Online monitoring system and method for machining state of grinding equipment

Abstract

The invention relates to the field of online monitoring, and particularly discloses an online monitoring system and method for the machining state of grinding equipment. The method comprises the steps that firstly, vibration values, collected by a vibration sensor, of grinding equipment at multiple preset time points in a preset time period and temperature values, collected by a temperature sensor, of the grinding equipment at the multiple preset time points are obtained; after Fourier transform is carried out on the vibration values, a vibration frequency domain feature vector is obtained through a sequence encoder of a joint encoder, and after the temperature values are arranged into a temperature input vector, a temperature time sequence correlation feature vector is obtained through a multi-scale neighborhood feature extraction module; the vibration frequency domain feature vector and the temperature time sequence correlation feature vector are fused to obtain a classification feature matrix, finally, the classification feature matrix passes through a classifier to obtain a classification result, and the classification result represents whether the machining state of the grinding equipment is abnormal or not, so that the machining efficiency and precision of the grinding equipment are further improved.

Description

Technical Field

[0001] The present application relates to the field of online monitoring, and more specifically, to an online monitoring system and method for the processing status of a grinding device. Background Art

[0002] The processing status of the grinding equipment directly affects the quality and efficiency of the processing process. If there are problems with the grinding equipment, problems such as increased surface roughness, dimensional deviation or uneven grinding will occur. It will also affect the processing efficiency, equipment life and maintenance costs.

[0003] In the current existing technology, manual monitoring of the processing status of grinding equipment is easily affected by the subjective consciousness and experience of the operator. Different operators may have different judgments on the same processing status, and long-term manual monitoring is likely to cause operator fatigue and inattention, which may miss some subtle changes or abnormal situations. In addition, manual monitoring may be intermittent, unable to obtain information on the processing status in real time, and difficult to achieve continuous monitoring.

[0004] Therefore, a more optimized online monitoring system and method for the processing status of grinding equipment is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an online monitoring system and method for the processing state of a grinding device, which first obtains the vibration values ​​of the grinding device at multiple predetermined time points within a predetermined time period collected by a vibration sensor and the temperature values ​​of the grinding device at multiple predetermined time points collected by a temperature sensor, and then the vibration values ​​are subjected to Fourier transformation and passed through a sequence encoder of a joint encoder to obtain a vibration frequency domain feature vector, and then the temperature values ​​are arranged as a temperature input vector and then obtained through a multi-scale neighborhood feature extraction module to obtain a temperature time series associated feature vector, and further, the vibration frequency domain feature vector and the temperature time series associated feature vector are fused to obtain a classification feature matrix, and finally, the classification feature matrix is ​​passed through a classifier to obtain a classification result, and the classification result indicates whether the processing state of the grinding equipment is abnormal, so as to further improve the efficiency and accuracy of the grinding equipment processing.

[0006] According to one aspect of the present application, there is provided an online monitoring system for a grinding equipment processing state, comprising:

[0007] A data acquisition module, used to acquire the vibration values ​​of the grinding device collected by the vibration sensor at a plurality of predetermined time points within a predetermined time period and the temperature values ​​of the grinding device collected by the temperature sensor at a plurality of predetermined time points;

[0008] A frequency domain feature extraction module, used for converting the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points within a predetermined time period into the frequency domain based on Fourier transformation to obtain multiple vibration frequency domain statistical features;

[0009] A frequency domain feature encoding module, used for passing the plurality of vibration frequency domain statistical features through a sequence encoder of a joint encoder to obtain a vibration frequency domain feature vector, wherein the sequence encoder is a context encoder including an embedding layer;

[0010] A temperature feature encoding module, used for arranging the temperature values ​​of the grinding equipment collected by the temperature sensor at the plurality of predetermined time points into a temperature input vector and then passing it through a multi-scale neighborhood feature extraction module to obtain a temperature time series correlation feature vector;

[0011] A fusion module, used for fusing the vibration frequency domain feature vector and the temperature time series correlation feature vector to obtain a classification feature matrix;

[0012] The classification module is used to obtain a classification result by passing the classification feature matrix through a classifier, and the classification result indicates whether the processing state of the grinding equipment is abnormal.

