A non-circular gear emulsion motor operating state fault monitoring and early warning method

Abstract

The present application relates to the technical field of non-circular gear emulsion motor operation data processing, and particularly relates to a non-circular gear emulsion motor operation state fault monitoring and early warning method. The method comprises: obtaining a sample data set, setting an initial penalty factor and classifying samples in the sample data set to obtain outliers; calculating the importance of vibration data in the outliers when the outliers are positive samples; calculating the importance of vibration data in the outliers when the outliers are negative samples; calculating the importance of current data in the outliers when the outliers are positive samples; calculating the importance of current data in the outliers when the outliers are negative samples; further obtaining the importance of the outliers; obtaining an optimized penalty factor according to the importance of the outliers and the initial penalty factor; obtaining an optimized SVM model based on the optimized penalty factor of each outlier, and monitoring the operation state of the motor. The present application can improve the accuracy of non-circular gear emulsion motor operation fault monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of non-circular gear emulsion motor operation data processing, and particularly relates to a non-circular gear emulsion motor operation state fault monitoring and early warning method. BACKGROUND

[0002] The core component of the non-circular gear emulsion motor is a non-circular gear train, which is usually composed of a non-circular plum blossom inner ring gear, a non-circular square sun gear and a planetary gear. Its working principle mainly utilizes the mechanical properties of liquid to drive the equipment to run through the change of the pressure difference and the flow direction of the liquid. Specifically, when the emulsion enters the motor, the flow direction and speed of the liquid change after the action of the rotor and the stator, thereby generating a pressure difference. This pressure difference drives the rotation of the rotor, and in turn drives the equipment to run.

[0003] The non-circular gear emulsion motor is widely used in flammable and explosive environments such as coal mines, because it uses difficult-to-burn liquid-high water-based emulsion as a working medium, and once a fault occurs, a series of safety problems will be caused. And as a key equipment, the running state of the motor directly affects the production efficiency, so it is necessary to detect and warn the running state of the motor to improve the safety and reliability of the equipment.

[0004] Traditionally, the data generated when the non-circular gear emulsion motor is running is monitored to determine whether the data exceeds the threshold for early warning. However, due to differences in working environment and fault types, the method of determining whether the threshold is exceeded is not accurate enough for early warning of running data with complex linear relationships. Because support vector machines (SVM) can handle complex nonlinear problems and are good at handling multiple sources of data collected during motor fault monitoring (such as vibration data and current data), support vector machines (SVM) are used for fault monitoring and early warning. However, the support vector machine is affected by the size of the penalty factor when identifying abnormalities. If the penalty factor is too large, it will cause overfitting, making the model less generalizable. If the penalty factor is too small, it may ignore some important classification information during model training. Both cases will affect the abnormal classification results of the current vibration data. SUMMARY

[0005] To solve the above technical problems, the purpose of the present application is to provide a non-circular gear emulsion motor operation state fault monitoring and early warning method, and the technical solution adopted is as follows:

[0006] One embodiment of the present application provides a non-circular gear emulsion motor operation state fault monitoring and early warning method, which comprises:

[0007] The positive samples and the negative samples are obtained according to current data and vibration data when the motor is running, and a sample data set is formed; current sequences and vibration sequences in which the current data and the vibration data are located are determined; an initial penalty factor is set, and outliers are obtained by classifying the samples in the sample data set;

[0008] If the outlier is a positive sample, the importance of the vibration data in the outlier is calculated according to an abnormal point in a vibration sequence in which the vibration data is located, a coefficient of variation and data before and after the vibration data; if the outlier is a negative sample, the importance of the vibration data in the outlier is calculated according to an abnormal point in a vibration sequence in which the vibration data is located and a spectral entropy.

[0009] If the outlier is a positive sample, the importance of the current data in the outlier is calculated according to an information entropy of a current sequence in which the current data is located and data before and after the current data; if the outlier is a negative sample, the importance of the current data in the outlier is calculated according to an abnormal point in a current sequence in which the current data is located and data before and after the current data.

[0010] The importance of the outlier is calculated according to the importance of the vibration data and the current data in the outlier; an optimized penalty factor of the outlier is obtained according to the importance of the outlier and the initial penalty factor; an optimized SVM model is obtained based on the optimized penalty factors of the outliers, and the running state of the motor is monitored.

