Gearbox drive shaft signal data acquisition and processing system based on machine learning

Abstract

The present invention relates to the field of abnormality detection technology, and in particular to a data acquisition and processing system for a transmission shaft signal based on machine learning; obtaining a local fluctuation factor and an instantaneous change stability factor according to data features in a preset first neighborhood window and a preset second neighborhood window of a data point; clustering is performed according to the local fluctuation factor and the instantaneous change stability factor, and a reference cluster is obtained according to the number of data points in the cluster cluster; an abnormal cluster is obtained according to the data difference feature between the reference cluster and other cluster clusters; an abnormal index is obtained according to the distance feature between the abnormal cluster and the reference cluster, the data discrete feature in the abnormal cluster, and the local density feature of the abnormal data points in the abnormal cluster; and the score of the data point is obtained according to the isolation forest algorithm. The present invention obtains a modified score according to the abnormal index and the score and detects the transmission shaft, thereby improving the detection accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormality detection, and in particular to a gearbox drive shaft signal data acquisition and processing system based on machine learning. Background Art

[0002] The transmission shaft of the gearbox is responsible for transmitting the power generated by the engine to the wheels, and is an important component for driving the vehicle. The vibration signal of the transmission shaft can reflect faults such as power imbalance and gear wear in the transmission system of the gearbox. By analyzing the vibration signal, abnormal faults can be identified in time to improve the driving stability and safety of the vehicle. Since the fault characteristics in the early stage of the fault are not obvious and there are few abnormal vibration data, the isolation forest algorithm is usually used to detect a few abnormal data in the data set; however, due to the regular changes in the vibration signal, an abnormal value may be similar to the normal value at other times. The obtained transmission shaft vibration signal is prone to the situation where the boundary between the abnormal value and the normal value is not obvious, resulting in the poor isolation effect of the traditional isolation forest algorithm on the abnormal value, which ultimately affects the accuracy of the transmission shaft fault detection. Summary of the invention

[0003] In order to solve the above technical problem that the existing isolation forest algorithm has low detection accuracy for vibration signals, the purpose of the present invention is to provide a gearbox transmission shaft signal data acquisition and processing system based on machine learning. The technical solution adopted is as follows:

[0004] A data acquisition module, used to obtain the vibration intensity time series of the transmission shaft after removing high-frequency noise;

[0005] A vibration analysis module is used to construct a preset first neighborhood window and a preset second neighborhood window of data points in the vibration intensity time series; obtain a local fluctuation factor according to the data fluctuation characteristics within the preset first neighborhood window and the preset second neighborhood window of the data points; and obtain an instantaneous change stability factor according to the adjacent data change characteristics within the preset first neighborhood window and the preset second neighborhood window of the data points;

[0006] A feature processing module is used to cluster the data points according to the local fluctuation factor and the instantaneous change stability factor to obtain different clusters; obtain a reference cluster according to the number characteristics of the data points in the cluster; obtain an abnormal cluster according to the data difference characteristics between the reference cluster and other clusters; obtain a first abnormality degree according to the distance characteristics between the abnormal cluster and the reference cluster and the data discrete characteristics in the abnormal cluster; obtain a second abnormality degree according to the local density characteristics of the abnormal data points in the abnormal cluster; obtain an abnormal index of the abnormal data point according to the first abnormality degree and the second abnormality degree;

[0007] The anomaly detection module is used to obtain the score of the data points in the vibration intensity time series according to the isolation forest algorithm; obtain the corrected score according to the anomaly index and the score; and detect the transmission shaft according to the corrected score.

[0008] Furthermore, the step of constructing a preset first neighborhood window and a preset second neighborhood window of data points in the vibration intensity time series includes:

[0009] In the vibration intensity time series, a preset first neighborhood window and a preset second neighborhood window are constructed by taking the data point as the window center, and the length of the preset second neighborhood window is an odd multiple of the length of the preset first neighborhood window.

