A method, device, computer system and computer program product for monitoring the health of a rotating machinery

Abstract

The application discloses a health state monitoring method and device of rotating mechanical equipment, a computer component and a computer program product, relates to the field of mechanical equipment state monitoring and fault diagnosis, and comprises the following steps: acquiring a multi-channel vibration signal; constructing a graph fusion model, extracting a multi-channel graph frequency domain feature matrix through graph Fourier transform; taking the coherence of adjacent channel signals as a weight to fuse features and obtaining a channel fusion graph; adopting a random forest method to calculate the importance weight of each frequency domain feature, and obtaining a health index through weighted fusion; and dividing the health state based on the index, combining the 3σ principle and Mann-Kendall trend test. The application effectively solves the problems of multi-channel signal information redundancy and contradiction through graph model and feature fusion, constructs a health index sensitive to equipment degradation, and realizes accurate state division in combination with statistical test, thereby significantly improving the accuracy and reliability of monitoring.

Description

Technical Field

[0001] This application relates to the field of mechanical equipment condition monitoring and fault diagnosis, and in particular to a method, device, computer system and computer program product for monitoring the health status of rotating mechanical equipment. Background Technology

[0002] As industrial equipment develops towards larger scale and greater intelligence, health monitoring of rotating machinery is of great significance for ensuring production safety, avoiding unplanned downtime, and reducing maintenance costs. Currently, health monitoring of equipment often employs multi-channel sensors simultaneously to obtain richer machine status information than single-channel signals, thus providing a more comprehensive reflection of the equipment's operating status.

[0003] However, in multi-channel signal monitoring, due to the large number of sensors and high data dimensionality, there is often information redundancy or even contradiction between signals collected from different channels. This makes it difficult to effectively fuse multi-source heterogeneous information to construct health indicators that accurately reflect the equipment degradation trend. Related health monitoring methods typically perform simple data-level fusion or single feature extraction directly on the original multi-channel signals. This fails to fully utilize the inherent correlation and complementarity between multi-channel data, making it difficult to resolve information redundancy and contradiction issues. Consequently, feature information extraction is insufficient or severely affected by noise interference, thus impacting the accuracy and reliability of equipment health status classification.

[0004] Therefore, there is an urgent need for a health status monitoring method for rotating machinery to solve the problem of low accuracy in health indicator construction caused by redundancy and contradiction of multi-channel signal information, thereby improving the accuracy and reliability of health status monitoring of rotating machinery. Summary of the Invention

[0005] The purpose of this application is to provide a method, device, computer system, and computer program product for monitoring the health status of rotating machinery, which can effectively integrate complementary information from multi-channel vibration signals, suppress redundancy and interference, construct health indicators that can accurately reflect the equipment degradation process, and achieve stable and reliable classification of the equipment health status.

[0006] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for monitoring the health status of rotating machinery, comprising: Acquire multi-channel vibration signals of the rotating mechanical equipment to be monitored within the current time period; Based on the multi-channel vibration signals, a graph fusion model is constructed; Based on the graph fusion model, the graph frequency domain features of each channel are extracted using the graph Fourier transform method to obtain the multi-channel graph frequency domain feature matrix for the current time period. Using the signal coherence between adjacent channels as fusion weights, the frequency domain features of each channel in the multi-channel graph frequency domain feature matrix are fused to obtain the channel fusion graph for the current time period. The importance weight of each frequency domain feature relative to all frequency domain features in the channel fusion map is calculated using the random forest method. Based on the importance weights corresponding to each frequency domain feature, all frequency domain features in the channel fusion map are weighted and fused to obtain the health index of the rotating mechanical equipment to be monitored in the current time period. Based on the health indicators of the rotating machinery to be monitored in the current time period, a statistical analysis method is used to classify the health status, and the health status monitoring results of the rotating machinery to be monitored in the current time period are obtained. The statistical analysis method includes the 3σ principle detection method and the Mann-Kendall trend test method. The health status monitoring results include normal status, minor fault status and severe fault status.

[0007] Secondly, this application provides a health status monitoring device for rotating machinery, comprising: The signal acquisition module is used to acquire multi-channel vibration signals of the rotating mechanical equipment to be monitored within the current time period; The graph construction module is used to construct a graph fusion model based on the multi-channel vibration signal; The frequency domain feature extraction module is used to extract the graph frequency domain features of each channel based on the graph fusion model by using the graph Fourier transform method, so as to obtain the multi-channel graph frequency domain feature matrix for the current time period. The channel fusion module is used to fuse the frequency domain features of each channel in the multi-channel graph frequency domain feature matrix by using the signal coherence between adjacent channels as fusion weights to obtain the channel fusion graph for the current time period. The weight calculation module is used to calculate the importance weight of each frequency domain feature in the channel fusion map relative to all frequency domain features using the random forest method; The health indicator construction module is used to perform weighted fusion of all frequency domain features in the channel fusion map based on the importance weights corresponding to each frequency domain feature, so as to obtain the health indicators of the rotating mechanical equipment to be monitored in the current time period. The health status classification module is used to classify the health status of the rotating machinery to be monitored based on its health indicators in the current time period using statistical analysis methods, and to obtain the health status monitoring results of the rotating machinery to be monitored in the current time period. The statistical analysis methods include the 3σ principle detection method and the Mann-Kendall trend test method. The health status monitoring results include normal status, minor fault status, and severe fault status.

[0008] Thirdly, this application provides a computer system comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the health status monitoring method for rotating mechanical equipment as described above.

[0009] Fourthly, this application provides a computer program product having a computer program stored thereon, which, when executed by a processor, implements the steps of the health status monitoring method for rotating mechanical equipment described above.

[0010] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method, device, computer system, and computer program product for monitoring the health status of rotating machinery. By converting multi-channel vibration signals into a graph fusion model and extracting physically related frequency domain feature matrices using graph Fourier transform, it solves the key problems of information mixing and feature redundancy caused by direct fusion of multi-channel signals, and achieves efficient extraction of core feature information representing equipment status from high-dimensional heterogeneous data. By using the signal coherence between adjacent channels as fusion weights to fuse multi-channel graph frequency domain features, and combining the random forest method to adaptively calculate the importance weights of each frequency domain feature for weighted fusion to obtain health indicators, it solves the problems of traditional methods that cannot fully suppress noise or highlight key degradation features due to fixed weights or simple average fusion, and achieves the construction of health indicators that are more sensitive to and robust to equipment degradation trends. By adopting a health status classification method that combines the 3σ principle and the Mann-Kendall trend test, it solves the problem that single threshold or trend judgment is easily affected by fluctuations and has a high misjudgment rate when distinguishing between normal, minor and severe fault states of equipment, and achieves accurate and stable identification of the progressive evolution of equipment health status, significantly improving the reliability and engineering practicality of rotating machinery status monitoring. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating a method for monitoring the health status of rotating mechanical equipment according to an embodiment of this application.