[0011] Preferably, the positive samples and the negative samples are obtained according to current data and vibration data when the motor is running, and a sample data set is formed, comprising:

[0012] The current data and the vibration data are preprocessed, including data cleaning and standardization processing; the current data and the vibration data at one time point after preprocessing form a feature vector, one feature vector is one sample, the samples in a normal running state of the motor are positive samples, and the samples in an abnormal running state of the motor are negative samples; the positive samples and the negative samples are divided into a training set and a test set, and the test set and the training set form a sample data set.

[0013] Preferably, the initial penalty factor is set, and the outliers are obtained by classifying the samples in the sample data set, comprising:

[0014] The initial SVM model is obtained by training the sample data set according to the set initial penalty factor, and the outliers are obtained by classifying the sample data set by using the initial SVM model.

[0015] Preferably, the importance of the vibration data in the outlier is calculated according to an abnormal point in a vibration sequence in which the vibration data is located, a coefficient of variation and data before and after the vibration data, comprising:

[0016] A normalized value of a ratio of a number of abnormal points in the vibration sequence to which the vibration data in the outlier point belongs to a number of abnormal points in all vibration sequences after box plot screening is obtained, and is denoted as a non-abnormal proportion; a normalized value of a ratio of a coefficient of variation of the vibration sequence to which the vibration data in the outlier point belongs to a sum of coefficients of variation of all vibration sequences is obtained, and is denoted as a stability degree; a difference between the vibration data in the outlier point and a mean of data before the vibration data in the vibration sequence to which the vibration data belongs is obtained, and an absolute value is taken to obtain a first difference; a difference between the vibration data in the outlier point and a mean of data after the vibration data in the vibration sequence to which the vibration data belongs is obtained, and an absolute value is taken to obtain a second difference; an absolute value of a difference between the first difference and the second difference is obtained, and is normalized to obtain a turning point possibility; and the non-abnormal proportion, the stability degree and the turning point possibility are weighted and summed to obtain the importance of the vibration data.

[0017] Preferably, the importance of the vibration data in the outlier point is calculated according to the abnormal points in the vibration sequence to which the vibration data belongs and the spectral entropy, and includes:

[0018] A normalized value of a ratio of a number of abnormal points in the vibration sequence to which the vibration data in the outlier point belongs to a number of abnormal points in all vibration sequences after box plot screening is obtained, and is denoted as an abnormal proportion; a ratio of a spectral entropy of the vibration sequence to which the vibration data in the outlier point belongs to a sum of spectral entropies of all vibration sequences is obtained, and is normalized to obtain an abnormal degree; and the abnormal proportion and the abnormal degree are weighted and summed to obtain the importance of the vibration data.

[0019] Preferably, the importance of the current data in the outlier point is calculated according to the information entropy of the current sequence to which the current data belongs and data before and after the current data, and includes:

[0020] A current sequence stability degree is obtained by comparing the information entropy of the current sequence to which the current data in the outlier point belongs with a sum of information entropies of all current sequences and normalizing; a first slope and a second slope are calculated respectively, the first slope being a slope between the current data and a previous adjacent current data, and the second slope being a slope between the current data and a next adjacent current data; an absolute value of an inverse of a difference between the first slope and the second slope is normalized to obtain a slope change condition; and the current sequence stability degree and the slope change condition are weighted and summed to obtain the importance of the current data.

[0021] Preferably, the importance of the current data in the outlier point is calculated according to the abnormal points in the current sequence to which the current data belongs and data before and after the current data, and includes:

[0022] A normalized value of a ratio of a number of abnormal points in a current sequence where current data in an outlier is screened by a box plot to a number of abnormal points in all current sequences screened by a box plot is recorded as a current abnormality proportion; an absolute value of a difference between the first slope and the second slope is normalized to obtain a slope difference; and the current abnormality proportion and the slope difference are weighted and summed to obtain the importance of the current data.

[0023] Preferably, the importance of the outlier is calculated according to the importance of the vibration data and the current data in the outlier, and includes:

[0024] For the discrete point being the negative sample, the importance of the vibration data and the current data in the discrete point obtained under the negative sample is weighted and summed to obtain the importance of the discrete point; and for the discrete point being the positive sample, the importance of the vibration data and the current data in the discrete point obtained under the positive sample is weighted and summed to obtain the importance of the discrete point.

[0025] Preferably, the optimized penalty factor of the outlier is obtained according to the importance of the outlier and the initial penalty factor, and includes:

[0026] The importance of the discrete point is added to the initial penalty factor to obtain the optimized penalty factor of the discrete point.

