Gearbox fault diagnosis method and system based on graph structure signal processing

Abstract

The invention relates to the technical field of mechanical fault diagnosis, in particular to a gearbox fault diagnosis method and system based on graph structure signal processing, and the method comprises the following steps: calling a vibration data sequence, detecting slope change and marking nodes, constructing edge connection and edge weight to obtain a graph structure, calculating a spectral coefficient to recognize a high-density frequency interval, and carrying out the fault diagnosis. Extracting a spectrum boundary, obtaining a gradient difference value of a spectrum edge graph, positioning a node sequence, extracting a peak value, judging a fault type, and outputting a gearbox fault diagnosis result. According to the method, graph structure reconstruction is carried out through a node slope hopping trend, an edge connection direction and a weight reference quantity are constructed through inter-node amplitudes and time intervals, the sensitivity of a graph structure to weak impact characteristics is enhanced, disturbance frequency band screening is carried out based on spectrum coefficient slope inversion frequency, and the node response degree is mapped through spectrum edge amplitude gradient differences. And establishing a node amplitude direction sequence, matching the node amplitude direction sequence with a standard direction template, judging a fault boundary and a graph structure breakpoint position, and realizing accurate identification of an early fault of the gearbox.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical fault diagnosis, and in particular to a gearbox fault diagnosis method and system based on graph structure signal processing. Background Art

[0002] The technical field of mechanical fault diagnosis aims to detect, identify and classify abnormal conditions or fault types that occur in mechanical equipment during operation. By collecting and analyzing physical signals such as vibration, sound, and current of key components of mechanical systems such as gears, bearings, and shafts during operation, it is determined whether there are abnormalities or potential faults. It adopts steps such as signal acquisition and analysis, feature extraction, state modeling and classification judgment, and combines statistical analysis, spectral analysis, time series analysis, and graph structure modeling. It is applied to scenarios such as aerospace, rail transportation, wind power generation, and manufacturing processing that have high requirements for equipment stability. It has important engineering value in ensuring the safe and stable operation of equipment and reducing maintenance costs and downtime risks. Among them, the gearbox fault diagnosis method based on graph structure signal processing refers to the gear After the box vibration signal is converted into a graph structure, the technical method of identifying different fault states inside the gearbox is carried out through signal analysis means on the graph structure, including converting the continuous vibration time series into a graphical form with a topological structure, such as constructing the connection relationship between the graph nodes and edges through the horizontal visibility graph algorithm, and determining the weight of the edge according to the numerical relationship between the nodes. Based on the constructed graph structure, the spectral coefficient distribution law in the graph Fourier transform or the total variation function of the graph is applied. By screening high-order spectral coefficients or calculating the gradient change of the graph signal, the characteristic components reflecting the impact or periodic anomaly are extracted. Specifically, the vibration signal is constructed as a weighted horizontal visibility graph, the adjacency matrix is ​​constructed, the eigenvector decomposition is performed, the graph signal is reconstructed, etc., to complete the extraction of typical fault information contained in the original vibration signal and generate the classification basis.

[0003] Traditional gearbox fault diagnosis methods rely on road maps or horizontal visibility graphs to construct the gearbox vibration signal topology structure, and define edge weights based on the numerical relationship between nodes. They do not deeply consider the slope jump trend of the sampling nodes and the dynamic amplitude change relationship between adjacent nodes. The graph structure expression is not sensitive enough to the weak periodic impact characteristics in the early fault signal, and it is easy to miss small fluctuation abnormal information. As a result, when the vibration amplitude changes slightly or the impact characteristics are not obvious in the initial stage of the fault, it is impossible to accurately distinguish the fault state, thereby reducing the accuracy of fault diagnosis. Summary of the Invention

[0004] In order to solve the technical problems existing in the prior art, the embodiments of the present invention provide a gearbox fault diagnosis method and system based on graph structure signal processing. The technical solution is as follows:

[0005] In order to achieve the above object, the present invention adopts the following technical solution, a gearbox fault diagnosis method based on graph structure signal processing, comprising the following steps:

[0006] S1: Call the vibration data sequence, analyze the slope change differences of multiple consecutive sampling points in the gearbox input end bearing vibration sequence, detect the continuous jump trend, mark the corresponding sampling points as node marks, and output the slope mutation node set;

[0007] S2: Using the slope mutation node set, construct the edge connection direction according to the amplitude change and sampling time difference between each node in the point set, and construct the edge weight reference based on the amplitude difference between the nodes and the sampling point interval to generate the graph structure weight matrix;

[0008] S3: Calculate the spectral coefficient sequence based on the graph structure weight matrix, record the number of slope reversals within a fixed window, identify the turning point concentrated segment and construct a linear trend for the segment boundary, compare the fluctuation intensity of the central area and the difference in the rate of change of the trend of the two sides, screen the frequency bands with significant disturbances, and output the high-density frequency interval;

[0009] S4: Call the high-density frequency interval, extract the symmetrically distributed spectrum boundary, and compare the average level of slope change at both ends of the edge. Combined with the corresponding node distribution area in the graph signal, calculate the degree of difference in amplitude gradient between the nodes on both sides, and map it to the graph domain response index in the graph structure to obtain the spectral edge graph gradient difference.

[0010] As a further solution of the present invention, the slope mutation node set includes the mutation inflection point position, time series index, and node amplitude change direction; the graph structure weight matrix is ​​specifically the edge connection direction set, edge amplitude weight parameter, and node time interval index; the high-density frequency interval is specifically the frequency sequence index, turning frequency band boundary, and center frequency band fluctuation label; the spectrum edge graph gradient difference includes the spectrum edge distribution area, graph node slope response intensity, and structural gradient difference interval.