[0013] Figure 2 This is a flowchart illustrating a method for monitoring the health status of rotating mechanical equipment, provided as another embodiment of this application.

[0014] Figure 3 A bearing health index curve obtained using the method of this application based on a bearing dataset is provided as an embodiment of this application.

[0015] Figure 4 An embodiment of this application provides a bearing health index curve obtained based on a bearing dataset using a comparative method; wherein, Figure 4 (a) in the figure is a curve of bearing health index based on root mean square value; Figure 4 (b) in the figure is a curve of bearing health index based on the Gini index; Figure 4 (c) in the figure is a bearing health index curve based on negative entropy; Figure 4 (d) in the figure is a curve of bearing health index based on spectral energy weighting.

[0016] Figure 5 This is a schematic diagram of the X-axis provided in one embodiment of this application.

[0017] Figure 6 A health index curve obtained using the method of this application based on a machining center translation axis dataset, provided as an embodiment of this application.

[0018] Figure 7 An embodiment of this application provides a health index curve obtained based on a machining center translation axis dataset using a comparative method; wherein, Figure 7 (a) in the figure is a health index curve based on the root mean square value and shifted along the axis; Figure 7 (b) in the figure is a shifted axis health index curve based on the Gini index; Figure 7 (c) in the figure is a translation axis health index curve based on negative entropy; Figure 7 (d) in the figure is a translation axis health index curve based on spectral energy weighting.

[0019] Figure 8 This is a schematic diagram of the functional modules of a health status monitoring device for rotating machinery provided in one embodiment of this application.

[0020] Figure 9 This is a schematic diagram of the structure of a computer system provided in an embodiment of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] In one exemplary embodiment, such as Figure 1 As shown, a method for monitoring the health status of rotating mechanical equipment is provided, including the following steps 101 to 107. Wherein: Step 101: Obtain multi-channel vibration signals of the rotating mechanical equipment to be monitored within the current time period.

[0024] Step 102: Construct a graph fusion model based on the multi-channel vibration signal.

[0025] Step 103: Based on the graph fusion model, extract the graph frequency domain features of each channel using the graph Fourier transform method to obtain the multi-channel graph frequency domain feature matrix for the current time period.

[0026] Step 104: Using the signal coherence between adjacent channels as fusion weights, the frequency domain features of each channel in the multi-channel graph frequency domain feature matrix are fused to obtain the channel fusion graph for the current time period.

[0027] Step 105: Calculate the importance weight of each frequency domain feature in the channel fusion map relative to all frequency domain features using the random forest method.

[0028] Step 106: Based on the importance weights corresponding to each frequency domain feature, perform weighted fusion of all frequency domain features in the channel fusion spectrum to obtain the health index of the rotating mechanical equipment to be monitored in the current time period.

[0029] Step 107: Based on the health indicators of the rotating machinery to be monitored in the current time period, a statistical analysis method is used to classify the health status, and the health status monitoring results of the rotating machinery to be monitored in the current time period are obtained; the statistical analysis method includes the 3σ principle detection method and the Mann-Kendall trend test method; the health status monitoring results include normal status, minor fault status and severe fault status.

[0030] By implementing steps 101 to 107 above, this application can effectively solve the problem of information redundancy and the difficulty in fully utilizing complementarity in multi-channel vibration signal monitoring. By converting multi-channel signals into a graph fusion model and extracting frequency domain features using graph Fourier transform, core state features are efficiently extracted from high-dimensional heterogeneous data. Furthermore, feature-level fusion is performed by combining the coherence of adjacent channel signals, and a random forest method is introduced to adaptively determine feature weights, enabling the construction of a health index that is sensitive to equipment degradation and robust. Finally, by using a state classification method combining the 3σ principle and the Mann-Kendall trend test, the progressive evolution process of equipment from normal to minor faults to severe faults can be accurately identified, significantly improving the accuracy, stability, and engineering practicality of health status monitoring of rotating machinery.

[0031] In another exemplary embodiment of this application, a graph fusion model is constructed based on the multi-channel vibration signal, specifically including: The multi-channel envelope signal is obtained by performing a Hilbert transform on the multi-channel vibration signal using the following formula: .

[0032] in, Indicates the first n Time of the first i The envelope signal of each channel; Indicates the first n Time of the first i Vibration signals from each channel; This represents the modulo operation; j Represents the imaginary unit; Indicates to Perform a Hilbert transform.

[0033] Construct a graph fusion model using the following expression: .

[0034] .

[0035] in, Representation graph fusion model; , , and Representing time 1, time 2, and time 3 respectively. Time and the N The vertex vector at time t. N Indicates the total number of moments; , , and They represent the first n Time slot 1, 2, 3i The first channel and the first q The envelope signal of each channel, q This indicates the total number of channels.

[0036] In another exemplary embodiment of this application, based on the graph fusion model, the graph frequency domain features of each channel are extracted using the graph Fourier transform method to obtain the multi-channel graph frequency domain feature matrix for the current time period, specifically including: The frequency domain features of each channel are extracted using the following expression: .

[0037] in, Indicates the first The frequency of the first graph i The frequency domain characteristics of each channel; express The complex conjugate; Indicates the first The orthogonal eigenvectors of the Laplacian matrix of the graph fusion model corresponding to each graph frequency are in the th order. n The value at time; In the graph fusion model, the first The first moment i The signal values ​​of each channel.

[0038] Based on the extracted image frequency domain features of each channel, the multi-channel image frequency domain feature matrix for the current time period is obtained through the following expression: .

[0039] .

[0040] in, This represents the frequency domain feature matrix of the multi-channel graph for the current time period; , and Representing the frequency of the first graph, the second graph, and the third graph respectively. The frequency of the first graph and the first The frequency domain characteristics of each graph frequency; , and They represent the first The first, second, and third channels of the image frequency. q The frequency domain characteristics of each channel.

[0041] In another exemplary embodiment of this application, the signal coherence between adjacent channels is used as the fusion weight to fuse the frequency domain features of each channel in the multi-channel graph frequency domain feature matrix to obtain the channel fusion graph for the current time period, specifically including: The signal coherence between adjacent channels is calculated using the following formula: .