[0010] Furthermore, the step of obtaining the local fluctuation factor according to the data fluctuation characteristics within the preset first neighborhood window and the preset second neighborhood window of the data point includes:

[0011] Calculate the difference between the maximum and minimum values ​​in the preset first neighborhood window and normalize it to obtain a first extreme value difference; calculate the product of the variance of the data in the preset first neighborhood window and the first extreme value difference to obtain a first fluctuation characteristic value; evenly divide the preset second neighborhood window into sub-windows with the same length as the preset first neighborhood window; calculate the difference between the maximum and minimum values ​​in the sub-window and normalize it to obtain a second extreme value difference; calculate the product of the variance of the data in the sub-window and the second extreme value difference to obtain a second fluctuation characteristic value; calculate the sum of the absolute values ​​of the difference between the first fluctuation characteristic value and the second fluctuation characteristic value of each sub-window to obtain a fluctuation difference value; calculate the product of the first fluctuation characteristic value and the fluctuation difference value and positively correlate them to obtain a local fluctuation factor of the data point.

[0012] Furthermore, the step of obtaining the instantaneous change stability factor according to the adjacent data change characteristics within the preset first neighborhood window and the preset second neighborhood window of the data point includes:

[0013] Calculate the variance of the instantaneous change rate of the data in the preset first neighborhood window to obtain the first instantaneous change characteristic value; calculate the variance of the instantaneous change rate of the data in the preset second neighborhood window to obtain the second instantaneous change characteristic value; calculate the absolute value of the difference between the first instantaneous change characteristic value and the second instantaneous change characteristic value and negatively map them to obtain the instantaneous change stability factor of the data point.

[0014] Furthermore, the step of obtaining a reference cluster according to the quantity characteristics of the data points in the clustering clusters includes:

[0015] The cluster with the most data points in the cluster is used as the reference cluster.

[0016] Furthermore, the step of obtaining an abnormal cluster according to the data difference characteristics between the benchmark cluster and other clustering clusters includes:

[0017] The data points in the cluster are sorted in chronological order to obtain the intra-cluster data sequence; the dynamic time-warped distance between the benchmark cluster and the intra-cluster data sequence of other clusters is calculated and normalized to obtain the difference between the benchmark cluster and the other clusters; the other clusters whose difference exceeds a preset difference threshold are regarded as the abnormal clusters.

[0018] Furthermore, the step of obtaining the first abnormality degree according to the distance feature between the abnormal cluster and the reference cluster and the discrete feature of the data in the abnormal cluster comprises:

[0019] The Euclidean distance between the cluster center of the abnormal cluster and the benchmark cluster is calculated in the cluster space to obtain the difference distance; the coefficient of variation of the local fluctuation factor of the data points in the abnormal cluster and the average value of the coefficient of variation of the instantaneous change stability factor are calculated; the discrete eigenvalue is obtained; the product of the difference distance and the discrete eigenvalue is calculated and normalized to obtain the first abnormality degree of the abnormal cluster.

[0020] Furthermore, the step of obtaining a second abnormality degree according to the local density characteristics of the abnormal data points in the abnormal cluster includes:

[0021] In the clustering space, the average value of the Euclidean distances between the abnormal data point in the abnormal cluster and a preset number of other abnormal data points closest to the abnormal data point is calculated and normalized to obtain a second abnormality degree of the abnormal data point.

[0022] Further, the step of obtaining an abnormality index of the abnormal data point according to the first abnormality degree and the second abnormality degree includes:

[0023] The product of the second abnormality degree of the abnormal data point and the first abnormality degree of the abnormal cluster is calculated to obtain a comprehensive abnormality degree; the sum of the comprehensive abnormality degree and a preset constant is calculated to obtain an abnormality index of the abnormal data point; the preset constant is not less than 1.

[0024] Furthermore, the step of obtaining a revised score according to the abnormality index and the score comprises:

[0025] The product of the abnormal index of the abnormal data point and the corresponding score is calculated to obtain a revised score of the abnormal data point; for non-abnormal data points, the score is used as the revised score.

