time, j = 1, 2, ..., m, t = 1, 2, ..., q, and the calculation formulas of each statistic are as follows: Step 2.2: The multivariate statistics of the vibration signal are obtained from step 2.
1. The matrix expression of the multivariate statistics sequence of the vibration signal corresponding to the jth sensor is: in, represents the variance sequence, represents the RMS sequence, represents the peak-to-peak sequence, represents the square root amplitude sequence; Step 2.3: Calculate the minimum mean square error The turning point The turning point The turning point The turning point The calculation formulas are as follows: Combination Determine whether the jth sensor is performing the The selected component fails during sampling. in, represents the set of multivariate statistical sequences of the jth sensor; Step 2.4: In order to reflect early signs of failure and prevent potential major damage, the failure occurrence point of the selected component is d = (bkp-1)p+1, where 4. The signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation according to claim 3, characterized in that: The step 3 is as follows: Step 3.1: Divide the fault signal into the signal after the fault occurs at point d. The matrix of the fault signal is expressed as: Among them, Y v ∈R m ,y jv (j=1,2,..,m,v=1,2,...,u) represents the value of the vth fault signal collected by the jth sensor, u=n-(d-1)p, represents the length of the fault signal; Step 3.2: Use the average standard deviation σ of the fault signal as the threshold for judging the signal fluctuation. The calculation formula of σ is as follows: Design an adaptive window to segment fault data. The adaptive window includes an initial window w0, a large window w1, and a small window w2, w2<w0<w1, where w0 is used to evaluate the fluctuation size of the signal to be segmented, w1 is used to capture more signal features, and w2 is used to capture the detailed features of the signal; Step 3.3: Get the fault occurrence point of the selected component from step 2.4, and start segmenting the fault signal from the fault occurrence point: First, set up the first window The window value α1=w0, calculate The standard deviation of the mean Compare σ1 with σ. If σ1 < σ, update α1 = w1. Otherwise, update α1 = w2. Then, the next window The window value α2 = w0, calculate The standard deviation of the mean And compare it with σ to update α2; And so on, until the entire fault signal is traversed; Divide the fault signal into β segments of different sizes of initial segmentation windows The initial segmentation point set is α = {α1, .., α r ,....α β-1 }.
5. The signal processing method based on MS-Dynp inflection point detection and AW-GGS segmentation according to claim 4, characterized in that: The step 4 is specifically as follows: Step 4.1: The preliminary segmentation window of the β segment is obtained from step 3.
3. The matrix of the rth preliminary segmentation window is expressed as: in, represents the e-th fault signal collected by the j-th sensor in the r-th preliminary segmentation window; Step 4.2 can be divided into K+1 segments by K segmentation points. Let the segmentation point set make represent The f r Segments, Each timestamp in All follow Gaussian distribution By constructing the objective function ξ r , optimize the split point ξ r The calculation formula is: Among them, C r =-(mα r / 2)(log(2π)+1), represents a constant, represents the empirical covariance of segment i, δ is the regularization parameter and δ ≥ 0, and we have: Step 4.3: By maximizing the objective function ξ r , for the window Re-segmentation is performed, r = 1, 2, ..., β.
6. The signal processing method based on MS-Dynp inflection point detection and AW-GGS signal segmentation according to claim 5, characterized in that: The step 4.3 is specifically implemented according to the following steps: First, initialize Traversal Each sampling point in in Display Window The sampling starting point, Display Window The sampling end point is selected according to formula (15) so that Maximum sampling point make Will Add to collection λ r ,renew Then, traverse the interval and Each sampling point calculate and Select so that Maximum sampling point Make Maximum sampling point like make On the contrary, Will Add to collection λ r ,renew Finally, when When the the redivision of The above process results in θ final segmentation windows, θ = (K + 1) * β. The segmentation point set obtained in step 4.3 is Combined with step 3.3, the set of all segmentation points is: Finally, the processing of the vibration signals of the selected components is completed, including the detection of the fault occurrence point and the segmentation of the fault signal, which further improves the accuracy and reliability of subsequent fault diagnosis work.