[0013] In the above-mentioned online monitoring system of the processing status of the grinding equipment, the frequency domain feature encoding module includes: a vector embedding unit, which is used to use the embedding layer of the sequence encoder to convert the multiple vibration frequency domain statistical features into input vectors respectively to obtain a sequence of multiple vibration frequency domain embedded vectors; a context encoding unit, which is used to use the converter-based Bert model of the sequence encoder to perform global semantic encoding based on the global mask structure on the sequence of the multiple vibration frequency domain embedded vectors to obtain multiple statistical feature vectors; a feature vector cascading unit, which is used to cascade the multiple statistical feature vectors to generate the vibration frequency domain feature vector.

[0014] In the above-mentioned online monitoring system for the processing status of the grinding equipment, the context encoding unit includes: a query vector construction subunit, which is used to arrange the sequence of the multiple vibration frequency domain embedding vectors in one dimension to obtain a vibration frequency domain sequence feature vector; a self-attention subunit, which is used to calculate the product between the vibration frequency domain sequence feature vector and the transposed vector of each vibration frequency domain embedding vector in the sequence of multiple vibration frequency domain embedding vectors to obtain multiple self-attention association matrices; a normalization subunit, which is used to standardize each self-attention association matrix in the multiple self-attention association matrices to obtain multiple standardized self-attention association matrices; an attention calculation subunit, which is used to pass each standardized self-attention association matrix in the multiple standardized self-attention association matrices through the Softmax classification function to obtain multiple probability values; an attention application subunit, which is used to weight each vibration frequency domain embedding vector in the sequence of the multiple vibration frequency domain embedding vectors with each probability value in the multiple probability values ​​as a weight to obtain the multiple statistical feature vectors.

[0015] In the above-mentioned online monitoring system for the processing state of the grinding equipment, the temperature feature encoding module includes: a first scale convolution encoding unit, which is used to use the first convolution layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the temperature input vector respectively using the following first convolution formula to obtain the first scale feature vector;

[0016] Among them, the first convolution formula is:

[0017]

[0018] Wherein, a is the width of the first convolution kernel in the x direction, F(a) is the first convolution kernel parameter vector, G(xa) is the local vector matrix operated with the first convolution kernel function, w is the size of the first convolution kernel, X represents the temperature input vector, and Cov(X) represents one-dimensional convolution encoding of the temperature input vector respectively;

[0019] A second-scale convolutional encoding unit, configured to use the second convolutional layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolutional encoding on the temperature input vector using the following second convolutional formula to obtain the second-scale feature vector;

[0020] Among them, the second convolution formula is:

[0021]

[0022] Wherein, b is the width of the second convolution kernel in the x direction, F(b) is the second convolution kernel parameter vector, G(xb) is the local vector matrix operated with the second convolution kernel function, m is the size of the second convolution kernel, X represents the temperature input vector, and Cov(X) represents one-dimensional convolution encoding of the temperature input vector respectively;

[0023] A cascading unit is used to cascade the first scale feature vector and the second scale feature vector to obtain the temperature time series correlation feature vector.

[0024] In the above-mentioned online monitoring system for the processing status of the grinding equipment, the fusion module includes: respectively calculating the covariance matrix of the vibration frequency domain eigenvector and the temperature time series associated eigenvector to obtain a first covariance matrix and a second covariance matrix; performing singular value decomposition on the first covariance matrix to obtain multiple first singular values ​​and multiple first singular vectors corresponding to the multiple first singular values; performing singular value decomposition on the second covariance matrix to obtain multiple second singular values ​​and multiple second singular vectors corresponding to the multiple second singular values; extracting the first singular vectors corresponding to the first k first singular values ​​from the multiple first singular vectors and extracting the second singular vectors corresponding to the first k second singular values ​​from the multiple second singular vectors as the basis of the low-dimensional sparse intermediate space; mapping the vibration frequency domain eigenvector and the temperature time series associated eigenvector to the low-dimensional sparse intermediate space respectively to obtain a first type of homography mapping eigenvector and a second type of homography mapping eigenvector; multiplying the transposed vectors of the first type of homography mapping eigenvector and the second type of homography mapping eigenvector to obtain the classification feature matrix.

[0025] In the above-mentioned online monitoring system for the processing status of the grinding equipment, the classification module includes: using the classifier to process the classification feature matrix with the following classification formula to obtain the classification result;

[0026] Wherein, the formula is: O = softmax{(W c ,B c )│Project(M)}, where Project(M) represents converting the classification feature matrix into a vector, W c is the weight matrix, B c represents a bias vector, softmax represents a normalized exponential function, and O represents the classification result.