[0026] The present invention has the following beneficial effects:

[0027] In the present invention, constructing a preset first neighborhood window and a preset second neighborhood window can be used to analyze the local neighborhood data of the data point; obtaining a local fluctuation factor can reflect the data fluctuation characteristics of the neighborhood of the data point, thereby distinguishing normal and abnormal data points according to the local fluctuation factor; obtaining an instantaneous change stability factor can reflect the regularity of the change of the adjacent data of the neighborhood of the data point, thereby distinguishing normal and abnormal data points according to the regularity of the change of the adjacent data. Obtaining clusters can distinguish data representing different vibration intensity characteristics, which is convenient for determining the degree of abnormality of abnormal data points; obtaining a reference cluster can determine the position of normal data points in the cluster space, which is convenient for determining the degree of abnormality of abnormal data points. Obtaining abnormal clusters can preliminarily determine the range of abnormal data points, thereby improving the accuracy of fault detection. Obtaining the first abnormality degree and the second abnormality degree can characterize the degree of abnormality of abnormal data points according to the location of the abnormal cluster and the abnormal data point, thereby improving the accuracy of the revised score. Obtaining the revised score can reduce the detection error of the isolation forest algorithm, and finally perform fault detection of the transmission shaft according to the revised score, thereby improving the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. 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 paying creative work.

[0029] Figure 1 A block diagram of a gearbox drive shaft signal data acquisition and processing system based on machine learning provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a gearbox transmission shaft signal data acquisition and processing system based on machine learning proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0031] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0032] The specific scheme of a gearbox drive shaft signal data acquisition and processing system based on machine learning provided by the present invention is described in detail below with reference to the accompanying drawings.

[0033] See also Figure 1 , which shows a block diagram of a gearbox transmission shaft signal data acquisition and processing system based on machine learning provided by an embodiment of the present invention, the system includes the following modules:

[0034] The data acquisition module S1 is used to obtain the vibration intensity time series of the transmission shaft after removing the high-frequency noise.

[0035] In an embodiment of the present invention, the implementation scenario is to perform data analysis and fault detection on the transmission shaft of the gearbox to improve the accuracy of fault detection. First, the vibration intensity time series of the transmission shaft after removing high-frequency noise is obtained, and a vibration sensor is installed on the transmission shaft of the gearbox to collect vibration intensity data. The implementer can determine the collection frequency according to the implementation scenario; in order to improve the accuracy of subsequent analysis, it is necessary to remove obvious high-frequency noise; first, the collected sequence is obtained through fast Fourier transform to obtain a spectrum, and the frequency components in the spectrum that are greater than the preset cutoff frequency are set to 0, and the modified spectrum is obtained through inverse Fourier transform to obtain the vibration intensity time series after removing high-frequency noise. The preset cutoff frequency can be determined by the implementer according to the implementation scenario.

[0036] The vibration analysis module S2 is used to construct a preset first neighborhood window and a preset second neighborhood window of data points in the vibration intensity time series; obtain a local fluctuation factor based on the data fluctuation characteristics within the preset first neighborhood window and the preset second neighborhood window of the data points; and obtain an instantaneous change stability factor based on the adjacent data change characteristics within the preset first neighborhood window and the preset second neighborhood window of the data points.

[0037] Since the vibration intensity time series has fluctuations as a whole, some abnormal values ​​may be similar to normal values ​​at other times, resulting in low detection accuracy of the isolated forest algorithm for abnormal data points; and the data features of the abnormal points and the local neighborhood will be different from the data features of the normal points and the local neighborhood, so the local data features of the data points can be analyzed to correct the scores of the data points obtained by the isolated forest algorithm. First, a preset first neighborhood window and a preset second neighborhood window of the data points in the vibration intensity time series are constructed, and the preset first neighborhood window and the preset second neighborhood window can be used to analyze the local neighborhood data features of the data points; preferably, in an embodiment of the present invention, the data point is used as the window center to construct the preset first neighborhood window and the preset second neighborhood window in the vibration intensity time series, and the length of the preset second neighborhood window is an odd multiple of the length of the preset first neighborhood window; in the embodiment of the present invention, the length of the preset first neighborhood window is set to 7, that is, the data point and the 6 data points before and after it are composed, and the length of the preset second neighborhood window is 3 times the length of the preset first neighborhood window, which can be determined by the implementer according to the implementation scenario; data points that do not meet the window construction conditions in the vibration intensity time series do not participate in subsequent calculations.