[0027] Preferably, the optimized SVM model is obtained based on the optimized penalty factor of each outlier, and the running state of the motor is monitored, and includes:

[0028] The initial SVM model is trained by using each discrete point in the sample data set, the corresponding optimized penalty factor and other samples to obtain the optimized SVM model; and the running data of the non-circular gear emulsion motor is classified by using the optimized SVM model to monitor the running state of the motor.

[0029] The application has at least the following beneficial effects: the initial penalty factor is set, the initial SVM model is combined with the initial penalty factor to classify the samples in the sample data set to obtain outliers; further, the importance of the current data and the vibration data in the outlier is calculated on the premise that the outlier is the positive sample or the negative sample, the importance of the current data and the vibration data is integrated to obtain the importance of the discrete point, the initial penalty factor is optimized by using the importance of each outlier to obtain the optimized penalty factor of the discrete point, and then the optimized penalty factor is brought into the original optimization problem, the sample data set is trained to obtain a more accurate optimized SVM model, and the accuracy of the fault monitoring of the running state of the non-circular gear emulsion motor is improved. BRIEF DESCRIPTION OF DRAWINGS

[0030] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0031] Figure 1 A method flow chart of a non-circular gear emulsion motor running state fault monitoring and early warning method provided by the embodiment of the present application;

[0032] Figure 2 A non-circular gear emulsion motor running state fault monitoring and early warning method provided by the embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the specific implementation, structure, features and effects of the non-circular gear emulsion motor running state fault monitoring and early warning method according to the present application are described in detail as follows. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0034] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0035] The specific scheme of the non-circular gear emulsion motor running state fault monitoring and early warning method provided by the present application is specifically described below with reference to the drawings.

[0036] Embodiment:

[0037] The main application scenario of the present application is: collecting running data of the non-circular gear emulsion motor during running, including current data and vibration data, then analyzing the running data to obtain the abnormal running state, and then realizing the purpose of non-circular gear emulsion motor running fault monitoring and early warning.

[0038] Please refer to Figure 1 , which shows a method flow chart of a non-circular gear emulsion motor running state fault monitoring and early warning method provided by the embodiment of the present application, which comprises the following steps:

[0039] Step S1, according to the current data and vibration data of the motor running, positive samples and negative samples are obtained, and a sample data set is formed; the current sequence and the vibration sequence of each current data and vibration data are determined; an initial penalty factor is set, and the samples in the sample data set are classified to obtain outliers.

[0040] During the operation of the non-circular gear emulsion motor, the current data can reflect the load condition of the motor, and too high or too low current may mean that the motor has a problem, and the change of the vibration data can provide an early warning in the initial stage of failure, and provide more specific fault information, so the current data and vibration data of the non-circular gear emulsion motor during operation are collected. Since the operation of the non-circular gear emulsion motor is not a fixed period, the current data and vibration data are collected during the operation period, the current data of one operation period forms a current sequence, and the voltage data of one operation period forms a voltage sequence. The current data and vibration data are obtained according to the operation period of the current data and vibration data.

[0041] Further, the current data and vibration data need to be preprocessed, mainly including data cleaning and standardization processing. Since the current data generated during the operation of the motor needs to be classified as an anomaly using SVM, the historical emulsion motor operation data needs to be cleaned first, including outlier rejection, missing value completion, and denoising. The current data and vibration data after data cleaning need to be standardized to ensure that each feature has the same dimension. At the same time, since the subsequent feature space vectors need to be reflected, the corresponding timestamps of the vibration data and current data before mapping to the feature space are needed.

[0042] Further, the current data and vibration data of one time after preprocessing form a feature vector, and one feature vector is a sample. The sample in the normal operation state of the motor is a positive sample, and the sample in the abnormal operation state of the motor is a negative sample. It should be noted that when dividing the positive and negative samples, manual division is needed to divide them into positive and negative samples. The positive samples and negative samples are divided into a training set and a test set, the test set and the training set form a sample data set, the method of dividing into a training set and a test set is a cross-validation method, the test set is used to test the performance of the model, and the commonly used cross-validation methods include K-fold cross-validation and leave-one-out cross-validation.

[0043] Then, a suitable kernel function needs to be selected to process the nonlinear relationship of the positive and negative samples of the sample data set, so that the data becomes more separable, such as a radial basis kernel function (RBF), an S-shaped kernel function or a polynomial kernel function to reduce the computational complexity.