[0011] As a further solution of the present invention, the step of obtaining the slope mutation node set is specifically as follows:

[0012] S101: calling a vibration data sequence, analyzing a slope change of continuous sampling points in a gearbox input end bearing vibration sequence, sequentially comparing amplitude and time information between adjacent sampling points in the sampling point sequence, establishing a continuous slope change trend, and obtaining a slope change trend sequence;

[0013] S102: comparing the change differences between a plurality of adjacent slope segments according to the slope change trend sequence, detecting a continuous jump trend, and marking corresponding sampling points as node markers to obtain a node marker set;

[0014] S103: using the node marking set and combining it with the time sequence, sorting the marking points in the node marking set, removing the unmarked non-turning points, forming a graph structure node set, and outputting a slope mutation node set.

[0015] As a further solution of the present invention, the steps for obtaining the graph structure weight matrix are specifically as follows:

[0016] S201: calling the slope mutation node set, comparing the increase and decrease trends of the amplitudes between the nodes and the time sequence according to the amplitude changes of each pair of adjacent nodes in the node set and the sampling time interval, and generating an amplitude and time change sequence;

[0017] S202: Based on the amplitude and time change sequence, the consistency of the amplitude change direction of each pair of nodes is judged, and the connection order and connection direction between the nodes are identified in combination with the node amplitude change direction, and an edge connection direction set is established;

[0018] S203: Analyze the amplitude difference and sampling point interval between nodes according to the edge connection direction set, construct an edge weight reference, output the structural connection relationship between nodes, and generate a graph structure weight matrix.

[0019] As a further solution of the present invention, the specific formula for constructing the edge weight reference is:

[0020]

[0021] Calculate edge weight reference value;

[0022] Among them, W ij is the reference value of the edge weight between node i and node j, ΔV ij is the normalized value of the vibration amplitude difference between node i and node j, t i is the normalized value of the sampling time of node i, t j is the normalized value of the sampling time of node j, γ ij is the normalized value of the slope disturbance change between node i and node j, n i is the number of node i in the graph structure, n j is the number of node j in the graph structure.

[0023] As a further solution of the present invention, the step of obtaining the high-density frequency interval is specifically as follows:

[0024] S301: calling the graph structure weight matrix, analyzing the spectral coefficient sequence corresponding to each node in the matrix, determining the number of switching times of the slope direction in the spectral coefficient sequence according to the node order, and counting the frequency of occurrence of the slope reversal phenomenon to obtain a slope reversal frequency sequence;

[0025] S302: According to the slope inversion frequency sequence, the spectral coefficients are sorted according to the frequency order, the spectral coefficient sequence is divided into windows with a fixed width, the number of inversion points is counted for each window, the turning point distribution dense area is marked, and a turning point set is obtained.

[0026] S303: The turning point set is called, a linear trend is constructed at each section boundary, the fluctuation intensity of the section center area and the change rate of the two sides of the section are compared, the disturbance significant frequency band is screened, and a high-density frequency interval is output.

[0027] As a further scheme of the present application, the step of obtaining the spectral edge graph gradient difference value is specifically:

[0028] S401: The high-density frequency interval is called, the symmetrically distributed spectral boundaries in the frequency interval are extracted, the slope changes of the left and right boundaries are counted, the average level of the slope changes on each side is calculated, and a boundary slope change mean value group is obtained.

[0029] S402: In combination with the boundary slope change mean value group, the node distribution area corresponding to the spectral boundary in the graph signal is located, the change amount of the nodes on the amplitude gradient on the left and right sides is calculated, and a node gradient difference index is obtained.

[0030] S403: Based on the node gradient difference index, the structural change strength caused by spectral disturbance is evaluated, and is mapped into a graph domain response index in the graph structure, and a spectral edge graph gradient difference value is obtained.

[0031] As a further scheme of the present application, the method further comprises:

[0032] S5: The spectral edge graph gradient difference value is obtained, the corresponding node sequence in the graph structure is located, the continuous peak values in the signal fluctuation are extracted, and a node amplitude direction sequence is established, the positive and negative change sequences are compared, the matching similarity of a plurality of training sample direction templates is compared, the fault boundary and the graph structure breakpoint are identified, the gear box fault type is judged, and a gear box fault diagnosis result is obtained.

[0033] The gear box fault diagnosis result specifically refers to the graph structure breakpoint distribution position, the fault boundary direction sequence, and the fault type identification label.

[0034] As a further scheme of the present application, the step of obtaining the gear box fault diagnosis result is specifically:

[0035] S501: The spectral edge graph gradient difference value is obtained, the node sequence corresponding to the difference value in the graph structure is located, the continuous peak values of the signal fluctuation in the node sequence are identified, the amplitudes are extracted according to the node order, the amplitude change direction between adjacent nodes is established, and a node amplitude direction sequence is generated.

[0036] S502: Based on the node amplitude direction sequence, a similarity comparison is performed between the positive and negative change sequences and a plurality of training sample direction templates. By comparing the matching degrees, the fault boundary area is identified and divided to obtain a fault boundary interval set.

[0037] The specific formula for performing similarity comparison between the positive and negative change sequences and multiple training sample direction templates is:

[0038]

[0039] Calculate the directional trend difference value;

[0040] Among them, S km is the direction trend difference value, k is the number index of the amplitude direction sequence of the node to be identified, m is the number index of the direction sequence of the training sample template, d ku is the normalized amplitude direction value of the u-th sampling point in the k-th node sequence, t mu is the normalized amplitude direction value of the u-th sampling point in the m-th template direction sequence, u is the index of the sampling point in the sequence, L k is the total number of sampling points in the k-th node sequence;

[0041] S503: Based on the fault boundary interval set, analyze the amplitude change trajectory between adjacent segments in the boundary area, identify the distribution characteristics of the graph structure breakpoints, determine the fault type of the gearbox, and obtain the gearbox fault diagnosis result.