[0042] in, Indicates the first i The first channel and the first The first channel in the The signal coherence value at each frequency; ,and ; Indicates the first i The first channel in the Image frequency place, and the first The first channel in the Image frequency Cross-power spectral density at; Indicates the first i The first channel in the Image frequency The self-power spectral density at the location; Indicates the first The first channel in the Image frequency The self-power spectral density at that location.

[0043] The following formula is used to fuse the frequency domain features of each channel in the multi-channel graph frequency domain feature matrix, using the signal coherence between adjacent channels as fusion weights, to obtain the channel fusion graph for the current time period: .

[0044] in, Indicates the first Channel fusion spectrum values ​​at individual frequency points; , and These represent the first and second channels in the [missing information - likely a number] section. The signal coherence value at the frequency of the first image, the second channel and the third channel at the second image. The signal coherence value at the frequency of the first graph and the first q -1 channel and the first q The first channel in the The signal coherence value at each frequency.

[0045] In another exemplary embodiment of this application, based on the importance weights corresponding to each frequency domain feature, all frequency domain features in the channel fusion spectrum are weighted and fused to obtain the health index of the rotating machinery to be monitored in the current time period, specifically including: Normalize the channel fusion map for the current time period to obtain the normalized channel fusion map for the current time period.

[0046] Based on the fault characteristic frequency and frequency resolution, a first preset number of frequency domain features are selected from the normalized channel fusion spectrum to obtain a key frequency domain feature set.

[0047] The health index of the rotating machinery to be monitored in the current time period can be obtained using the following formula: .

[0048] in, Indicates the first m Health indicator values ​​for a given time period; K represents the preset quantity; Represents the key frequency domain feature set. The importance weights corresponding to each key frequency domain feature; Indicates the first m In the normalized channel fusion map of the time period, the first The eigenvalues ​​of key frequency domain features.

[0049] In another exemplary embodiment of this application, based on the health indicators of the rotating machinery to be monitored in the current time period, a statistical analysis method is used to classify the health status, and the monitoring results of the rotating machinery to be monitored in the current time period are obtained, specifically including: If the current time period is within the early operating time period, or the health index of the current time period is less than or equal to the first preset threshold, then the health status monitoring result of the rotating machinery to be monitored in the current time period is normal. The early operating time period includes the period from the first time period when the rotating machinery to be monitored starts operating to the second preset number of time periods. The first preset threshold is calculated based on the health index of the early operating time period using the 3σ principle detection method.

[0050] If the health index for the current time period is greater than the first preset threshold, and the Mann-Kendall trend test result for the current time period shows no statistically significant upward trend, then the health status monitoring result of the rotating machinery to be monitored in the current time period is a minor fault state; the Mann-Kendall trend test result is determined by the Mann-Kendall trend test method based on the health index of the current time period and all time periods before the current time period.

[0051] If the health index for the current time period is greater than the first preset threshold, and the Mann-Kendall trend test result for the current time period shows a statistically significant upward trend, then the health status monitoring result of the rotating machinery to be monitored during the current time period is a serious fault state.

[0052] In another exemplary embodiment of this application, based on health indicators of the current time period and all time periods before the current time period, the Mann-Kendall trend test result for the current time period is determined using the Mann-Kendall trend test method, specifically including: The Mann-Kendall statistic for the current time period is determined using the following expression: .

[0053] .

[0054] in, R represents the Mann-Kendall statistic for the current time period; R represents the total number of time periods. Represents a symbolic function; Indicates the first r Health indicator values ​​over a period of time.

[0055] The standardized statistic of S is calculated using the following formula: .

[0056] .

[0057] in, The standardized statistic representing the Mann-Kendall statistic for the current time period; This represents the variance of S.

[0058] like ,or If the Mann-Kendall trend test result for the current time period is that there is no statistically significant upward trend; among which, This represents the absolute value of the standardized Mann-Kendall statistic for the current time period. This represents the critical value for a two-tailed test of the standard normal distribution at the preset significance level α.

[0059] like ,and If the Mann-Kendall trend test result for the current time period is that there is a statistically significant upward trend.

[0060] The following example illustrates this application using a specific health status monitoring process for rotating mechanical equipment.

[0061] like Figure 2 The diagram shows a flowchart of a method for monitoring the health status of rotating machinery, which includes: Step 1: Acquire multi-channel vibration signals, calculate the envelope signals of the multi-channel vibration signals, use the multi-channel envelope signals as vertices V of the graph fusion model, set the edges to 1, and construct the graph fusion model FG.

[0062] As an optional implementation method, step 1 specifically includes: In the path graph model of vibration signals, it is usually defined that To represent two adjacent vertices and The edge between, Indicates the first Time vertex and the There are no connections between vertices at any given time. The Laplacian matrix L of the graph is defined as: (1).

[0063] in, It is the degree matrix of the vertices, defined as ,in It is the first Peak of Time The degree is expressed as W denotes the weighted adjacency matrix. By definition, the Laplace matrix is ​​a real symmetric matrix and has completely orthogonal eigenvectors, denoted as { } . N This represents the total number of moments.

[0064] Assuming multi-channel vibration signal ,in, ; q This represents the total number of channels. , i =1, 2, ..., q The vibration signal is captured by the sensor. Next, the envelope signal is obtained through Hilbert transform. As shown in Equation 2 below: (2).

[0065] in, Indicates the first n Time of the first i The envelope signal of each channel; Indicates the first n Time of the first i Vibration signals from each channel; This represents the modulo operation; j Represents the imaginary unit; Indicates to The Hilbert transform is performed on the envelope signal of the multi-channel vibration signal. For the envelope signal Y of a multi-channel vibration signal It still corresponds to the structure of the graph fusion model, while The vertices can be represented as .

[0066] The graph fusion model FG integrates graph vertices... The vector concatenation is performed to obtain the expression: (3).

[0067] Unlike traditional graph models, where vertices represent magnitudes, in the graph fusion model FG, vertices represent column vectors. It can be represented as Different vibration signals can be selected to construct However, the number of data points for different signals should be consistent to ensure correct acquisition. Furthermore, the graph fusion model FG is structurally similar to the path graph model, and the graph Fourier transform (GFT) theory can be directly applied.

[0068] Step 2: Calculate the Laplacian matrix L of the graph fusion model FG to obtain the graph signal of the graph fusion model FG. .

[0069] Step 3: Match the image signal Perform a graphical Fourier transform to obtain the multi-channel graphical frequency domain feature matrix U.