[0044] In an ideal case, using SVM classification will divide the motor running training data into normal data and fault data, but due to the nonlinear and dynamic characteristics and noise of the non-circular gear emulsion motor during operation, the initial classification model will have outliers, and since the penalty factor is the tolerance of outliers, the larger the penalty factor, the less willing to discard outliers, so it is necessary to evaluate the importance of outliers, so that each outlier corresponds to a corresponding penalty factor, and then the corresponding penalty factor of each outlier is brought into the model to train the training data set, and a more accurate classification model is obtained, so as to realize the fault evaluation of the current motor running state.

[0045] Therefore, it is necessary to set an initial penalty factor and classify the samples in the sample data set to obtain outliers. Specifically, an initial SVM model is obtained by training the sample data set according to the set initial penalty factor, and the outliers are obtained by classifying the sample data set using the initial SVM model, as shown in Figure 2 H is a hyperplane, and are support vectors, and the distance from the hyperplane is equal. The square represents the positive sample of the motor running data in the feature space, that is, the normal state of the motor running; the circle represents the negative sample of the motor running data in the feature space, that is, the fault state of the motor running; the square and the circle on the two straight lines are closest to the hyperplane, indicating the support vector; the square in the circular region is an outlier, which is recorded and corresponds to the initial vibration data and current data. It should be noted that a hyperplane is obtained after model training, and then the data is classified.

[0046] Step S2, if the outlier is a positive sample, the importance of the vibration data in the outlier is calculated according to the abnormal points in the vibration sequence where the vibration data is located, the coefficient of variation and the data before and after the vibration data; if the outlier is a negative sample, the importance of the vibration data in the outlier is calculated according to the abnormal points in the vibration sequence where the vibration data is located and the spectral entropy.

[0047] Since the penalty factor represents the tolerance of the classification error of the outliers, whether to pay attention to the classification results of these outliers needs to be judged according to the characteristics of the outliers, that is, the importance of the outliers needs to be calculated at this time, and the higher the importance, the more attention is paid to the classification results of the outliers, and the corresponding penalty factor should be larger at this time. Since the outlier is composed of two features, vibration data and current data, the corresponding importance needs to be obtained according to the data characteristics of the outliers in the two types of data, and then the comprehensive importance of the current outlier is obtained.

[0048] Since the importance of the outliers of different positive and negative categories is different, the importance needs to be calculated according to the positive and negative sample categories of the outliers.

[0049] If an outlier is a positive sample, the importance of the vibration data is calculated based on the outlier in the vibration sequence containing the outlier, the coefficient of variation, and the data before and after the outlier. Let the vibration sequence contain a total of... The vibration data in the i-th outlier point is at the i-th... In a vibration sequence.

[0050] Specifically, the normalized value of the ratio of the number of outliers in all vibration sequences after box plot screening to the number of outliers in the vibration sequence containing the outlier data after box plot screening is obtained, denoted as the non-outlier percentage; the normalized value of the ratio of the sum of the coefficients of variation of all vibration sequences to the coefficient of variation of the vibration sequence containing the outlier data is obtained, denoted as the stability; the difference between the vibration data in the outlier and the mean of the data preceding the outlier in the vibration sequence containing the outlier data is obtained, and the absolute value is calculated to obtain the first difference; the difference between the vibration data in the outlier and the mean of the data following the outlier in the vibration sequence containing the outlier data is obtained, and the absolute value is calculated to obtain the second difference; the absolute value of the difference between the first difference and the second difference is obtained and normalized to obtain the probability of a turning point; the weighted sum of the non-outlier percentage, stability, and probability of a turning point is calculated to obtain the importance of the vibration data.

[0051] The specific calculation formula is as follows:

[0052]

[0053] in, This indicates the importance of the i-th outlier as a positive sample in the corresponding vibration sequence, which is also the importance of the vibration data in the i-th outlier. This represents the number of outliers in the vibration sequence containing the vibration data of the i-th outlier when the i-th outlier is a positive sample, after being filtered by box plot. The number of outliers after screening all vibration sequences using box plots; Let be the coefficient of variation of the j-th vibration sequence; This indicates the location of the vibration data in the i-th outlier point. The value of the coefficient of variation in each vibration sequence segment; Let i be the value of the vibration data at the i-th outlier point. This represents the mean of the vibration data preceding the vibration data at the i-th outlier in the vibration sequence containing the vibration data at the i-th outlier. Let be the mean of the vibration data following the vibration data of the i-th outlier in the vibration sequence containing the vibration data of the i-th outlier, where norm represents the normalization operation.