[0042] On the other hand, a gearbox fault diagnosis system based on graph structure signal processing is provided. The system is applied to the gearbox fault diagnosis method based on graph structure signal processing. The system includes:

[0043] The slope detection module calls the vibration data sequence, analyzes the slope change differences composed of multiple consecutive sampling points in the gearbox input end bearing vibration sequence, detects the continuous jump trend, identifies the corresponding sampling points as node markers, and outputs the slope mutation node set;

[0044] The edge connection construction module uses the slope mutation node set to construct the edge connection direction according to the amplitude change and sampling time difference between each node in the point set, and constructs the edge weight reference based on the amplitude difference between the nodes and the sampling point interval to generate the graph structure weight matrix;

[0045] The spectrum segment analysis module calculates the spectrum coefficient sequence based on the graph structure weight matrix, records the number of slope reversals within a fixed window, identifies the turning point concentrated segment and constructs a linear trend for the segment boundary, compares the fluctuation intensity of the central area and the difference in the rate of change of the trend of the two sides, screens the frequency bands with significant disturbance, and outputs the high-density frequency interval;

[0046] The gradient evaluation module calls the high-density frequency interval, extracts the symmetrically distributed spectrum boundary, and compares the average level of slope change at both ends of the edge. Combined with the corresponding node distribution area in the graph signal, it calculates the difference in amplitude gradient between the nodes on both sides and maps it to the graph domain response index in the graph structure to obtain the spectral edge graph gradient difference;

[0047] The fault determination module obtains the gradient difference of the spectral edge graph, locates the corresponding node sequence in the graph structure, extracts the continuous peaks in the signal fluctuation, and establishes a node amplitude direction sequence. By comparing the matching similarity of the positive and negative change sequences and multiple training sample direction templates, the fault boundary and graph structure breakpoints are identified, the gearbox fault type is determined, and the gearbox fault diagnosis result is obtained.

[0048] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0049] The graph structure is reconstructed through the node slope jump trend, and the edge connection direction and weight reference are constructed through the amplitude and time interval between nodes. The sensitivity of the graph structure to weak impact characteristics is enhanced. The disturbance frequency band is screened based on the frequency of spectral coefficient slope inversion. The node response degree is mapped by the spectrum edge amplitude gradient difference. The node amplitude direction sequence is established and matched with the standard direction template. The fault boundary and graph structure breakpoint position are determined, and the accurate identification of early gearbox faults is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0051] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0052] Figure 2 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0053] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0054] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0055] In the embodiments of the present invention, the terms "image" and "picture" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same. The terms "of," "corresponding," and "corresponding" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same.

[0056] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0057] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0058] See also Figure 1 The present invention provides a technical solution, a gearbox fault diagnosis method based on graph structure signal processing, comprising the following steps:

[0059] S1: Call the vibration data sequence, analyze the slope change differences of multiple consecutive sampling points in the gearbox input end bearing vibration sequence, detect the continuous jump trend, mark the corresponding sampling points as node marks, and output the slope mutation node set;

[0060] S2: Using the slope mutation node set, the edge connection direction is constructed based on the amplitude change and sampling time difference between each node in the point set. Based on the amplitude difference between nodes and the sampling point interval, the edge weight reference is constructed to generate the graph structure weight matrix;

[0061] S3: Calculate the spectral coefficient sequence based on the graph structure weight matrix, record the number of slope reversals within a fixed window, identify the turning point concentrated segment and construct a linear trend for the segment boundary, compare the fluctuation intensity of the central area and the change rate difference of the trend in the two sides, screen the frequency bands with significant disturbances, and output the high-density frequency interval;

[0062] S4: Call the high-density frequency interval, extract the symmetrically distributed spectrum boundary, and compare the average level of slope change at both ends of the edge. Combined with the corresponding node distribution area in the graph signal, the difference in amplitude gradient between the nodes on both sides is calculated and mapped to the graph domain response index in the graph structure to obtain the spectral edge graph gradient difference;

[0063] S5: Obtain the gradient difference of the spectral edge graph, locate the corresponding node sequence in the graph structure, extract the continuous peaks in the signal fluctuation, and establish the node amplitude direction sequence. By comparing the matching similarity of the positive and negative change sequences and multiple training sample direction templates, the fault boundary and graph structure breakpoints are identified, the gearbox fault type is determined, and the gearbox fault diagnosis result is obtained.

[0064] The slope mutation node set includes the mutation inflection point position, time series index, and node amplitude change direction. The graph structure weight matrix specifically includes the edge connection direction set, edge amplitude weight parameter, and node time interval index. The high-density frequency interval specifically includes the frequency sequence index, turning frequency band boundary, and center frequency band fluctuation label. The spectrum edge graph gradient difference includes the spectrum edge distribution area, graph node slope response intensity, and structural gradient difference interval. The gearbox fault diagnosis result specifically refers to the graph structure breakpoint distribution position, fault boundary direction sequence, and fault type identification label.