[0027] According to another aspect of the present application, there is provided an online monitoring method for a grinding equipment processing state, comprising:

[0028] Acquire the vibration values ​​of the grinding device collected by the vibration sensor at a plurality of predetermined time points within a predetermined time period and the temperature values ​​of the grinding device collected by the temperature sensor at a plurality of predetermined time points;

[0029] Based on Fourier transformation, the vibration values ​​of the grinding equipment collected by the vibration sensor at a plurality of predetermined time points within a predetermined time period are converted into a frequency domain to obtain a plurality of vibration frequency domain statistical features;

[0030] Passing the plurality of vibration frequency domain statistical features through a sequence encoder of a joint encoder to obtain a vibration frequency domain feature vector, wherein the sequence encoder is a context encoder including an embedding layer;

[0031] Arranging the temperature values ​​of the grinding equipment collected by the temperature sensor at the plurality of predetermined time points as a temperature input vector and then performing the temperature extraction through a multi-scale neighborhood feature extraction module to obtain a temperature time series correlation feature vector;

[0032] Fusion of the vibration frequency domain feature vector and the temperature time series correlation feature vector to obtain a classification feature matrix;

[0033] The classification feature matrix is ​​passed through a classifier to obtain a classification result, and the classification result indicates whether the processing state of the grinding equipment is abnormal.

[0034] In the above-mentioned online monitoring method for the processing status of the grinding equipment, the multiple vibration frequency domain statistical features are passed through a sequence encoder of a joint encoder to obtain a vibration frequency domain feature vector. The sequence encoder is a context encoder including an embedding layer, including: using the embedding layer of the sequence encoder to convert the multiple vibration frequency domain statistical features into input vectors respectively to obtain a sequence of multiple vibration frequency domain embedded vectors; using the converter-based Bert model of the sequence encoder to perform global semantic encoding based on a global mask structure on the sequence of the multiple vibration frequency domain embedded vectors to obtain multiple statistical feature vectors; and cascading the multiple statistical feature vectors to generate the vibration frequency domain feature vector.

[0035] Compared with the prior art, the present application provides an online monitoring system and method for the processing status of a grinding device, which first obtains the vibration values ​​of the grinding device collected by the vibration sensor at multiple predetermined time points within a predetermined time period and the temperature values ​​of the grinding device collected by the temperature sensor at multiple predetermined time points, then transforms the vibration values ​​through Fourier transform and passes them through the sequence encoder of the joint encoder to obtain a vibration frequency domain feature vector, then arranges the temperature values ​​into a temperature input vector and obtains a temperature time series associated feature vector through a multi-scale neighborhood feature extraction module, further, fuses the vibration frequency domain feature vector and the temperature time series associated feature vector to obtain a classification feature matrix, finally, passes the classification feature matrix through a classifier to obtain a classification result, and the classification result indicates whether the processing status of the grinding equipment is abnormal, so as to further improve the efficiency and accuracy of the grinding equipment processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0037] Figure 1 It is a block diagram of an online monitoring system for the processing status of a grinding device according to an embodiment of the present application.

[0038] Figure 2 4 is a schematic diagram of an online monitoring system for the processing status of a grinding device according to an embodiment of the present application.

[0039] Figure 3 It is a block diagram of a frequency domain feature coding module in an online monitoring system for the processing status of a grinding device according to an embodiment of the present application.

[0040] Figure 4 It is a block diagram of a context encoding unit in a frequency domain feature encoding module according to an embodiment of the present application.

[0041] Figure 5 It is a block diagram of a temperature feature encoding module in an online monitoring system of a grinding equipment processing state according to an embodiment of the present application.

[0042] Figure 6 It is a flow chart of an online monitoring method of the processing status of a grinding device according to an embodiment of the present application. DETAILED DESCRIPTION

[0043] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0044] Scenario Overview

[0045] As mentioned in the background technology above, the processing status of the grinding equipment directly affects the quality and efficiency of the processing process. If there is a problem with the grinding equipment, problems such as increased surface roughness, dimensional deviation or uneven grinding will occur. It will also affect the processing efficiency, equipment life and maintenance cost.

[0046] In the current prior art, manual monitoring of the processing state of grinding equipment is easily affected by the subjective consciousness and experience of the operator. Different operators may have different judgments on the same processing state, and long-term manual monitoring is likely to cause fatigue and inattention of the operator, which may miss some subtle changes or abnormal situations. In addition, manual monitoring may be intermittent, and it is impossible to obtain information on the processing state in real time and it is difficult to achieve continuous monitoring. Therefore, a more optimized online monitoring system and method for the processing state of grinding equipment is desired.