[0054] the greater the value, the fewer the number of abnormal points in the vibration sequence where the i-th outlier point is located, indicating that the stability of the vibration sequence where the i-th outlier point is located is higher, and the greater importance should be given to this point; the greater the value, the fewer the number of abnormal points in the vibration sequence where the i-th outlier point is located, indicating that the stability of the vibration sequence where the i-th outlier point is located is higher, and the greater importance should be given to this point; the greater the value, the fewer the number of abnormal points in the vibration sequence where the i-th outlier point is located, indicating that the stability of the vibration sequence where the i-th outlier point is located is higher, and the greater importance should be given to this point; the greater the value, the lower the stability of the vibration sequence segment where the i-th outlier point is located, that is, the higher the stability of the vibration sequence segment where the vibration data in the i-th outlier point is located, and the greater importance should be given to this point, the greater the value, the lower the stability of the vibration sequence segment where the i-th outlier point is located, that is, the higher the stability of the vibration sequence segment where the vibration data in the i-th outlier point is located, and the greater importance should be given to this point, the greater the value, the lower the stability of the vibration sequence segment where the i-th outlier point is located, that is, the higher the stability of the vibration sequence segment where the vibration data in the i-th outlier point is located, and the greater importance should be given to this point, the greater the value, the lower the stability of the vibration sequence segment where the i-th outlier point is located, that is, the higher the stability of the vibration sequence segment where the vibration data in the i-th outlier point is located, and the greater importance should be given to this point, the greater the value, the lower the stability of the vibration sequence segment where the i-th outlier point is located, that is, the higher the stability of the vibration sequence segment where the vibration data in the i-th outlier point is located, and the greater importance should be given to this point, the greater the value, the lower the stability of the vibration sequence segment where the i-th outlier point is located, that is, the higher the stability of the vibration sequence segment where the vibration data in the i-th outlier point is located, and the greater importance should be given to this point,

[0055] Further, when the outlier point is a negative sample, the importance of the vibration data in the outlier point is calculated according to the abnormal points in the vibration sequence where the vibration data is located and the spectral entropy.

[0056] Specifically, a normalized value of a ratio of a number of abnormal points in a vibration sequence where vibration data in an outlier point is located to a number of abnormal points in all vibration sequences after screening by a box plot is obtained, denoted as an abnormality ratio; a ratio of a spectral entropy of the vibration sequence where the vibration data in the outlier point is located to a sum of spectral entropies of all vibration sequences is obtained, and an abnormality degree is obtained by normalization; and the importance of the vibration data is obtained by weighted summation of the abnormality ratio and the abnormality degree.

[0057] The calculation formula is:

[0058]

[0059] wherein, ​​The importance of the i-th outlier as a negative sample in the corresponding vibration data sequence is, in other words, the importance of the vibration data in the i-th outlier as a negative sample. Indicates the location of the i-th outlier. The ratio of the number of outliers in a vibration sequence after box plot screening to the total number of outliers in all vibration sequences. This represents the proportion of outliers. The larger the value, the more outliers there are in the vibration sequence segment where the i-th outlier is located. This indicates that the degree of anomalousness of the vibration sequence where the i-th outlier is located as a negative sample is higher, and this point should be given greater importance. The vibration data in the i-th outlier point is located at the i-th outlier point. The spectral entropy of a vibration sequence Let be the spectral entropy of the j-th vibration sequence. It is the sum of the spectral entropies of all vibration sequences. This value represents the degree of anomalousness. A larger value indicates a greater degree of anomalousness in the vibration sequence containing the i-th outlier, and thus, that point should be given a higher importance. Since the first term in the above formula has a greater impact on the importance calculation, it is weighted accordingly. , .

[0060] This allows us to calculate the importance of vibration data in outliers, whether they are positive or negative samples.

[0061] Step S3: If the outlier is a positive sample, calculate the importance of the current data based on the information entropy of the current sequence in which the current data is located and the data before and after the current data; if the outlier is a negative sample, calculate the importance of the current data based on the outliers in the current sequence in which the current data is located and the data before and after the current data.

[0062] After calculating the importance of vibration data at outliers, it is also necessary to analyze the current data at outliers, assuming the current sequence has a total of The current data of the i-th outlier is located at the i-th... There are several current sequences. Similarly, outliers can be positive or negative samples, and the methods for calculating the importance of their current data differ.