[0038] Furthermore, under normal circumstances, the fluctuation characteristics of the vibration intensity of the transmission shaft have certain regularity. If the transmission shaft fails, it will cause abnormal values ​​to affect the fluctuation law. Then, the local fluctuation factor can be obtained according to the data fluctuation characteristics in the preset first neighborhood window and the preset second neighborhood window of the data point; preferably, in the embodiment of the present invention, the step of obtaining the local fluctuation factor includes: calculating the difference between the maximum value and the minimum value in the preset first neighborhood window and normalizing it to obtain the first extreme value difference; the first extreme value difference reflects the maximum fluctuation amplitude characteristics in the preset first neighborhood window of the data point. Calculate the product of the variance of the data in the preset first neighborhood window and the first extreme value difference to obtain the first fluctuation characteristic value; the variance reflects the overall vibration fluctuation degree of the local neighborhood of the data point, and the first fluctuation characteristic value characterizes the vibration fluctuation characteristics in the preset first neighborhood window of the data point; when the first fluctuation characteristic value is larger, it means that the local vibration characteristics of the data point are more obvious, and the more likely it is to be abnormal. The preset second neighborhood window is evenly divided into sub-windows of the same length as the preset first neighborhood window. In the embodiment of the present invention, three sub-windows are obtained, one of which overlaps with the preset first neighborhood window; if there is no abnormal data in the local neighborhood of the data point, the vibration fluctuation characteristics of the three sub-windows should be similar, and if there is abnormal data, the vibration fluctuation characteristics will be different. The difference between the maximum and minimum values ​​in the sub-window is calculated and normalized to obtain the second extreme value difference; the second extreme value difference reflects the maximum fluctuation amplitude characteristic of the sub-window. The product of the variance of the data in the sub-window and the second extreme value difference is calculated to obtain the second fluctuation characteristic value. The sum of the absolute values ​​of the difference between the first fluctuation characteristic value and the second fluctuation characteristic value of each sub-window is calculated to obtain the fluctuation difference value; if there is no abnormal data, the second fluctuation characteristic value is similar to the first fluctuation characteristic value, and the fluctuation difference value is small; if the abnormal data is more obvious, the difference between the second fluctuation characteristic value and the first fluctuation characteristic value of two sub-windows is greater, and the fluctuation difference value is greater. Calculate the product of the first fluctuation characteristic value and the difference value and map them in a positive correlation to obtain the local fluctuation factor of the data point; when the local fluctuation factor is larger, it means that the local vibration characteristics of the data point are more abnormal, and the transmission shaft is more likely to be abnormal. The formula for obtaining the local fluctuation factor includes:

[0039]

[0040] In the formula, Q represents the local fluctuation factor of the data point, represents an exponential function with a natural constant as the base, S represents the variance of the data in the preset first neighborhood window, D represents the first extreme value difference, represents the first fluctuation characteristic value, N represents the number of sub-windows, represents the second fluctuation characteristic value of the nth sub-window, Indicates the fluctuation difference value.

[0041] Furthermore, under normal circumstances, although the vibration data obtained is fluctuating as a whole, there is a certain similarity in the change characteristics between adjacent data; if an abnormal value appears in the vibration intensity time series, the change characteristics of the abnormal value and the adjacent data will be different from the normal situation; therefore, the instantaneous change stability factor can be obtained based on the change characteristics of the adjacent data within the preset first neighborhood window and the preset second neighborhood window of the data point.