[0047] Accordingly, in the grinding equipment processing state, the increase in temperature can reduce the hardness of the material, thereby reducing the generation of cutting forces and vibrations. When the temperature of the material increases, its crystal structure may change, causing the material to become softer or plastic. This will reduce the cutting force during the cutting process and reduce the vibration caused by the cutting force. In addition, the increase in temperature can also improve the lubrication conditions, reduce the coefficient of friction, and further reduce the generation of cutting forces and vibrations. At high temperatures, the viscosity of the lubricant may decrease, making it easier to form a lubricating film in the cutting area, reducing friction and heat generation.

[0048] However, it should be noted that temperature increase does not always significantly reduce cutting force and vibration. Some materials may exhibit the phenomenon of reverse hardening at high temperature, that is, the temperature increase makes the material harder, thereby increasing the cutting force and the vibration caused by the cutting force. In addition, too high a temperature may cause the material to burn, soften or deform, and cause thermal expansion and thermal stress of the equipment, increasing the risk of wear and damage to the equipment. Temperature reduction increases the hardness of the material, thereby increasing the generation of cutting force and vibration, increasing the vibration value of the grinding equipment, and reducing the risk of the material not being softened or burned by high temperature. However, temperature reduction may increase the viscosity of the cutting fluid, which is not conducive to the lubrication and cooling effect of grinding, and too low a temperature may increase the brittleness of the material, which is prone to the risk of cracks and fractures. Therefore, there is a high-dimensional implicit characteristic logical association between the vibration value of the grinding equipment collected by the vibration sensor at multiple predetermined time points within a predetermined time period and the temperature value of the grinding equipment collected by the temperature sensor at the multiple predetermined time points. Balancing the association and connection between the two is the key to solving the problem of online monitoring of the processing state of the grinding equipment.

[0049] In recent years, deep learning and neural networks have been widely used in computer vision, natural language processing, text signal processing and other fields. In addition, deep learning and neural networks have also demonstrated a level close to or even beyond that of humans in image classification, object detection, semantic segmentation, text translation and other fields.

[0050] In particular, considering that the vibration values ​​of the grinding equipment under different processing states have different frequency components, in order to better determine the type of processing state, the inventors of the present application attempted to combine the traditional frequency domain analysis method and the deep neural network model to construct an abnormal diagnosis model for the processing state of the grinding equipment.

[0051] Specifically, in the technical solution of the present application, the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points within a predetermined time period are first converted into the frequency domain by Fourier transformation to obtain multiple vibration frequency domain statistical features. It should be understood that, compared with conventional vibration feature analysis methods, in the technical solution of the present application, a frequency domain analysis method based on Fourier transformation is adopted, and the deep neural network model can be used to extract high-dimensional implicit local correlation features for online monitoring from the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points within a predetermined time period.

[0052] Next, the plurality of vibration frequency domain statistical features are passed through a sequence encoder of a joint encoder to obtain a vibration frequency domain feature vector, where the sequence encoder is a context encoder including an embedding layer. In this way, encoding the plurality of vibration frequency domain statistical features through a context encoder can extract a feature distribution based on global statistical correlation features, thereby obtaining a vibration frequency domain feature vector.

[0053] Furthermore, it is considered that the temperature values ​​of the grinding equipment collected by the temperature sensor at the plurality of predetermined time points can be observed and analyzed from different perspectives by adopting a multi-scale approach, so as to fully explore the correlation characteristics in the time domain. Among them, the large convolution kernel will extract features from the large-scale time series neighborhood, and the influence of each numerical value in its neighborhood will be smaller, thereby reducing the fluctuation of the input data and reducing the influence of the noise on the output features. However, the large-scale convolution kernel weakens the difference in numerical changes, which is easy to cause the problem of excessive smoothing, making the output features lose the ability to distinguish. In contrast, the small-scale convolution kernel can better retain the information in the input data, but it is also more susceptible to the interference of the noise. Taking into account the characteristics of convolutions of different scales, convolution units of different sizes are combined to extract features of different time series scales, and then feature splicing is used to complete feature fusion, thereby obtaining multi-scale neighborhood features. In this way, the output features contain both the smoothed features and the original input features.