[0063] If the outlier is a positive sample, the importance of the current data is calculated based on the information entropy of the current sequence in which the current data in the outlier is located and the data before and after the current data.

[0064] Specifically, the sum of information entropy of all current sequences is compared with the information entropy of the current sequence in which the current data in the outlier point is located, and normalization is performed to obtain the stability degree of the current sequence; the slope between the current data and the previous adjacent current data and the slope between the current data and the next adjacent current data are calculated respectively, and are denoted as the first slope and the second slope; the absolute value of the difference between the first slope and the second slope is inverted and normalized to obtain the slope change condition; the stability degree of the current sequence and the slope change condition are weighted and summed to obtain the importance degree of the current data.

[0065] The calculation formula is:

[0066]

[0067] denotes the importance degree of the current data in the i th outlier point when the i th outlier point is a positive sample; denotes the information entropy of the j th current sequence; denotes the information entropy of the current sequence in which the current data in the i th outlier point is located; denotes the sum of information entropy of all current sequences, denotes the stability degree of the current sequence, and the result of normalization of the ratio between the sum of information entropy of all current sequences and the information entropy of the current sequence in which the current data in the outlier point is located, the greater the value, the higher the stability degree of the current sequence in which the current data in the outlier point is located, and the greater importance degree should be given to the outlier point as a positive sample; denotes the slope between the current data and the previous adjacent current data in the i th outlier point, that is, the first slope, denotes the slope between the current data and the next adjacent current data in the i th outlier point, that is, the second slope, denotes the slope change condition, and the result of normalization of the inverse of the absolute value of the difference between the two slopes, the greater the value, that is, the smaller the denominator in the second term, indicating that the difference between the slope of the current data and the slope of the adjacent current data is smaller when the i th outlier point is a positive sample, indicating that the stability degree of the point is greater, and the greater importance degree should be given to the point, is the weight.

[0068] Finally, the importance degree of the current data in the outlier point when the outlier point is a negative sample is calculated, and the importance degree of the current data is calculated according to the abnormal point in the current sequence in which the current data is located and the data before and after the current data.

[0069] ​​Specifically, a normalized value of a ratio of a number of abnormal points in a current sequence in which current data in an outlier point is located after being screened by a box plot to a number of abnormal points in all current sequences after being screened by a box plot is obtained, and is denoted as a current abnormality proportion; an absolute value of a difference between the first slope and the second slope is normalized to obtain a slope difference; and the current abnormality proportion and the slope difference are weighted and summed to obtain the importance of the current data.

[0070] The calculation formula is:

[0071] ,

[0072] Among them, represents the importance of the current data in the i-th discrete point when the i-th discrete point is a negative sample; represents the number of abnormal points in a current sequence in which current data in the i-th outlier point is located after being screened by a box plot when the i-th outlier point is a negative sample, is the number of abnormal points in all current sequences after being screened by a box plot; represents the slope of the current data in the i-th outlier point and the previous adjacent current data, that is, the first slope, represents the slope of the current data in the i-th outlier point and the next adjacent current data, that is, the second slope.

[0073] is the current abnormality proportion, and the greater the value, the more abnormal points there are in the current sequence in which the current data in the i-th outlier point is located, indicating that the i-th outlier point as a negative sample has a higher degree of abnormality in the current sequence, and the i-th outlier point should be given a greater importance; represents the slope difference, and represents the slope difference between the current data in the i-th outlier point and the previous and next two current data. The greater the value, the higher the degree of abnormality of the current point, and the greater importance should be given to the current outlier point.

[0074] Thus, the importance of the current data in the discrete point when the discrete point is a positive sample or a negative sample is obtained.

[0075] Step S4, calculating the importance of the outlier point according to the importance of the vibration data and the current data in the outlier point; obtaining an optimized penalty factor of the outlier point according to the importance of the outlier point and the initial penalty factor; obtaining an optimized SVM model based on the optimized penalty factors of the outlier points, and monitoring the running state of the non-circular gear emulsion motor.

[0076] The importance of the vibration data and the current data in the discrete points is obtained in steps S2 and S3 respectively when the discrete points are positive samples or negative samples, and the importance of the vibration data and the current data of the discrete points needs to be further integrated to calculate the importance of the discrete points. When calculating, the discrete points also need to be divided into positive samples or negative samples for calculation.

[0077] For the discrete points that are negative samples, the importance of the vibration data and the current data in the discrete points obtained under negative samples is weighted and summed to obtain the importance of the discrete points. For the discrete points that are positive samples, the importance of the vibration data and the current data in the discrete points obtained under positive samples is weighted and summed to obtain the importance of the discrete points.