[0042] Preferably, in an embodiment of the present invention, the step of obtaining the instantaneous change stability factor includes: calculating the variance of the instantaneous change rate of the data in the preset first neighborhood window to obtain the first instantaneous change characteristic value; the instantaneous change rate is the ratio of the difference between the adjacent data and the time, and the first instantaneous change characteristic value reflects the distribution characteristics of the instantaneous change rate of the data in the preset first neighborhood window. Calculate the variance of the instantaneous change rate of the data in the preset second neighborhood window to obtain the second instantaneous change characteristic value; calculate the absolute value of the difference between the first instantaneous change characteristic value and the second instantaneous change characteristic value and negatively correlate and map them to obtain the instantaneous change stability factor of the data point. If there is no abnormal value in the local neighborhood of the data point, the first instantaneous change characteristic value and the second instantaneous change characteristic value will be relatively similar; if the vibration abnormality feature is more obvious, the difference between the first instantaneous change characteristic value and the second instantaneous change characteristic value will be greater, and the instantaneous change stability factor will be smaller.

[0043] The feature processing module S3 is used to cluster the data points according to the local fluctuation factor and the instantaneous change stability factor to obtain different clustering clusters; obtain the benchmark cluster according to the number characteristics of the data points in the clustering cluster; obtain the abnormal cluster according to the data difference characteristics between the benchmark cluster and other clustering clusters; obtain the first abnormality degree according to the distance characteristics between the abnormal cluster and the benchmark cluster and the data discrete characteristics in the abnormal cluster; obtain the second abnormality degree according to the local density characteristics of the abnormal data points in the abnormal cluster; and obtain the abnormal index of the abnormal data points according to the first abnormality degree and the second abnormality degree.

[0044] The local fluctuation factor and instantaneous change stability factor obtained according to the data characteristics of the local neighborhood of the data point can distinguish normal vibration data from abnormal vibration data, so the data points can be clustered according to the local fluctuation factor and instantaneous change stability factor to obtain different cluster clusters; in the embodiment of the present invention, the existing mean shift clustering algorithm is used for clustering, and the specific steps are not repeated. The data points in different cluster clusters reflect different vibration intensity characteristics. Since the amount of normal data in the acquired vibration intensity time series is greater than the amount of abnormal data, the more data points in the cluster cluster, the more the cluster cluster can represent the normal vibration intensity characteristics; therefore, the reference cluster is obtained according to the number of data points in the cluster cluster, specifically including: taking the cluster cluster with the most data points in the cluster cluster as the reference cluster. The data points in the reference cluster represent the most normal vibration intensity characteristics in the vibration intensity time series.

[0045] Furthermore, since there are many vibration intensity features, not all clusters except the reference cluster represent abnormal conditions. In order to improve the accuracy of fault detection, it is necessary to screen out clusters that represent abnormal vibration intensity features from all clusters. The data points in the reference cluster represent normal vibration intensity features and change trends. If the vibration intensity features and change trends of data points in other clusters differ more from those of the reference cluster, it means that the other clusters are more likely to represent abnormal conditions. Therefore, the abnormal cluster is obtained according to the data difference characteristics between the reference cluster and other cluster clusters; preferably, in the embodiment of the present invention, the step of obtaining the abnormal cluster includes: sorting the data points in the cluster cluster in chronological order to obtain the data sequence within the cluster; calculating the dynamic time warping distance between the data sequence within the reference cluster and other cluster clusters and normalizing it to obtain the difference between the reference cluster and other cluster clusters; it should be noted that the dynamic time warping distance is obtained by the existing dynamic time warping algorithm, and the specific calculation steps are not repeated. The greater the difference between the two sequences, the greater the dynamic time warping distance; and then the greater the difference, it means that the greater the difference between the vibration intensity characteristics and the change trend law represented by the reference cluster and the other cluster clusters, the more likely the other cluster clusters are to represent abnormal situations. Other cluster clusters whose difference exceeds the preset difference threshold are regarded as abnormal clusters. In the embodiment of the present invention, the preset difference threshold is 0.8, which can be determined by the implementer according to the implementation scenario.