[0065] The specific steps for obtaining the slope mutation node set are:

[0066] S101: calling a vibration data sequence, analyzing a slope change of continuous sampling points in a gearbox input end bearing vibration sequence, sequentially comparing amplitude and time information between adjacent sampling points in the sampling point sequence, establishing a continuous slope change trend, and obtaining a slope change trend sequence;

[0067] Call the vibration data sequence of the bearing at the input end of the gearbox. First, determine the sampling frequency to be 10000 Hz. At this frequency, obtain 8192 sampling points in the complete speed cycle to obtain a complete one-dimensional vibration signal sequence. When the sequence is processed in sequence, for every two adjacent points x i and x i+1 , calculate the amplitude difference Δy=x i+1 -x i and time difference seconds, and then calculate the slope value of the adjacent points The first layer slope sequence is formed, where Δy is the amplitude difference between adjacent sampling points, Δt is the sampling time interval, S i is the i-th slope value, and then the slope pair S of three consecutive sampling points is formed i ,S i+1 Calculate the difference ΔS=S i+1 -S i , judge the direction and amount of change of adjacent slopes, and judge whether the fluctuation trend is rising, falling or stable based on this, and construct a label sequence of continuous slope change trend. If the ΔS fluctuation direction of multiple consecutive points remains consistent, the trend is linear continuation, otherwise there is an inflection point. For the simulation signal {0.5, 0.8, 0.6, 0.3, 0.7}, S iThe sequence is {+0.3, -0.2, -0.3, +0.4}, and the ΔS sequence is {-0.5, -0.1, +0.7}. This indicates that the slope changes significantly at the fourth point. In an experimental scenario with stable mechanical speed and a fixed vibration excitation frequency, the slope change trend can directly reveal the impact waveform boundary. After establishing a sequence record of the slope change trend corresponding to each point, a complete trend list is constructed in the sampling order, ultimately obtaining a slope change trend sequence.

[0068] S102: comparing the change differences between multiple adjacent slope segments according to the slope change trend sequence, detecting continuous jump trends, and marking corresponding sampling points as node markers to obtain a node marker set;

[0069] According to the slope change trend sequence, the fluctuation degree of each group of adjacent slope pairs is evaluated, and the fluctuation judgment benchmark is set as the slope difference threshold η S , select η according to the range of equipment vibration amplitude S =0.15, if ΔS between adjacent slope pairs is greater than η S , and the sign reversal occurs more than 3 times in a row, it is determined that there is a jump trend. At this time, the position of the middle point is recorded as the key jump point in the node list. For example, in the sequence {+0.25, -0.2, +0.3}, there are 2 sign reversals and the absolute differences are both greater than η S , then the middle point is regarded as the jump point, where η S is the threshold used for slope jump judgment, ΔS is the difference between adjacent slopes, and for the periodic shock caused by poor meshing in the gear transmission process, the slope jump often appears at the peak point after the meshing shock. All points that meet the above conditions are added to the node marker set as candidate node markers. This set not only retains the nonlinear inflection point information of the vibration signal, but also strengthens the expression of the periodic shock characteristics through slope reversal, and finally obtains the node marker set.

[0070] S103: using the node marking set and combining it with the time sequence, sorting the marking points in the node marking set, removing the unmarked non-turning points, forming a graph structure node set, and outputting the slope mutation node set;

[0071] Using the node marker set, combined with the sampling timing index, the marker points are reordered, and the marker points are numbered in ascending order according to the sampling time in the original signal. The points that do not meet the jump amplitude condition are eliminated, even if they appear in the change trend but do not reach ΔS>η. SIf there is no requirement, the node sequence is not included. For example, the index of the marked points in the original sequence is {100, 120, 130, 132}, where the difference between 130 and 132 is too small to have independent identification significance, then only 100 and 120 are retained as valid nodes, the vertex setting of the graph structure is completed, and a graph structure node set is constructed. This set clearly defines the actual index position of the node, the vibration amplitude value, and its position number in the original time series, establishing a data basis for the subsequent calculation of the edge and edge weight between nodes, and finally outputting the slope mutation node set.

[0072] The specific steps for obtaining the graph structure weight matrix are:

[0073] S201: calling the slope mutation node set, comparing the amplitude increase and decrease trends and time sequence between nodes based on the amplitude change of each pair of adjacent nodes in the node set and the sampling time interval, and generating an amplitude and time change sequence;

[0074] Call the slope mutation node set and select each pair of adjacent nodes v in the graph structure in turn i and v i+1 , respectively obtain the vibration amplitude a i 、a i+1 , sampling index t i , t i+1 , compare the vibration amplitude change and the sampling time sequence, for each pair of nodes, calculate the amplitude difference as a i+1 -a i , the time interval is t i+1 -t i ,If the amplitude is 0.25, 0.30 and the time index is 105, 106, then the amplitude change is 0.05, the time interval is 1, and the record change rate is 0.05, which is used to judge the change speed. Then, this kind of processing is performed on all node pairs in turn to form a complete series of joint changes of amplitude and time, where a i represents the vibration amplitude of the i-th node, a i+1 Indicates the vibration amplitude of the next node, t i , t i+1 are the indexes of the two nodes in the sampling sequence, (a i+1 -a i ) represents the direction and amplitude of vibration signal fluctuation, (t i+1 -t i ) represents the sampling distance between nodes. This process traverses the entire graph nodes. After obtaining the amplitude and time relationship of each pair of nodes, all node relationships are summarized to obtain the amplitude and time change sequence.