[0070] As an optional implementation, steps 2-3 specifically include: For a graph fusion model FG containing rich information, useful feature information can be extracted to build a graph fusion spectrum to achieve feature-level data fusion of the graph fusion model. The graph signal of the graph fusion model is subjected to graph Fourier transform (GFT) operation to obtain a multi-channel graph frequency domain feature matrix. Then, by introducing the coherence of the signal as a feature-level fusion method, the channel fusion spectrum is obtained through the multi-channel graph frequency domain feature matrix.

[0071] The graph fusion model FG has a graph structure similar to the path graph model. The adjacency matrix W is constructed using the path graph structure, and the Laplacian matrix L of the graph fusion model FG is defined as shown in formula (1). The orthogonal eigenvectors of the Laplacian matrix are derived from [ ; [Represents] the graph signal defined on the graph fusion model FG. It can be represented in the following matrix form: (4).

[0072] in, , indicating the first in the graph fusion model The graph signal value of the vertex at time t. Define the graph signal. Then, the envelope signal can be re-expressed using formula (5). : (5).

[0073] in, , representing the elements used to obtain the multi-channel graph frequency domain feature matrix U; express The complex conjugate; Indicates the first The value of the orthogonal eigenvector of the Laplacian matrix of the graph fusion model corresponding to each graph frequency at time n; In the graph fusion model, the first The first moment i The signal values ​​of each channel. Represented as: (6).

[0074] in, This represents the frequency domain characteristic matrix of a multi-channel graph, which is used as a graph signal. A bridge to obtain the spectrum of fused graphics.

[0075] Step 4: Through signal coherence The graph signals in the multi-channel graph frequency domain feature matrix are fused to obtain the channel fused graph spectrum (FGS).

[0076] Unlike performing GFT on a single graph signal to obtain a spectrum, performing GFT on a multi-channel graph signal requires further operations to obtain a spectrum.

[0077] As an optional implementation, step 4 specifically includes: By calculating the coherence and cross-power spectra of neighboring signals, graph fusion models can achieve information fusion of different signals, thereby extracting useful feature information from the signals. Unlike the Fourier Transform (FT), the process of multiplying Fourier coefficients by their conjugate complex numbers results in the loss of some feature information. To address the problem of feature information loss, the average coherence of neighboring signals is introduced as a feature-level fusion method to design the graph fusion spectrum. Here, to better measure the coherence of the signal at each frequency, the coherence is improved to average correlation, calculated as follows: (7).

[0078] in ， Indicates the first i The first channel and the first The first channel in the The signal coherence value at each frequency; ,and ; Indicates the first i The first channel in the Image frequency place, and the first The first channel in the Image frequency Cross-power spectral density at; Indicates the first i The first channel in the Image frequency The self-power spectral density at the location; Indicates the first The first channel in the Image frequency The self-power spectral density at that location. Channel fusion map. As shown in Equation 8 below: (8).

[0079] in, Indicates the first Channel fusion spectrum values ​​at individual frequency points; , and These represent the first and second channels in the [missing information - likely a number] section. The signal coherence value at the frequency of the first image, the second channel and the third channel at the second image. The signal coherence value at the frequency of the first graph and the first q- 1 channel and the first q The first channel in the The signal coherence value at each frequency. This represents the modulo-square operation. This represents the dot product operator. Each frequency corresponds to a frequency in the Fourier spectrum of the vibration signal. The frequency can be determined by the graph frequency. To establish, It is the sampling frequency.

[0080] As an optional implementation, assume the number of channels q =2 (two sensors), total number of time points N=5 (5 sampling time points).

[0081] Step 1: Acquire multi-channel vibration signals. The envelope signal data of the two channels are shown in Table 1 below.

[0082] Table 1. Envelope signal data for the two channels

[0083] Step 2: ① Construct a graph fusion model.

[0084] According to Table 1, the graph fusion model... , of which each It is a column vector: .

[0085] ② Construct the graph signal matrix According to step 1, we can obtain: =[ ],and The number of rows in the signal matrix equals the number of channels q=2, and the number of columns equals the number of sampling times (sampling points) N=5.

[0086] = .

[0087] ③ Construct the Laplace matrix.

[0088] Adjacency matrix W (path graph structure, connecting adjacent points): .

[0089] Degree matrix D: .

[0090] Laplace matrix = .

[0091] ④ Calculate the orthogonal eigenvectors of the Laplace matrix.

[0092] .

[0093] therefore, , , , , .

[0094] Step 3: Graph Fourier Transform.

[0095] .

[0096] Similarly, we can obtain: .

[0097] Based on the extracted image frequency domain features of each channel, a fused feature information matrix is ​​constructed.

[0098] .

[0099] Step 4: Calculate the channel fusion map.

[0100] ① The signal coherence value is calculated.

[0101] hour: , , ,so .

[0102] Similarly, the calculated coherence values ​​for all signals are 1.

[0103] ② The channel fusion map was calculated.

[0104] , , , , .

[0105] Step 5: Construct a training dataset Tr using partial channel fusion maps, and use the random forest method to classify Tr to obtain the importance weights corresponding to each frequency domain feature of the channel fusion map.

[0106] Step 6: Calculate the Health Indicator (HI).

[0107] As an optional implementation, steps 5-6 specifically include: After the fusion map features are constructed, Random Forest (RF) is used to evaluate the importance of the features and determine the weights of different features to improve the reliability of the Health Index (HI). When evaluating the importance of each feature in the fusion map, the Random Forest algorithm is used to classify the constructed training dataset, and the importance of each feature in the channel fusion map is indirectly obtained by optimizing the classification results.

[0108] First, we need to create a dataset for the classification task. Let's set the training dataset as follows: (9).

[0109] in, Let e ​​be the category label of the e-th sample. 0 represents the normal state. 1 represents an abnormal state; E represents the total number of samples in the training dataset; Composed of frequency domain features from the channel fusion spectrum, defined as follows: (10).

[0110] in, This represents the normalized channel fusion map of the e-th sample; The fault characteristic frequency; This refers to frequency resolution, ensuring focus on the main fault frequencies and their harmonics. A random forest method is used to classify this dataset, with the classification objective being: (11).

[0111] Here, θ represents all the parameters of the random forest model. It is the prediction function of the random forest classifier; This is the classification error loss function, such as cross-entropy loss. After the random forest classification training is completed, the feature importance score can be obtained, that is, the score of the key frequency domain feature set. Importance weights corresponding to key frequency domain features : (12).

[0112] in, The number of decision trees in the random forest; Indicates the first Variables in a tree Importance to classification tasks (e.g., Gini index or information gain); Indicates the first The total number of nodes in the tree. After obtaining the feature importance scores, HI can be calculated as a weighted sum: (13).