[0046] After obtaining the abnormal cluster, it is necessary to determine the abnormal degree of the abnormal cluster, thereby improving the accuracy of abnormal detection. In the clustering space, if the distance between the abnormal cluster and the reference cluster is farther, it means that the difference between the abnormal cluster and the reference cluster is greater, and the abnormal degree of the abnormal cluster is greater. If the characteristic distribution of the abnormal data points in the abnormal cluster is more discrete, it means that the vibration characteristics in the abnormal cluster are rarer, and the transmission shaft is more likely to be abnormal. Therefore, the first abnormal degree is obtained according to the distance characteristics of the abnormal cluster and the reference cluster and the data discrete characteristics in the abnormal cluster; preferably, in an embodiment of the present invention, the step of obtaining the first abnormal degree includes: calculating the Euclidean distance between the cluster centers of the abnormal cluster and the reference cluster in the clustering space to obtain the difference distance; when the difference distance is larger, it means that the abnormal degree of the abnormal cluster is greater. Calculate the average value of the coefficient of variation of the local fluctuation factor of the data points in the abnormal cluster and the coefficient of variation of the instantaneous change stability factor; obtain discrete eigenvalues; it should be noted that the coefficient of variation belongs to the prior art, and the specific calculation steps are no longer repeated. The larger the coefficient of variation, the more discrete the data; therefore, the larger the discrete eigenvalue, the greater the abnormal degree of the abnormal cluster. The product of the difference distance and the discrete eigenvalue is calculated and normalized to obtain the first abnormality degree of the abnormal cluster; the larger the first abnormality degree, the higher the abnormality degree of the abnormal cluster as a whole.

[0047] Because there are still certain differences in the degree of abnormality of abnormal data points within the abnormal cluster, in order to further improve the accuracy of abnormality detection, it is necessary to analyze the degree of abnormality of each abnormal data point. If the local density of the abnormal data points within the abnormal cluster is smaller, it can further explain that the vibration intensity characteristics represented by the abnormal data point are rarer, the difference with the vibration intensity characteristics of the reference cluster is greater, and the farther away from the reference cluster. Therefore, the second abnormality is obtained according to the local density characteristics of the abnormal data points within the abnormal cluster; preferably, in an embodiment of the present invention, the step of obtaining the second abnormality includes: calculating the average value of the Euclidean distance between the abnormal data point within the abnormal cluster and the nearest preset number of other abnormal data points in the cluster space and normalizing them to obtain the second abnormality of the abnormal data point. In an embodiment of the present invention, the preset number is 5, which can be determined by the implementer according to the implementation scenario; when the abnormal data point is farther away from the other data points closest to the distance, it means that the abnormal data point is more outlier, the vibration intensity characteristics represented are rarer, and the degree of abnormality is higher.

[0048] Furthermore, after obtaining the first abnormality degree of the abnormal cluster and the second abnormality degree of the abnormal data point, the abnormality degree of each abnormal data point can be accurately characterized in combination with the first abnormality degree and the second abnormality degree. Therefore, the abnormality index of the abnormal data point is obtained according to the first abnormality degree and the second abnormality degree; preferably, in the embodiment of the present invention, the step of obtaining the abnormality index includes: calculating the product of the second abnormality degree of the abnormal data point and the first abnormality degree of the abnormal cluster to obtain the comprehensive abnormality degree; the larger the comprehensive abnormality degree, the more abnormal the vibration intensity characteristics represented by the abnormal data point. Calculate the sum of the comprehensive abnormality degree and the preset constant to obtain the abnormality index of the abnormal data point; the preset constant is not less than the constant 1. In the embodiment of the present invention, the preset constant is 1, the purpose of which is to change the value range of the comprehensive abnormality degree, thereby increasing the score.

[0049] The anomaly detection module S4 is used to obtain the score of the data points in the vibration intensity time series according to the isolation forest algorithm; obtain the corrected score according to the anomaly index and the score; and detect the transmission shaft according to the corrected score.

[0050] After obtaining the abnormal index of the abnormal data point in the abnormal cluster, the abnormal detection result of the isolation forest algorithm can be corrected according to the abnormal index; first, the score of the data point in the vibration intensity time series is obtained according to the isolation forest algorithm. It should be noted that the isolation forest algorithm belongs to the prior art, and the specific calculation steps of the score are not repeated. The larger the score, the more the data point represents the vibration anomaly. Further, a corrected score can be obtained according to the abnormal index and the score; preferably, in an embodiment of the present invention, the step of obtaining the corrected score includes: calculating the product of the abnormal index of the abnormal data point and the corresponding score to obtain a corrected score of the abnormal data point; the corrected score of the abnormal data point is larger than the original score, thereby improving some abnormal but lower-scoring data points. For non-abnormal data points, the score is used as the corrected score without adjustment.