[0075] S202: Based on the amplitude and time change sequence, the consistency of the amplitude change direction of each pair of nodes is judged. Combined with the node amplitude change direction, the connection order and connection direction between the nodes are identified, and an edge connection direction set is established;

[0076] Based on the amplitude and time change sequence, the vibration trend direction of each pair of nodes is judged. The amplitude difference values ​​of each group recorded in the previous paragraph are called. If the two adjacent change rates are positive, the trends are considered to be consistent. If one is positive and the other is negative, the trend is considered to be reversed. The signed product method is used to determine the consistency of direction. For example, if the amplitude difference of the previous group is +0.05 and the amplitude difference of the second group is +0.02, the product of the two is positive, indicating that the direction is consistent. If the amplitude difference of the second group is -0.04, the product is negative, indicating that the direction is reversed. According to the continuity of the trend, the node chains with the same direction are connected and marked. The connection direction is determined by the size of the node index value. For example, if the vibration amplitudes of nodes v5 and v6 are 0.45 and 0.50, and the indexes are 150 and 151, the direction is rising and the time sequence is correct. The connection direction is recorded from v5 to v6. Here, the connection direction set will be recorded as (v5→v6). The entire set of node pairs is traversed in this way. Finally, all connection paths that meet the consistency of amplitude change direction and continuous sampling order are extracted in time sequence, and a complete edge connection direction set is established.

[0077] S203: Analyze the amplitude difference and sampling point interval between nodes based on the edge connection direction set, construct edge weight reference, output the structural connection relationship between nodes, and generate a graph structure weight matrix;

[0078] The specific formula for constructing the edge weight reference is:

[0079]

[0080] Calculate edge weight reference value;

[0081] Among them, W ij is the reference value of the edge weight between node i and node j, ΔV ij is the normalized value of the vibration amplitude difference between node i and node j, t i is the normalized value of the sampling time of node i, t j is the normalized value of the sampling time of node j, γ ij is the normalized value of the slope disturbance change between node i and node j, n i is the number of node i in the graph structure, n j is the number of node j in the graph structure.

[0082] formula:

[0083]

[0084] Detailed explanation of the formula and the process of formula calculation and derivation:

[0085] The formula is used to calculate the structural connection edge weight reference value W between node i and node j ij , used for the adjacency matrix element A in the graph structure weight matrix ij Weight setting;

[0086] Parameter meaning and setting value:

[0087] ΔV ij It represents the normalized value of the vibration amplitude difference between node i and node j. Based on the peak acceleration data of the bearing vibration sensor monitored at a sampling frequency of 10 kHz, the vibration amplitude values ​​of nodes i and j corresponding to the sampling points are set to 1.85 mm / s and 1.30 mm / s, respectively. Compared with the maximum amplitude of 3.20 mm / s and the minimum amplitude of 0.95 mm / s in the whole sequence, the normalized calculation is: It reflects the relative difference level of amplitude between nodes;

[0088] t j -t i It represents the normalized value of the time difference between two sampling points. Assuming that in the same vibration data sequence, nodes i and j are separated by 15 sampling points, the corresponding time difference is seconds, compared with the maximum interval of 0.0020 seconds in the monitoring sequence, the normalized value is Reflects the intensity of the sampling time interval between nodes;

[0089] γ ij It represents the normalized value of the slope disturbance change between node i and node j. The local trend slope values ​​of the five sampling points near the two nodes are set to be 3.2 and 1.4, with a slope difference of 1.8. The maximum slope fluctuation in the comparison sample is 4.5. The normalized calculation is Reflects the impact of trend fluctuations between adjacent nodes on connection strength;

[0090] n j -n i Indicates the node number difference. Set node i to 8, node j to 11, the difference to 3, and the modulus to 3.

[0091] Substitute the above parameters into the formula for calculation:

[0092]

[0093] Results ij ≈0.0519 indicates that the connection edge has a relatively weak edge weight connection strength in the graph structure, and the structural relationship between the participating nodes is relatively sparse. It can be used as a quantification basis for non-critical connection edges in the graph structure weight matrix, and used for the subsequent construction of structural difference indicators and feature extraction basis in the graph domain.

[0094] The specific steps for obtaining the high-density frequency interval are as follows:

[0095] S301: Calling the graph structure weight matrix, analyzing the spectral coefficient sequence corresponding to each node in the matrix, determining the number of times the slope direction in the spectral coefficient sequence switches according to the node order, and counting the frequency of slope reversal to obtain a slope reversal frequency sequence;

[0096] Call the graph structure weight matrix, select the row corresponding to each node in the weight matrix W in turn, obtain the connection weight information between the node and all other nodes, and for node v i , read W i ={w i1 ,w i2 ,…,w in}, where w ij Represents node v i With v j The edge weights are selected, and the node index and weight value corresponding to the non-zero weight are selected. The spectral coefficient sequence is reorganized from small to large according to the node index, and then for any two adjacent items in the sequence, k ,λ k+1 , calculate the spectral coefficient difference Δλ k =λ k+1 -λ k If the previous difference is positive and the next difference is negative, or vice versa, a slope reversal phenomenon is recorded. All positive and negative transformation points in the spectral coefficient sequence of the node are counted as the number of slope reversals of the node. This process is applied to all nodes to form a slope reversal frequency sequence. For example, the corresponding weight of node v1 is {0.13, 0.18, 0.15, 0.22, 0.17}, and the spectral coefficient difference sequence is {+0.05, -0.03, +0.07, -0.05}. There are 3 sign changes in total, and the slope reversal frequency of node v1 is 3. The slope reversal frequency of each node is sorted in sequence to finally obtain the slope reversal frequency sequence of all nodes.

[0097] S302: sorting the spectral coefficients according to the frequency order based on the slope inversion frequency sequence, dividing the spectral coefficient sequence into windows of fixed width, counting the number of inversion points in each window, marking the turning point dense distribution area, and obtaining a set of turning point concentrated segments;

[0098] Based on the slope reversal frequency sequence, all spectral coefficient data points are rearranged according to their distribution order in the frequency domain. Each frequency segment is partitioned using a fixed window width of 20 Hz. The frequency range of 0 to 2000 Hz is divided into 100 windows, each containing 20 frequency points. The number of slope reversals corresponding to all spectral coefficients in each window is cumulatively counted to determine the number of slope reversal points in the window. The judgment threshold γ is set to the window average reversal frequency plus the standard deviation. In practical application, if the number of slope reversal points in a window occurs 15 times, while the overall average is 8 times, with a standard deviation of 2, then γ = 10. This window reversal point is greater than the threshold γ, indicating a dense turning point distribution area. The frequency segment corresponding to the window is recorded as a candidate turning point segment. For example, the frequency windows [400, 420] Hz and [900, 920] Hz both meet the dense area criteria. All window frequency bands that meet the criteria are merged to form an interval set, and the final output is a concentrated turning point segment set.