[0078] The specific calculation formula is:

[0079]

[0080]

[0081] Among them, and are the importance of the discrete points when they are positive samples and the importance of the discrete points when they are negative samples respectively; and are the importance of the vibration data and the current data of the discrete points when they are positive samples; and are the importance of the vibration data and the current data of the discrete points when they are negative samples.

[0082] Thus, the importance of each discrete point can be obtained.

[0083] Since the penalty factor represents the tolerance of the classification error of the outlier, the higher the importance of the outlier, the more attention is paid to the classification result of the point, and the more hope is given to the correct classification of the point. Therefore, a larger penalty factor should be used for the outlier with high importance to ensure that the point is classified as correctly as possible.

[0084] The optimized penalty factor of the outlier is obtained according to the importance of the outlier and the initial penalty factor. Specifically, the importance of the discrete point is added to the initial penalty factor to obtain the optimized penalty factor of the discrete point. For the discrete points that are positive samples, the importance of the discrete points calculated under positive samples is added to the initial penalty factor to obtain the optimized penalty factor. For the discrete points that are negative samples, the importance of the discrete points calculated under negative samples is added to the initial penalty factor to obtain the optimized penalty factor.

[0085] Before obtaining the optimized penalty factor, the importance of the discrete point needs to be processed for subsequent use. The importance of the discrete point is kept to one decimal place, and then added to the initial penalty factor.

[0086] ​​After the optimization penalty factor is obtained, the optimization penalty factor is brought into the original optimization problem, the running data is trained using the training model, and a more accurate classification model is obtained. Specifically, the initial SVM model is trained using each discrete point in the sample data set, the corresponding optimization penalty factor and other samples, to obtain an optimized SVM model. The running data of the non-circular gear emulsion motor is classified using the optimized SVM model to monitor the running state of the motor. If the running data is found to be abnormal, a warning is given to remind the staff to repair.

[0087] The present scheme first classifies the motor running training data, obtains outliers according to the initial classification result, calculates the importance of the outliers in the corresponding data sequence, adjusts the penalty factor according to the importance, and then trains the running data using the adjusted training model to obtain a more accurate classification model, thereby achieving more accurate abnormal identification of the current motor running data. This method of adaptively adjusting the penalty factor according to the importance of the outliers is more flexible than the traditional SVM classification, and is more suitable for the unique environment of the non-circular gear emulsion motor, thereby improving the accuracy of fault monitoring.

[0088] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. The above describes a specific embodiment of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0089] Each embodiment in the present specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment focuses on the difference from other embodiments.

[0090] The above only describes the preferred embodiments of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A non-circular gear emulsion motor operating state fault monitoring and early warning method, characterized in that, The method comprises: According to the current data and vibration data of the motor in operation, positive samples and negative samples are obtained, and a sample data set is formed; current sequences and vibration sequences in which the current data and the vibration data are located are determined; an initial penalty factor is set, and the samples in the sample data set are classified to obtain outliers; If the outliers are positive samples, the importance of the vibration data is calculated according to the abnormal points in the vibration sequence in which the vibration data is located, the coefficient of variation and the data before and after the vibration data; if the outliers are negative samples, the importance of the vibration data is calculated according to the abnormal points in the vibration sequence in which the vibration data is located and the spectral entropy; If the outliers are positive samples, the importance of the current data is calculated according to the information entropy of the current sequence in which the current data is located and the data before and after the current data; if the outliers are negative samples, the importance of the current data is calculated according to the abnormal points in the current sequence in which the current data is located and the data before and after the current data; The importance of the outliers is calculated according to the importance of the vibration data and the current data in one outlier; the optimized penalty factor of the outlier is obtained according to the importance of the outlier and the initial penalty factor; the optimized SVM model is obtained based on the optimized penalty factors of the outliers, and the operating state of the motor is monitored.

2. The method according to claim 1, characterized in that, According to the current data and vibration data of the motor in operation, positive samples and negative samples are obtained, and a sample data set is formed, comprising: The current data and the vibration data are preprocessed, including data cleaning and standardization processing; the current data and the vibration data at one time after preprocessing form a feature vector, one feature vector is one sample, the samples under the normal operating state of the motor are positive samples, and the samples under the abnormal operating state of the motor are negative samples; the positive samples and the negative samples are divided into a training set and a test set, and the test set and the training set form a sample data set.