[0051] Furthermore, the transmission shaft can be tested according to the modified score. In the embodiment of the present invention, the data points whose modified scores exceed the preset score threshold can be used as fault data points. When the proportion of fault data points exceeds the preset proportion threshold, it means that the transmission shaft is faulty and needs to be repaired in time. The implementer can determine the detection method of the modified score according to the implementation scenario, which is not limited here. At this point, the abnormal data points and the corresponding abnormal index are obtained through the neighborhood data features of the data points, and the scores obtained by the isolated forest algorithm are corrected according to the abnormal index, so that the modified score can accurately find the fault data points in the vibration intensity time series, thereby improving the fault detection accuracy of the transmission shaft.

[0052] In summary, the embodiment of the present invention provides a data acquisition and processing system for transmission shaft signals of a gearbox based on machine learning; local fluctuation factors and instantaneous change stability factors are obtained according to data features within a preset first neighborhood window and a preset second neighborhood window of data points; clustering is performed according to the local fluctuation factors and instantaneous change stability factors, and a reference cluster is obtained according to the number characteristics of data points within the cluster cluster; abnormal clusters are obtained according to the data difference characteristics between the reference cluster and other cluster clusters; an abnormal index is obtained according to the distance characteristics between the abnormal cluster and the reference cluster, the data discrete characteristics within the abnormal cluster, and the local density characteristics of abnormal data points within the abnormal cluster; and the score of the data point is obtained according to the isolation forest algorithm. The present invention obtains a modified score according to the abnormal index and the score and detects the transmission shaft, thereby improving the detection accuracy.

[0053] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0054] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A gearbox transmission shaft signal data acquisition and processing system based on machine learning, characterized in that: The system includes the following modules: A data acquisition module, used to obtain the vibration intensity time series of the transmission shaft after removing high-frequency noise; A vibration analysis module is used to construct a preset first neighborhood window and a preset second neighborhood window of data points in the vibration intensity time series; obtain a local fluctuation factor according to the data fluctuation characteristics within the preset first neighborhood window and the preset second neighborhood window of the data points; and obtain an instantaneous change stability factor according to the adjacent data change characteristics within the preset first neighborhood window and the preset second neighborhood window of the data points; A feature processing module, used for clustering data points according to the local fluctuation factor and the instantaneous change stability factor to obtain different clusters; and obtaining a reference cluster according to the number characteristics of data points in the clusters; Obtain an abnormal cluster according to the data difference characteristics between the reference cluster and other clustering clusters; obtain a first abnormality degree according to the distance characteristics between the abnormal cluster and the reference cluster and the data discrete characteristics within the abnormal cluster; obtain a second abnormality degree according to the local density characteristics of the abnormal data points within the abnormal cluster; obtain an abnormal index of the abnormal data point according to the first abnormality degree and the second abnormality degree; An anomaly detection module, used for obtaining scores of data points in the vibration intensity time series according to an isolation forest algorithm; and obtaining a modified score according to the anomaly index and the score; Testing the drive shaft according to the modified score; The step of obtaining a reference cluster according to the number characteristics of data points in the cluster comprises: The cluster with the most data points in the cluster is used as the benchmark cluster; The step of obtaining an abnormal cluster according to the data difference characteristics between the benchmark cluster and other clustering clusters includes: The data points in the cluster are sorted in chronological order to obtain the intra-cluster data sequence; the dynamic time-warped distance between the benchmark cluster and the intra-cluster data sequence of other clusters is calculated and normalized to obtain the difference between the benchmark cluster and the other clusters; the other clusters whose difference exceeds a preset difference threshold are regarded as the abnormal clusters.