[0099] S303: Call the turning point concentrated segment set, construct a linear trend at the boundary of each segment, compare the fluctuation intensity in the segment center area and the change rate on both sides of the segment, screen the disturbance significant frequency band, and output the high-density frequency interval;

[0100] Call the turning point concentrated segment set and perform feature extraction on each frequency segment in turn. First, extract 5 consecutive spectral coefficient points on the left and right ends of each segment respectively. Use these points to perform linear trend fitting using the least squares method to obtain the trend slopes on the left and right sides. Extract 10 spectral points in the middle of the segment and calculate the average of their change amplitudes, which is set as the central area fluctuation intensity. Then compare the difference between the central area fluctuation intensity and the slope fitting results on both sides of the segment. If the central area intensity is significantly greater than the fitting slopes on both sides, the frequency segment is determined to be a significantly disturbed frequency segment. For example, in the frequency segment [700,720] Hz, the slope on the left end is 0.015, the slope on the right end is 0.012, the average fluctuation of the 10 central points is 0.035, and the difference between the central and boundary slopes is greater than the set criterion 0.01, which is determined to be a high disturbance segment. Finally, sort out all the frequency intervals determined to be significantly disturbed, number and merge their frequency index ranges, and output all high-density frequency intervals that meet the requirements.

[0101] The specific steps for obtaining the gradient difference of the spectral edge map are:

[0102] S401: Calling a high-density frequency interval, extracting symmetrically distributed spectrum boundaries within the frequency interval, counting slope changes at both ends of the boundaries, calculating the average level of slope changes on each side, and obtaining a boundary slope change mean value group;

[0103] Call the high-density frequency interval, first traverse the spectrum points in each high-density frequency interval, determine the left and right sides of the interval, take five spectrum points respectively as the left boundary and right boundary, sort the spectrum amplitude sequences within the two boundaries, and then for the left boundary, calculate the difference in the amplitude of each two adjacent frequency points and divide it by the frequency point spacing to obtain the left boundary slope change group. Similarly, calculate the right boundary slope change group on the right side, average all slope values ​​in each slope group, and obtain the average slope level of the left and right boundaries, that is, the boundary slope change mean group. For example, assuming that the amplitudes of the five spectrum points on the left are 0.22, 0.24, 0.27, and 0. .25, 0.28, corresponding to frequencies of 395, 396, 397, 398, 399 Hz respectively. The slope groups on the left are 0.02, 0.03, -0.02, 0.03, with an average of 0.015. Similar operations are performed on the right. Assuming the five amplitudes are 0.33, 0.35, 0.34, 0.38, 0.37, corresponding to frequencies of 421, 422, 423, 424, 425 Hz, the slope groups on the right are 0.02, -0.01, 0.04, -0.01, with an average of 0.01. Finally, the average slope levels on the left and right ends are statistically sorted to obtain the boundary slope change mean group of the current high-density frequency interval.

[0104] S402: combining the boundary slope change mean value group, locating the node distribution area corresponding to the spectrum boundary in the graph signal, calculating the change in the amplitude gradient of the nodes on the left and right sides, and obtaining the node gradient difference index;

[0105] Combined with the boundary slope change mean group, the node index and signal amplitude corresponding to the left and right boundaries of the spectrum interval are located in the graph signal structure. According to the average slope at both ends of the high-density frequency interval, the node sets represented by the corresponding frequency points in the graph structure are selected respectively, and the amplitude value of each node is extracted. The amplitude change between the nodes is calculated one by one. The difference between the maximum and minimum amplitude values ​​of the node sets on the left and right sides is divided by the index span between the nodes to obtain the average gradient of the amplitude change of the nodes on each side. The amplitude gradients of the nodes on the left and right sides are compared respectively, and the difference between the gradients of the nodes on both sides is calculated. The obtained value is the node gradient difference index. For example, on the left side, the node amplitudes are 0.20, 0.22, 0.25, 0.26, and 0.28, corresponding to node indexes 101, 102, 103, 104, and 105, and the average gradient is (0.28-0.20) / (105-101)=0.02. On the right side, the amplitudes are 0.32, 0.33, 0.35, 0.38, and 0.37, corresponding to indexes 121, 122, 123, 124, and 125, and the average gradient is (0.37-0.32) / (125-121)=0.0125. The final node gradient difference index is 0.02-0.0125=0.0075.

[0106] S403: Based on the node gradient difference index, the intensity of the structural change caused by the spectrum perturbation is evaluated and mapped into a graph domain response index in the graph structure to obtain the spectral edge graph gradient difference;

[0107] Based on the node gradient difference index and combined with the graph signal structure, the gradient difference value is used as a quantitative parameter to characterize the structural changes caused by spectral perturbations. The node gradient difference index corresponding to all high-density frequency intervals is traversed and normalized to the range of 0-1. The normalized gradient difference index is directly used as a numerical measure of the intensity of structural change. This value is assigned to the corresponding area of ​​the graph structure. For each high-density frequency interval, a corresponding graph domain response block is established. Finally, the spectral edge graph gradient difference in each interval is obtained. For example, if the gradient difference before normalization is 0.008, the maximum gradient difference of all segments is 0.010 and the minimum gradient difference is 0.002. Then, after normalization, it is (0.008-0.002) / (0.010-0.002)=0.75. The normalized results of all intervals are sequentially filled into the graph structure matrix. The final output spectral edge graph gradient difference is the response distribution of the structural change in each interval.