[0113] in, This represents the health indicator value for the m-th time period; K represents the preset quantity. Represents the key frequency domain feature set. The importance weights corresponding to each key frequency domain feature; This represents the normalized channel fusion map of the m-th time period. The eigenvalues ​​of key frequency domain features.

[0114] Step 7: Detect minor faults in the health indicator HI using the 3σ principle, and detect serious faults in the health indicator HI using the Mann-Kendall method, thereby dividing the operation phase of HI.

[0115] To accurately identify abnormal states during equipment operation, this application, after calculating health indicators, further employs an anomaly detection method based on statistical analysis and trend detection to determine whether the equipment has entered a fault state. The 3σ principle is used for minor fault detection, and the Mann-Kendall trend test (MK Test) is introduced to identify the occurrence of serious faults.

[0116] In the early stages of equipment operation, HI fluctuations are typically small. Therefore, a first preset threshold can be established using the mean and standard deviation of the health status. When HI exceeds the first preset threshold, the device is considered to have entered a minor fault state.

[0117] (14).

[0118] (15).

[0119] After the equipment enters a minor fault state, the Mann-Kendall trend test is used to detect the changing trend of health indicators, thereby identifying the moment when a serious fault occurs. The principle of the Mann-Kendall trend test is to calculate the difference between each pair of data points in the time series and count the number of positive and negative trends. By detecting the moment when the HI trend rises significantly, it is determined that the equipment has entered a serious fault stage.

[0120] (16).

[0121] (17).

[0122] in, R represents the Mann-Kendall statistic for the current time period, which is the cumulative sum of trend directions among all sample points in the entire time series; R represents the total number of time periods. Represents a symbolic function; Let S represent the health indicator value for time period r. If S is positive, it indicates that HI is generally trending upward, and the equipment may be entering a fault state; if S is negative, it indicates that HI is generally trending downward, and the equipment's health status may be recovering; if S is close to 0, it indicates that HI has not changed significantly. When R is large, the Mann-Kendall statistic S approximately follows a normal distribution with a mean of 0 and a variance as follows: (18).

[0123] Therefore, its standardized statistic Z is: (19).

[0124] in, This represents the standardized statistic of S; This represents the variance of S.

[0125] Then a significance test is performed. The Mann-Kendall test is based on the p-value, with α being the significance level. If If p < α, then the trend change is significant. If p ≥ α, then the trend change is not significant. Therefore, the condition for judging a serious fault point is: and .

[0126] Step 8: Experimental verification.

[0127] Example 1: Bearing Dataset Experiment.

[0128] The method of this application was validated using the XJTU-SY public bearing dataset from Xi'an Jiaotong University. Vibration signal data from bearing 1-1 was used. The bearing model is LDK UER 204, and the parameters are shown in Table 2 below. Three types of operating conditions were designed for the experiment, with five bearings in each condition. The first set of bearing data from condition 1 was selected to validate the method of this application. In the test, the bearing rotation frequency was 35Hz, and the signal sampling frequency was 25.6kHz. A total of 123 sets of monitoring data were collected, with each sample lasting 1.28 seconds and spaced one minute apart. The characteristic frequency of the outer race fault can be calculated as follows: =107.91Hz.

[0129] Table 2 Bearing Parameter Table

[0130] use Figure 2 The method described above analyzes and processes the vibration signal data of bearing 1-1 in the horizontal and vertical directions. First, the envelope signal is calculated. Then, a graphical model is constructed according to the path graph pattern to achieve data-level fusion of multi-channel signals. Next, the graphical signal is obtained through the Laplace matrix. The coherence of the signal is calculated using formula (7). Finally, the coherence is used to fuse the multi-channel graph frequency domain feature matrix U obtained after the Laplace transform into a channel fusion graph FGS. The FGS is then normalized to obtain NFGS for the subsequent construction of HI.

[0131] The data in files 1-3 of the normalized fusion graph are taken as normal data, while the data in files 67-69 are taken as abnormal data. The training dataset is constructed using the method in steps 5-6. The parameters of the random forest method are: in the random forest classifier, the number of trees is set to 100 (n_estimators=100), the maximum depth is set to None, the minimum number of samples in the leaf nodes is 2, and the default Gini index (criterion='gini') is used for splitting. All parameters are tuned on the validation set based on grid search (GridSearchCV). The optimized weights of the normalized fusion graph are obtained by classifying the training data, and HI is calculated by formula (13). Anomaly detection is performed on the health indicators. Early fault points are detected in file 68, and serious fault points are detected in file 80. Therefore, the health operation stages of bearing 1-1 can be divided as follows: 0-68 is the normal operation stage, 69-80 is the minor fault stage, and 81-123 is the serious fault stage, such as Figure 3As shown. To illustrate the superiority of the method in this application, the RMS method is used as a typical traditional HI, the Gini index and negative entropy are used as commonly used HIs, and the HI constructed by the spectral energy weighting method is used as a comparative method. After constructing the HI, the 3σ principle and the Mann-Kendall method are still used to detect early and severe fault points of the HI, and the health stages of the equipment are divided accordingly. The comparison results are shown below. Figure 4 As shown, where, Figure 4 (a) in the figure is a curve of bearing health index based on root mean square value; Figure 4 (b) in the figure is a curve of bearing health index based on the Gini index; Figure 4 (c) in the figure is a bearing health index curve based on negative entropy; Figure 4 (d) in the figure is a curve of bearing health index based on spectral energy weighting.

[0132] Depend on Figure 4 As shown in (a), the HI algorithm built based on RMS (Root Mean Square) is insensitive to early faults. Early fault points are detected in file 77, while severe fault points are located in file 79. This results in the HI algorithm being almost unable to identify minor fault stages. Figure 4 As shown in (b), the HI algorithm based on the Gini index detected an early fault point in document 74 and a severe fault point in document 78, but it also struggled to identify minor fault stages. Figure 4 As shown in (c), HI based on negative entropy is significantly affected by noise, resulting in large fluctuations in HI. This leads to early fluctuations being misjudged as minor fault points, while serious fault points are masked by noise, causing recognition lag. Figure 4 As shown in (d), the HI based on spectral energy weighting is constructed by weighting the power spectral energy on the basis of the normalized fusion spectrum. It can be seen that it is affected by noise and cannot accurately identify early fault points.

[0133] To further quantify the superiority of the HI constructed in this application, the monotonicity, trend and robustness of the method in this application and the comparative method were calculated, as shown in Table 3 below.