2. The gearbox transmission shaft signal data acquisition and processing system based on machine learning according to claim 1 is characterized in that: The step of constructing a preset first neighborhood window and a preset second neighborhood window of data points in the vibration intensity time series includes: In the vibration intensity time series, a preset first neighborhood window and a preset second neighborhood window are constructed by taking the data point as the window center, and the length of the preset second neighborhood window is an odd multiple of the length of the preset first neighborhood window.

3. The gearbox transmission shaft signal data acquisition and processing system based on machine learning according to claim 1 is characterized in that: The step of obtaining the local fluctuation factor according to the data fluctuation characteristics within the preset first neighborhood window and the preset second neighborhood window of the data point comprises: Calculate the difference between the maximum and minimum values ​​in the preset first neighborhood window and normalize it to obtain a first extreme value difference; calculate the product of the variance of the data in the preset first neighborhood window and the first extreme value difference to obtain a first fluctuation characteristic value; evenly divide the preset second neighborhood window into sub-windows with the same length as the preset first neighborhood window; calculate the difference between the maximum and minimum values ​​in the sub-window and normalize it to obtain a second extreme value difference; calculate the product of the variance of the data in the sub-window and the second extreme value difference to obtain a second fluctuation characteristic value; calculate the sum of the absolute values ​​of the difference between the first fluctuation characteristic value and the second fluctuation characteristic value of each sub-window to obtain a fluctuation difference value; calculate the product of the first fluctuation characteristic value and the fluctuation difference value and positively correlate them to obtain a local fluctuation factor of the data point.

4. The gearbox transmission shaft signal data acquisition and processing system based on machine learning according to claim 1 is characterized in that: The step of obtaining the instantaneous change stability factor according to the adjacent data change characteristics within the preset first neighborhood window and the preset second neighborhood window of the data point comprises: Calculate the variance of the instantaneous change rate of the data in the preset first neighborhood window to obtain the first instantaneous change characteristic value; calculate the variance of the instantaneous change rate of the data in the preset second neighborhood window to obtain the second instantaneous change characteristic value; calculate the absolute value of the difference between the first instantaneous change characteristic value and the second instantaneous change characteristic value and negatively map them to obtain the instantaneous change stability factor of the data point.

5. The gearbox transmission shaft signal data acquisition and processing system based on machine learning according to claim 1 is characterized in that: The step of obtaining the first abnormality degree according to the distance feature between the abnormal cluster and the reference cluster and the discrete feature of the data in the abnormal cluster comprises: The Euclidean distance between the cluster center of the abnormal cluster and the benchmark cluster is calculated in the cluster space to obtain the difference distance; the coefficient of variation of the local fluctuation factor of the data points in the abnormal cluster and the average value of the coefficient of variation of the instantaneous change stability factor are calculated; the discrete eigenvalue is obtained; the product of the difference distance and the discrete eigenvalue is calculated and normalized to obtain the first abnormality degree of the abnormal cluster.

6. The gearbox transmission shaft signal data acquisition and processing system based on machine learning according to claim 1 is characterized in that: The step of obtaining a second abnormality degree according to the local density characteristics of the abnormal data points in the abnormal cluster comprises: In the clustering space, the average value of the Euclidean distances between the abnormal data point in the abnormal cluster and a preset number of other abnormal data points closest to the abnormal data point is calculated and normalized to obtain a second abnormality degree of the abnormal data point.

7. The gearbox transmission shaft signal data acquisition and processing system based on machine learning according to claim 1 is characterized in that: The step of obtaining the abnormality index of the abnormal data point according to the first abnormality degree and the second abnormality degree comprises: The product of the second abnormality degree of the abnormal data point and the first abnormality degree of the abnormal cluster is calculated to obtain a comprehensive abnormality degree; the sum of the comprehensive abnormality degree and a preset constant is calculated to obtain an abnormality index of the abnormal data point; the preset constant is not less than 1.

8. The gearbox transmission shaft signal data acquisition and processing system based on machine learning according to claim 1 is characterized in that: The step of obtaining a revised score according to the abnormal index and the score comprises: The product of the abnormal index of the abnormal data point and the corresponding score is calculated to obtain a revised score of the abnormal data point; for non-abnormal data points, the score is used as the revised score.

Images