[0054] It should be understood that the temperature values ​​of different grinding equipment collected by temperature sensors at the plurality of predetermined time points may produce different temperature variation patterns. The use of a multi-scale approach can better adapt to the temperature variation of different grinding equipment processing states. By using information at multiple scales, more temperature features can be obtained, thereby improving the accuracy and stability of the classifier. Therefore, after the temperature values ​​of the grinding equipment collected by the temperature sensor at the plurality of predetermined time points are arranged as temperature input vectors, a multi-scale neighborhood feature extraction module is used to obtain a temperature time series correlation feature vector.

[0055] Next, the vibration frequency domain feature vector and the temperature time series associated feature vector are fused to obtain a classification feature matrix. Specifically, the vibration frequency domain feature vector and the temperature time series associated feature vector respectively extract different feature information, and there may be complementarity between them, that is, the feature expression in one feature information may be compensated in another feature information. Therefore, fusing these two feature vectors can make full use of their complementarity and improve the accuracy and robustness of classification. In addition, the use of a classification feature matrix can also reduce computational complexity. It is more efficient to fuse the feature vectors of two scales together than to use each feature vector separately. Finally, the classification feature matrix is ​​passed through a classifier to obtain a classification result, which is used to adaptively determine whether the processing state of the grinding equipment is abnormal.

[0056] In particular, considering that the vibration frequency domain feature vector and the temperature time series correlation feature vector usually have high dimensions, this will lead to the problem of dimensionality curse. The data points in the high-dimensional feature space are sparsely distributed, and the distance between samples is difficult to measure accurately, and it is also easy to cause overfitting problems. By reducing the dimension of the feature space, the dimensionality curse problem can be alleviated. In addition, each dimension in a high-dimensional feature vector may represent a different feature or feature combination, but not all features contribute to the classification task. In a low-dimensional feature space, it is easier to understand the meaning and role of each feature, and improve the interpretability of the feature. There may be redundant or noisy features in the high-dimensional feature vector, which may not be of substantial help to the classification task and may even interfere with the learning process of the model. By reducing the dimension of the feature space, the impact of redundant and noisy features can be reduced and the robustness of the feature can be improved.

[0057] Specifically, the vibration frequency domain eigenvector and the temperature time series associated eigenvector are fused to obtain a classification feature matrix, including: respectively calculating the covariance matrix of the vibration frequency domain eigenvector and the temperature time series associated eigenvector to obtain a first covariance matrix and a second covariance matrix; performing singular value decomposition on the first covariance matrix to obtain a plurality of first singular values ​​and a plurality of first singular vectors corresponding to the plurality of first singular values; performing singular value decomposition on the second covariance matrix to obtain a plurality of second singular values ​​and a plurality of second singular vectors corresponding to the plurality of second singular values. extract first singular vectors corresponding to the first k first singular values ​​from the multiple first singular vectors and extract second singular vectors corresponding to the first k second singular values ​​from the multiple second singular vectors as the basis of a low-dimensional sparse intermediate space; map the vibration frequency domain eigenvector and the temperature time series association eigenvector to the low-dimensional sparse intermediate space respectively to obtain a first type of homography mapping eigenvector and a second type of homography mapping eigenvector; multiply the transposed vectors of the first type of homography mapping eigenvector and the second type of homography mapping eigenvector to obtain the classification feature matrix.

[0058] Here, the vibration frequency domain feature vectors and temperature time series associated feature vectors with different feature manifold modes are mapped to a common low-dimensional sparse intermediate space through homography-like transformation to linearly embed and fuse the semantic features of different modes in the low-dimensional sparse space to obtain a classification feature matrix with more expressive and discriminative capabilities. In other words, the feature fusion problem of high-dimensional feature vectors is transformed into the problem of homography-like mapping and linear fusion of low-dimensional feature space, and in this way, the interpretability and robustness of the classification feature matrix are increased.

[0059] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0060] Exemplary Systems

[0061] Figure 1 FIG. 1 is a block diagram of an online monitoring system for the processing status of a grinding device according to an embodiment of the present application. Figure 1 As shown, according to the embodiment of the present application, the online monitoring system for the processing state of the grinding equipment includes: a data acquisition module 110, which is used to obtain the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points in a predetermined time period and the temperature values ​​of the grinding equipment collected by the temperature sensor at multiple predetermined time points; a frequency domain feature extraction module 120, which is used to convert the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points in a predetermined time period into the frequency domain based on Fourier transform to obtain multiple vibration frequency domain statistical features; a frequency domain feature encoding module 130, which is used to convert the multiple vibration frequency domain statistical features into the frequency domain through a joint encoder. A sequence encoder is used to obtain a vibration frequency domain feature vector, and the sequence encoder is a context encoder including an embedding layer; a temperature feature encoding module 140 is used to arrange the temperature values ​​of the grinding equipment collected by the temperature sensor at the plurality of predetermined time points as a temperature input vector and then pass it through a multi-scale neighborhood feature extraction module to obtain a temperature time series associated feature vector; a fusion module 150 is used to fuse the vibration frequency domain feature vector and the temperature time series associated feature vector to obtain a classification feature matrix; a classification module 160 is used to pass the classification feature matrix through a classifier to obtain a classification result, and the classification result indicates whether the processing state of the grinding equipment is abnormal.