[0134] Table 3 Comparison of Monotonicity, Trendality and Robustness of Bearing HI

[0135] Table 3 shows that RMS is the most stable and highest-scoring health indicator overall, but when combined with... Figure 3 The analysis results show that, since RMS is not sensitive to early failures and can hardly identify the early failure stage, the method of this application is more advantageous in terms of overall performance.

[0136] Example 2: Degradation experiment of translation axis of high precision vertical machining center.

[0137] This embodiment of the experiment was conducted on the X-axis of the machining center under service conditions, and the experiment lasted for the entire maintenance cycle of the X-axis. A schematic diagram of the X-axis is shown below. Figure 5 As shown, Figure 5 G#1, G#2, G#3, and G#4 in the figure represent four different gears with 24, 31, 31, and 56 teeth respectively. The experiment was conducted during the entire maintenance period of the X-axis, with a feed rate of 550 mm / min and stable table feed without cutting. The three-phase AC current signal of the servo motor was measured synchronously by three current sensors, and the motor torque was obtained by the DQ transformation method of equations (20) and (21). The total experimental time was 192 days, with a data acquisition interval of approximately 26 days. The sampling frequency of each data point was 1000 Hz. Eight days of equivalent torque data samples were taken. Each data sample was a collected data series containing 60,000 data points. Each file was divided into 60 files of 1,000 points each.

[0138] (20).

[0139] (twenty one).

[0140] in, It represents the quadrature-axis current, in a rotating coordinate system, the component perpendicular to the direction of the rotor magnetic field, which is proportional to the electromagnetic torque of the motor; This represents the direct-axis current, and in a rotating coordinate system, the component parallel to the direction of the rotor magnetic field. It is typically used to adjust the magnetic field of a motor (field weakening control). , , These represent the current values ​​before the transformation, which are the actual phase currents of the motor's three-phase windings. The subscripts indicate the current values ​​before the transformation. u , v , w This typically corresponds to the three phases of a motor; This indicates the number of pole pairs in the motor. If the motor has 4 pole pairs (i.e., 8 magnetic poles), then... =4; The electrical angle (or mechanical angle multiplied by the number of pole pairs) of the rotor is usually measured by an encoder; This refers to electromagnetic torque, which is generated by the interaction between the internal magnetic field and the current in the motor. This represents the torque coefficient (or torque constant). This represents the system's moment of inertia; Indicates frictional torque; This indicates the load torque.

[0141] The experiment also simultaneously acquired encoder data, and used a frequency domain weighting method to convert the encoder data into instantaneous angular acceleration (IAA). The discrete IAA sequence was obtained through the following steps: (1) Use FFT (Fast Fourier Transform) to calculate the position sequence after detrending processing. Fourier coefficients: (twenty two).

[0142] Where B is the total length of the position sequence after detrending. For discrete frequencies.

[0143] (2) Multiply each calculated Fourier coefficient by the corresponding weighting factor. ; (twenty three).

[0144] (3) Through calculation The inverse Fourier transform yields the corresponding IAA sequence. Represents discrete frequency The corresponding intermediate variables.

[0145] The IAA signal is divided into 60 files of 1000 points each, corresponding to the equivalent torque. The equivalent torque and IAA signal are used as data from two separate channels. Figure 2 The method described above analyzes and processes the signal. First, the envelope signal is calculated. Then, a graph fusion model is constructed according to the path graph pattern to achieve data-level fusion of multi-channel signals. Next, the graph signal is obtained through the Laplacian matrix. The coherence of the signal is calculated using formula (7). Finally, the coherence is used to fuse the multi-channel graph frequency domain feature matrix U obtained after the Laplace transform into a channel fusion graph FGS. The FGS is then normalized to obtain NFGS for subsequent HI construction.

[0146] Data from files 0-30 in the normalized fusion graph are considered normal data, while data from files 300-330 are considered abnormal data. A training dataset is constructed using the methods in steps 5-6. The optimized weights of the normalized fusion graph are obtained by classifying the training data, and HI is calculated using formula (13). The HI constructed using the method of this application is as follows: Figure 6 As shown, by Figure 6It can be seen that this HI accurately reflects the degradation trend of the translation axis of the high-precision vertical machining center. Anomaly detection of the health indicators revealed an early fault point in file 282 and a severe fault point in file 361. Therefore, the healthy operating stages of the translation axis of the high-precision vertical machining center can be divided as follows: 0-282 represents the normal operating stage, 283-361 represents the minor fault stage, and 362-480 represents the severe fault stage. Based on the data acquisition time and equipment maintenance cycle, it can be determined that the early fault point occurred 89 days before the next maintenance, and the severe fault point occurred 32 days before the next maintenance.

[0147] The RMS method was used as a typical traditional fault indicator (HI), while the Gini index and negative entropy were used as commonly used HIs. The HI constructed using the spectral energy weighting method was also used as a comparison method. The 3σ principle and the Mann-Kendall method were still employed to detect early and severe fault points in all the HIs constructed by the comparison methods, and this was used to classify the operational stages of the translation axis of the high-precision vertical machining center. The comparison results are as follows: Figure 7 As shown, where, Figure 7 (a) in the figure is a health index curve based on the root mean square value and shifted along the axis; Figure 7 (b) in the figure is a shifted axis health index curve based on the Gini index; Figure 7 (c) in the figure is a translation axis health index curve based on negative entropy; Figure 7 (d) in the figure is a translational axis health index curve based on spectral energy weighting. From Figure 7 As can be seen, the HI based on RMS, the HI based on negative entropy, and the HI based on spectral energy weighting all exhibit a certain periodicity, with poor trend and monotonicity. This may be related to the working principle of the translation axis of the high-precision vertical machining center. The ball screw reciprocates during operation, and the impact energy is greatest each time it passes through the fault location. The HI based on the Gini index has relatively better trend and monotonicity, but the fluctuations of these four comparison methods are all large, and they are greatly affected by noise interference. Another reason is the redundancy and contradiction in the dual-channel data, which these comparison methods cannot handle well. The method in this application can solve the problems of redundancy and contradiction in multi-channel data and reduce the impact of noise interference. The method in this application can accurately reflect the true degradation trend of the translation axis of the high-precision vertical machining center, providing a basis for subsequent RUL prediction.

[0148] Further quantitative comparisons were made of the HI constructed by these methods, and their monotonicity, trend, and robustness were calculated respectively. The calculation results are shown in Table 4 below. It can be seen that the method of this application has good scores in monotonicity, trend, and robustness, indicating that the method of this application can better reflect the true degradation trend of the translation axis of the high-precision vertical machining center in practical applications, and accurately divide its degradation stage into normal operation stage, minor fault stage, and severe fault stage, providing a new idea for the construction of health indicators for multi-source data fusion.