[0108] The specific steps for obtaining gearbox fault diagnosis results are as follows:

[0109] S501: Obtain the gradient difference of the spectral edge graph, locate the node sequence corresponding to the difference in the graph structure, identify the continuous peaks of signal fluctuations in the node sequence, extract the amplitude according to the node sequence, establish the amplitude change direction between adjacent nodes, and generate a node amplitude direction sequence;

[0110] After obtaining the gradient difference of the spectral edge graph, traverse the graph structure node set, locate all nodes corresponding to the gradient difference interval one by one, read the amplitude value of the node, and arrange them in order of node index. Then, for the entire node sequence, detect the signal amplitude change point by point, find the local maximum and minimum values, and judge the amplitude increase and decrease relationship between adjacent nodes. If the amplitude of the latter node is greater than that of the previous node, it is marked as the positive direction, otherwise it is the negative direction. Construct the node amplitude direction sequence in this order. For example, in a certain segment, the node amplitude sequence is 0.22, 0.25, 0.24, 0.27, 0.23, and 0.28, and the corresponding direction sequence is positive, negative, positive, negative, and positive. The changes between each node are fully listed to form a direction sequence with a length equal to the number of nodes minus one. Finally, complete the qualitative induction of the amplitude change direction of all nodes in the interval and obtain the node amplitude direction sequence.

[0111] S502: Based on the node amplitude direction sequence, the positive and negative change sequences are compared with multiple training sample direction templates for similarity, and the fault boundary area is identified and divided by comparing the matching degree to obtain a fault boundary interval set;

[0112] The specific formula for similarity comparison between positive and negative change sequences and multiple training sample direction templates is:

[0113]

[0114] Calculate the directional trend difference value;

[0115] Among them, S km is the direction trend difference value, k is the number index of the amplitude direction sequence of the node to be identified, m is the number index of the direction sequence of the training sample template, d ku is the normalized amplitude direction value of the u-th sampling point in the k-th node sequence, t mu is the normalized amplitude direction value of the u-th sampling point in the m-th template direction sequence, u is the index of the sampling point in the sequence, L k is the total number of sampling points in the k-th node sequence.

[0116] formula:

[0117]

[0118] Detailed explanation of the formula and the process of formula calculation and derivation:

[0119] The formula is used to calculate the directional trend difference between the node sequence and the training template. The result is used to identify similar patterns to the node fluctuations in the classification graph structure and locate the fault boundary interval;

[0120] Parameter meaning and setting value:

[0121] d ku Represents the normalized amplitude direction value of the u-th sampling point in the k-th sample direction sequence. It is normalized by detecting the continuous amplitude change direction of the graph nodes. The normalization method is the ratio of each sampling point to the maximum absolute amplitude of the sequence. The value range is limited to [-1, 1] and is set to [-0.9, 0.6, -0.3, 0.7, -0.8];

[0122] t mu It represents the normalized amplitude direction value of the u-th point in the m-th training template direction sequence, which is obtained by normalizing the signal direction in the sample template. The standard template is constructed by summarizing the typical waveforms under various gear fault types in the measured data samples. The normalization method is the same as d ku Same, set to [-1.0, 0.5, -0.5, 0.8, -0.7];

[0123] L k Indicates the current sequence length, which is equal to the number of sampling points and is set to 5;

[0124] Substitute the parameters into the formula for calculation:

[0125]

[0126]

[0127] Results km ≈0.1893 indicates that there is a low degree of difference between the current graph node direction sequence and the training sample direction template. This value is close to 0, indicating that the two sequences have a high degree of match in direction trend and have strong pattern consistency. The value can be used to determine the minimum difference template in multi-template comparison, and finally the minimum S km The corresponding training template type is used as the fault category to which the node sequence belongs, and a set of fault boundary intervals is further obtained.

[0128] S503: Based on the fault boundary interval set, analyze the amplitude change trajectory between adjacent segments in the boundary area, identify the distribution characteristics of the graph structure breakpoints, determine the gearbox fault type, and obtain the gearbox fault diagnosis result;

[0129] Based on the fault boundary interval set, for each boundary segment, the amplitude values ​​of all nodes within it are extracted, and the amplitude differences between adjacent nodes are analyzed to determine whether there are amplitude mutation points or abnormal jump segments in each segment. The node indexes with significant amplitude changes are recorded as graph structure breakpoints, and their distribution characteristics are statistically analyzed. For example, in a certain boundary segment, the amplitude change sequence is 0.23, 0.27, 0.45, 0.26, and 0.28. The amplitude suddenly increases by 0.18 from node 2 to node 3, which is much higher than other adjacent differences. Node 3 is then determined to be a graph structure breakpoint. In this way, all breakpoints in each fault boundary segment are identified. Combined with the number and location of the breakpoints and the overall direction change characteristics of the segment, and compared with the known fault modes in the diagnosis database, the fault type of the current gearbox state is determined, and the type number of the interval is output. Finally, the type determination results of all segments are summarized to output a comprehensive gearbox fault diagnosis result.