3. The method according to claim 1, characterized in that, The initial penalty factor is set, and the samples in the sample data set are classified to obtain outliers, comprising: According to the set initial penalty factor, the sample data set is trained to obtain an initial SVM model, and the initial SVM model is used for classifying the sample data set to obtain outliers.

4. The method according to claim 1, wherein The importance of the vibration data is calculated according to the abnormal points in the vibration sequence in which the vibration data is located, the coefficient of variation and the data before and after the vibration data, comprising: A normalized value of a ratio of a number of abnormal points in the vibration sequence to which the vibration data in the outlier point belongs to a number of abnormal points in all vibration sequences is obtained, and is denoted as a non-abnormal proportion; a normalized value of a ratio of a coefficient of variation of the vibration sequence to which the vibration data in the outlier point belongs to a sum of coefficients of variation of all vibration sequences is obtained, and is denoted as a stability degree; a difference between the vibration data in the outlier point and an average of data before the vibration data in the vibration sequence to which the vibration data belongs is obtained, and an absolute value of the difference is obtained, to obtain a first difference; a difference between the vibration data in the outlier point and an average of data after the vibration data in the vibration sequence to which the vibration data belongs is obtained, and an absolute value of the difference is obtained, to obtain a second difference; an absolute value of a difference between the first difference and the second difference is obtained, and is normalized to obtain a turning point possibility; and the non-abnormal proportion, the stability degree and the turning point possibility are weighted and summed to obtain an importance of the vibration data.

5. The method according to claim 1, wherein The importance of the vibration data is calculated according to the abnormal points in the vibration sequence to which the vibration data in the outlier point belongs and the spectral entropy, and includes: A normalized value of a ratio of a number of abnormal points in the vibration sequence to which the vibration data in the outlier point belongs to a number of abnormal points in all vibration sequences is obtained, and is denoted as an abnormal proportion; an abnormal degree is obtained by normalizing a ratio of a spectral entropy of the vibration sequence to which the vibration data in the outlier point belongs to a sum of spectral entropies of all vibration sequences; and the abnormal proportion and the abnormal degree are weighted and summed to obtain the importance of the vibration data.

6. The method according to claim 1, wherein The importance of the current data is calculated according to the information entropy of the current sequence to which the current data in the outlier point belongs and the data before and after the current data, and includes: A current sequence stability degree is obtained by normalizing a ratio of a sum of information entropies of all current sequences to the information entropy of the current sequence to which the current data in the outlier point belongs; a first slope and a second slope are calculated respectively, the first slope being a slope between the current data and a previous adjacent current data, and the second slope being a slope between the current data and a next adjacent current data; an absolute value of a difference between the first slope and the second slope is inverted and normalized to obtain a slope change condition; and the current sequence stability degree and the slope change condition are weighted and summed to obtain the importance of the current data.

7. The method according to claim 1, wherein The importance of the current data is calculated according to the abnormal points in the current sequence to which the current data in the outlier point belongs and the data before and after the current data, and includes: A normalized value of a ratio of a number of abnormal points in the current sequence to which the current data in the outlier point belongs to a number of abnormal points in all current sequences is obtained, and is denoted as a current abnormal proportion; an absolute value of a difference between the first slope and the second slope is normalized to obtain a slope difference; and the current abnormal proportion and the slope difference are weighted and summed to obtain the importance of the current data.

8. The method according to claim 1, wherein, The importance of the outlier point is calculated according to the importance of the vibration data and the importance of the current data in the outlier point, and includes: For the discrete points being negative samples, the importance of the vibration data and the current data in the discrete points under the negative samples is weighted and summed to obtain the importance of the discrete points; for the discrete points being positive samples, the importance of the vibration data and the current data in the discrete points under the positive samples is weighted and summed to obtain the importance of the discrete points.

9. The method according to claim 1, wherein The obtaining of the optimized penalty factor of the outlier point according to the importance of the outlier point and the initial penalty factor comprises: The importance of the discrete point is added to the initial penalty factor to obtain the optimized penalty factor of the discrete point.

10. The method according to claim 1, wherein The obtaining of the optimized SVM model based on the optimized penalty factor of each outlier point and the monitoring of the running state of the motor comprise: The initial SVM model is trained by using each discrete point in the sample data set, the corresponding optimized penalty factor and other samples, to obtain the optimized SVM model; the running data of the non-circular gear emulsion motor is classified by using the optimized SVM model, to monitor the running state of the motor.

Images