[0149] Table 4. Comparison of Monotonicity, Trendality, and Robustness of Translation Axis HI

[0150] In summary, based on the current trend of multi-faceted and comprehensive monitoring of equipment in industrial production, this application proposes a method for constructing health indicators (HI) based on multi-channel fusion graphs. This method utilizes multi-source data collected by multi-channel sensors to construct HI, accurately reflecting the current health status and degradation trend of equipment. The method first transforms the multi-channel data fusion problem into a graph model fusion problem. Multi-channel envelope signals are used in the path graph model, and vector concatenation is employed as a data-level fusion method to fuse these path graph models. Secondly, a graph Fourier transform operation is performed on the graph signals of the graph fusion model to obtain a fusion feature information matrix, which serves as a computational bridge. Simultaneously, the coherence of adjacent signals is used as a feature-level fusion method to obtain the fused graph, demonstrating excellent performance in feature extraction. Thirdly, the fused graph is introduced into a sparse metric framework, and the optimization weights are calculated using a random forest method. Finally, the fused graph is weighted and summed using these optimization weights to obtain the health indicator. The effectiveness of the proposed method was verified using a bearing dataset and a high-precision vertical machining center translation axis dataset. Fault point detection and monotonicity, trend, and robustness were compared between the HI constructed using the proposed method and those constructed using a comparative method. This demonstrates that the proposed method can better reflect equipment degradation trends and anomalies. The proposed method provides an effective approach for health monitoring of multi-channel signals, offering theoretical basis and technical support for equipment lifecycle health management.

[0151] Based on the same inventive concept, this application also provides a health status monitoring device for rotating machinery to implement the aforementioned health status monitoring method. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the health status monitoring device for rotating machinery provided below can be found in the limitations of the health status monitoring method for rotating machinery described above, and will not be repeated here.

[0152] In one exemplary embodiment, such as Figure 8As shown, a health status monitoring device for rotating machinery is provided, comprising: The signal acquisition module 201 is used to acquire multi-channel vibration signals of the rotating mechanical equipment to be monitored within the current time period.

[0153] The graph construction module 202 is used to construct a graph fusion model based on the multi-channel vibration signal.

[0154] The frequency domain feature extraction module 203 is used to extract the graph frequency domain features of each channel based on the graph fusion model by using the graph Fourier transform method, so as to obtain the multi-channel graph frequency domain feature matrix for the current time period.

[0155] The channel fusion module 204 is used to fuse the frequency domain features of each channel in the multi-channel graph frequency domain feature matrix by using the signal coherence between adjacent channels as fusion weights to obtain the channel fusion graph for the current time period.

[0156] The weight calculation module 205 is used to calculate the importance weight of each frequency domain feature in the channel fusion map relative to all frequency domain features using the random forest method.

[0157] The health indicator construction module 206 is used to perform weighted fusion of all frequency domain features in the channel fusion map based on the importance weights corresponding to each frequency domain feature, so as to obtain the health indicators of the rotating mechanical equipment to be monitored in the current time period.

[0158] The health status classification module 207 is used to classify the health status of the rotating machinery to be monitored based on its health indicators in the current time period using statistical analysis methods, and to obtain the health status monitoring results of the rotating machinery to be monitored in the current time period. The statistical analysis methods include the 3σ principle detection method and the Mann-Kendall trend test method. The health status monitoring results include normal status, minor fault status, and severe fault status.

[0159] In one exemplary embodiment, a computer system is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 9As shown, the computer system includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores health status monitoring and processing data for rotating machinery. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for monitoring the health status of rotating machinery.

[0160] Those skilled in the art will understand that Figure 9 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer system to which the present application is applied. A specific computer system may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer system is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0161] In one exemplary embodiment, a computer program product is provided, which stores a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0162] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0163] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0164] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0165] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0166] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A health condition monitoring method of a rotating machinery equipment, characterized by, include: Acquire multi-channel vibration signals of the rotating mechanical equipment to be monitored within the current time period; Based on the multi-channel vibration signals, a graph fusion model is constructed; Based on the graph fusion model, the graph frequency domain features of each channel are extracted using the graph Fourier transform method to obtain the multi-channel graph frequency domain feature matrix for the current time period. Using the signal coherence between adjacent channels as fusion weights, the frequency domain features of each channel in the multi-channel graph frequency domain feature matrix are fused to obtain the channel fusion graph for the current time period. The importance weight of each frequency domain feature relative to all frequency domain features in the channel fusion map is calculated using the random forest method. Based on the importance weights corresponding to each frequency domain feature, all frequency domain features in the channel fusion map are weighted and fused to obtain the health index of the rotating mechanical equipment to be monitored in the current time period. Based on the health indicators of the rotating machinery to be monitored in the current time period, a statistical analysis method is used to classify the health status, and the health status monitoring results of the rotating machinery to be monitored in the current time period are obtained. The statistical analysis method includes the 3σ principle detection method and the Mann-Kendall trend test method. The health status monitoring results include normal status, minor fault status and severe fault status.

2. The health state monitoring method of a rotating machinery apparatus according to claim 1, characterized by, Based on the multi-channel vibration signals, a graph fusion model is constructed, specifically including: The multi-channel envelope signal is obtained by performing a Hilbert transform on the multi-channel vibration signal using the following formula: ； wherein, represents the envelope signal of the n represents the envelope signal of the i channel at time instant represents the vibration signal of the n represents the vibration signal of the i channel at time instant represents a modulo operation; j represents the imaginary unit; represents a Hilbert transform of ; Construct a graph fusion model using the following expression: ； ； in, Representation graph fusion model; , , and Representing time 1, time 2, and time 3 respectively. Time and the N The vertex vector at time t. N Indicates the total number of moments; , , and They represent the first Time slot 1, 2, 3 i The first channel and the first q The envelope signal of each channel, q This indicates the total number of channels.

3. The method for monitoring the health status of rotating mechanical equipment according to claim 2, characterized in that, Based on the graph fusion model, the graph frequency domain features of each channel are extracted using the graph Fourier transform method to obtain the multi-channel graph frequency domain feature matrix for the current time period, specifically including: The frequency domain features of each channel are extracted using the following expression: ； in, Indicates the first The frequency of the first graph i The frequency domain characteristics of each channel; express The complex conjugate; Indicates the first The orthogonal eigenvectors of the Laplacian matrix of the graph fusion model corresponding to each graph frequency are in the th order. n The value at time; In the graph fusion model, the first The first moment i The signal values ​​of each channel; Based on the extracted image frequency domain features of each channel, the multi-channel image frequency domain feature matrix for the current time period is obtained through the following expression: ； ； in, This represents the frequency domain feature matrix of the multi-channel graph for the current time period; , and Representing the frequency of the first graph, the second graph, and the third graph respectively. The frequency of the first graph and the first The frequency domain characteristics of each graph frequency; , and They represent the first The first, second, and third channels of the image frequency. q The frequency domain characteristics of each channel.