[0130] See also Figure 2 , a gearbox fault diagnosis system based on graph structure signal processing, a gearbox fault diagnosis system based on graph structure signal processing is used to execute the above-mentioned gearbox fault diagnosis method based on graph structure signal processing, and the system includes:

[0131] The slope detection module calls the vibration data sequence, analyzes the slope change differences composed of multiple consecutive sampling points in the gearbox input end bearing vibration sequence, detects the continuous jump trend, identifies the corresponding sampling points as node markers, and outputs the slope mutation node set;

[0132] The edge connection construction module uses the slope mutation node set to construct the edge connection direction based on the amplitude change and sampling time difference between each node in the point set. Based on the amplitude difference between nodes and the sampling point interval, it constructs the edge weight reference and generates the graph structure weight matrix.

[0133] The spectrum segment analysis module calculates the spectrum coefficient sequence based on the graph structure weight matrix, records the number of slope reversals within a fixed window, identifies the turning point concentrated segment and constructs a linear trend for the segment boundary. It compares the fluctuation intensity of the central area and the difference in the rate of change of the trend of the two sides, selects the frequency band with significant disturbance, and outputs the high-density frequency interval.

[0134] The gradient evaluation module uses high-density frequency intervals to extract symmetrically distributed spectrum boundaries and compares the average level of slope changes at both ends of the edge. Combined with the corresponding node distribution area in the graph signal, it calculates the difference in amplitude gradient between the nodes on both sides and maps it to the graph domain response index in the graph structure to obtain the spectral edge graph gradient difference.

[0135] The fault judgment module obtains the gradient difference of the spectral edge graph, locates the corresponding node sequence in the graph structure, extracts the continuous peaks in the signal fluctuation, and establishes the node amplitude direction sequence. By comparing the matching similarity of the positive and negative change sequences and multiple training sample direction templates, it identifies the fault boundary and graph structure breakpoints, determines the gearbox fault type, and obtains the gearbox fault diagnosis result.

[0136] The above embodiments can be implemented in whole or in part through software, hardware (such as circuits), firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired method (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, or magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0137] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0138] In this disclosure, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, "at least one of a, b, or c" can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0139] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0140] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0141] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0142] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interface, indirect coupling or communication connection of the device or unit, which can be electrical, mechanical or other forms.

[0143] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0144] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0145] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0146] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A gearbox fault diagnosis method based on graph structure signal processing, characterized in that: The method comprises: S1: Call the vibration data sequence, analyze the slope change differences of multiple consecutive sampling points in the gearbox input end bearing vibration sequence, detect the continuous jump trend, mark the corresponding sampling points as node marks, and output the slope mutation node set; S2: Using the slope mutation node set, construct the edge connection direction according to the amplitude change and sampling time difference between each node in the point set, and construct the edge weight reference based on the amplitude difference between the nodes and the sampling point interval to generate the graph structure weight matrix; S3: Calculate the spectral coefficient sequence based on the graph structure weight matrix, record the number of slope reversals within a fixed window, identify the turning point concentrated segment and construct a linear trend for the segment boundary, compare the fluctuation intensity of the central area and the difference in the rate of change of the trend of the two sides, screen the frequency bands with significant disturbances, and output the high-density frequency interval; S4: Call the high-density frequency interval, extract the symmetrically distributed spectrum boundary, and compare the average level of slope change at both ends of the edge. Combined with the corresponding node distribution area in the graph signal, calculate the degree of difference in amplitude gradient between the nodes on both sides, and map it to the graph domain response index in the graph structure to obtain the spectral edge graph gradient difference.

2. The gearbox fault diagnosis method based on graph structure signal processing according to claim 1 is characterized in that: The slope mutation node set includes the mutation inflection point position, time series index, and node amplitude change direction. The graph structure weight matrix is ​​specifically the edge connection direction set, edge amplitude weight parameter, and node time interval index. The high-density frequency interval is specifically the frequency sequence index, turning frequency band boundary, and center frequency band fluctuation label. The spectrum edge graph gradient difference includes the spectrum edge distribution area, graph node slope response intensity, and structural gradient difference interval.

3. The gearbox fault diagnosis system based on graph structure signal processing is characterized by: The system is used to implement the gearbox fault diagnosis method based on graph structure signal processing according to any one of claims 1-2, and the system includes: The slope detection module calls the vibration data sequence, analyzes the slope change differences composed of multiple consecutive sampling points in the gearbox input end bearing vibration sequence, detects the continuous jump trend, identifies the corresponding sampling points as node markers, and outputs the slope mutation node set; The edge connection construction module uses the slope mutation node set to construct the edge connection direction according to the amplitude change and sampling time difference between each node in the point set, and constructs the edge weight reference based on the amplitude difference between the nodes and the sampling point interval to generate the graph structure weight matrix; The spectrum segment analysis module calculates the spectrum coefficient sequence based on the graph structure weight matrix, records the number of slope reversals within a fixed window, identifies the turning point concentrated segment and constructs a linear trend for the segment boundary, compares the fluctuation intensity of the central area and the difference in the rate of change of the trend of the two sides, screens the frequency bands with significant disturbance, and outputs the high-density frequency interval; The gradient evaluation module calls the high-density frequency interval, extracts the symmetrically distributed spectrum boundary, and compares the average level of slope change at both ends of the edge. Combined with the corresponding node distribution area in the graph signal, it calculates the difference in amplitude gradient between the nodes on both sides and maps it to the graph domain response index in the graph structure to obtain the spectral edge graph gradient difference; The fault determination module obtains the gradient difference of the spectral edge graph, locates the corresponding node sequence in the graph structure, extracts the continuous peaks in the signal fluctuation, and establishes a node amplitude direction sequence. By comparing the matching similarity of the positive and negative change sequences and multiple training sample direction templates, the fault boundary and graph structure breakpoints are identified, the gearbox fault type is determined, and the gearbox fault diagnosis result is obtained.

Images