4. The method for monitoring the health status of rotating mechanical equipment according to claim 3, characterized in that, Using the signal coherence between adjacent channels as fusion weights, the frequency domain features of each channel in the multi-channel graph frequency domain feature matrix are fused to obtain the channel fusion graph for the current time period, specifically including: The signal coherence between adjacent channels is calculated using the following formula: ； in, Indicates the first i The first channel and the first The first channel in the The signal coherence value at each frequency; ,and ; Indicates the first i The first channel in the Image frequency place, and the first The first channel in the Image frequency Cross-power spectral density at; Indicates the first i The first channel in the Image frequency The self-power spectral density at the location; Indicates the first The first channel in the Image frequency The self-power spectral density at the location; The following formula is used to fuse the frequency domain features of each channel in the multi-channel graph frequency domain feature matrix, using the signal coherence between adjacent channels as fusion weights, to obtain the channel fusion graph for the current time period: ; in, Indicates the first Channel fusion spectrum values ​​at individual frequency points; , and These represent the first and second channels in the [missing information - likely a number] section. The signal coherence value at the frequency of the first image, the second channel and the third channel at the second image. The signal coherence value at the frequency of the first graph and the first q -1 channel and the first q The first channel in the The signal coherence value at each frequency.

5. The method for monitoring the health status of rotating mechanical equipment according to claim 1, characterized in that, Based on the importance weights corresponding to each frequency domain feature, all frequency domain features in the channel fusion spectrum are weighted and fused to obtain the health indicators of the rotating machinery to be monitored in the current time period, specifically including: Normalize the channel fusion map for the current time period to obtain the normalized channel fusion map for the current time period. Based on the fault characteristic frequency and frequency resolution, a first preset number of frequency domain features are selected from the normalized channel fusion map to obtain a set of key frequency domain features. The health index of the rotating machinery to be monitored in the current time period can be obtained using the following formula: ； in, Indicates the first m Health indicator values ​​for a given time period; K represents the preset quantity; Represents the key frequency domain feature set. The importance weights corresponding to each key frequency domain feature; Indicates the first m In the normalized channel fusion map of the time period, the first The eigenvalues ​​of key frequency domain features.

6. The method for monitoring the health status of rotating mechanical equipment according to claim 5, characterized in that, Based on the health indicators of the rotating machinery to be monitored in the current time period, statistical analysis methods are used to classify the health status, and the monitoring results of the health status of the rotating machinery to be monitored in the current time period are obtained, specifically including: If the current time period is within the early operating time period, or the health index of the current time period is less than or equal to the first preset threshold, then the health status monitoring result of the rotating machinery to be monitored in the current time period is normal; the early operating time period includes the period from the first time period when the rotating machinery to be monitored starts operating to the second preset number of time periods; the first preset threshold is calculated based on the health index of the early operating time period using the 3σ principle detection method; If the health index for the current time period is greater than the first preset threshold, and the Mann-Kendall trend test result for the current time period shows no statistically significant upward trend, then the health status monitoring result of the rotating machinery to be monitored in the current time period is a minor fault state; the Mann-Kendall trend test result is determined by the Mann-Kendall trend test method based on the health index of the current time period and all time periods before the current time period. If the health index for the current time period is greater than the first preset threshold, and the Mann-Kendall trend test result for the current time period shows a statistically significant upward trend, then the health status monitoring result of the rotating machinery to be monitored during the current time period is a serious fault state.

7. The method for monitoring the health status of rotating mechanical equipment according to claim 6, characterized in that, Based on health indicators for the current time period and all time periods preceding it, the Mann-Kendall trend test result for the current time period is determined using the Mann-Kendall trend test method, specifically including: The Mann-Kendall statistic for the current time period is determined using the following expression: ； ； in, R represents the Mann-Kendall statistic for the current time period; R represents the total number of time periods. Represents a symbolic function; Indicates the first r Health indicator values ​​over a period of time; The standardized statistic of S is calculated using the following formula: ； ； in, This represents the standardized statistic of S; This represents the variance of S; like ,or If the Mann-Kendall trend test result for the current time period is that there is no statistically significant upward trend; among which, express The absolute value; This represents the critical value for a two-tailed test of the standard normal distribution at a pre-set significance level α. like ,and If the Mann-Kendall trend test result for the current time period is that there is a statistically significant upward trend.

8. A health status monitoring device for rotating mechanical equipment, characterized in that, The health status monitoring device for rotating machinery uses the health status monitoring method for rotating machinery according to any one of claims 1-7, and the health status monitoring device for rotating machinery comprises: The signal acquisition module is used to acquire multi-channel vibration signals of the rotating mechanical equipment to be monitored within the current time period; The graph construction module is used to construct a graph fusion model based on the multi-channel vibration signal; The frequency domain feature extraction module is used to extract the graph frequency domain features of each channel based on the graph fusion model by using the graph Fourier transform method, so as to obtain the multi-channel graph frequency domain feature matrix for the current time period. The channel fusion module is used to fuse the frequency domain features of each channel in the multi-channel graph frequency domain feature matrix by using the signal coherence between adjacent channels as fusion weights to obtain the channel fusion graph for the current time period. The weight calculation module is used to calculate the importance weight of each frequency domain feature in the channel fusion map relative to all frequency domain features using the random forest method; The health indicator construction module is used to perform weighted fusion of all frequency domain features in the channel fusion map based on the importance weights corresponding to each frequency domain feature, so as to obtain the health indicators of the rotating mechanical equipment to be monitored in the current time period. The health status classification module is used to classify the health status of the rotating machinery to be monitored based on its health indicators in the current time period using statistical analysis methods, and to obtain the health status monitoring results of the rotating machinery to be monitored in the current time period. The statistical analysis methods include the 3σ principle detection method and the Mann-Kendall trend test method. The health status monitoring results include normal status, minor fault status, and severe fault status.

9. A computer system, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the health status monitoring method for rotating mechanical equipment according to any one of claims 1-7.

10. A computer program product having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the health status monitoring method for rotating mechanical equipment as described in any one of claims 1-7.

Images