[0062] Figure 2 FIG. 1 is a schematic diagram of an online monitoring system for the processing status of a grinding device according to an embodiment of the present application. Figure 2As shown, according to the embodiment of the present application, the online monitoring system for the processing state of the grinding equipment includes: first, obtaining the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points within a predetermined time period and the temperature values ​​of the grinding equipment collected by the temperature sensor at multiple predetermined time points. Then, based on Fourier transformation, the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points within a predetermined time period are converted into the frequency domain to obtain multiple vibration frequency domain statistical features. Next, the multiple vibration frequency domain statistical features are passed through the sequence encoder of the joint encoder to obtain a vibration frequency domain feature vector, and the sequence encoder is a context encoder including an embedding layer. Further, after the temperature values ​​of the grinding equipment collected by the temperature sensor at the multiple predetermined time points are arranged as temperature input vectors, they are passed through a multi-scale neighborhood feature extraction module to obtain a temperature time series correlation feature vector. Specifically, the vibration frequency domain feature vector and the temperature time series correlation feature vector are fused to obtain a classification feature matrix. Finally, the classification feature matrix is ​​passed through a classifier to obtain a classification result, and the classification result indicates whether the processing state of the grinding equipment is abnormal.

[0063] In the above-mentioned online monitoring system 100 for the processing state of the grinding equipment, the data acquisition module 110 is used to obtain the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points within a predetermined time period and the temperature values ​​of the grinding equipment collected by the temperature sensor at multiple predetermined time points. It should be understood that in the processing state of the grinding equipment, the increase in temperature can reduce the hardness of the material, thereby reducing the generation of cutting force and vibration. When the temperature of the material increases, its crystal structure may change, causing the material to become softer or plastic. This will reduce the cutting force during the cutting process and reduce the vibration caused by the cutting force. In addition, the increase in temperature can also improve the lubrication conditions, reduce the friction coefficient, and further reduce the generation of cutting force and vibration. At high temperatures, the viscosity of the lubricant may decrease, making it easier to form a lubricating film in the cutting area, reducing friction and heat generation.

[0064] However, it should be noted that temperature increase does not always significantly reduce cutting force and vibration. Some materials may exhibit the phenomenon of reverse hardening at high temperature, that is, the temperature increase makes the material harder, thereby increasing the cutting force and the vibration caused by the cutting force. In addition, too high a temperature may cause the material to burn, soften or deform, and cause thermal expansion and thermal stress of the equipment, increasing the risk of wear and damage to the equipment. Temperature reduction increases the hardness of the material, thereby increasing the generation of cutting force and vibration, increasing the vibration value of the grinding equipment, and reducing the risk of the material not being softened or burned by high temperature. However, temperature reduction may increase the viscosity of the cutting fluid, which is not conducive to the lubrication and cooling effect of grinding, and too low a temperature may increase the brittleness of the material, which is prone to the risk of cracks and fractures. Therefore, there is an implicit correlation in the high-dimensional feature space between the vibration value of the grinding equipment collected by the vibration sensor at multiple predetermined time points in a predetermined time period and the temperature value of the grinding equipment collected by the temperature sensor at the multiple predetermined time points. Balancing the correlation and connection between the two is the key to solving the problem of online monitoring of the processing state of the grinding equipment. In particular, considering that the vibration values ​​of the grinding equipment under different processing states have different frequency components, in order to better determine the type of processing state, the inventors of the present application attempted to combine the traditional frequency domain analysis method and the deep neural network model to construct an abnormal diagnosis model for the processing state of the grinding equipment.

[0065] In the above-mentioned online monitoring system 100 for the processing state of the grinding equipment, the frequency domain feature extraction module 120 is used to convert the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points within a predetermined time period into the frequency domain based on Fourier transformation to obtain multiple vibration frequency domain statistical features. It should be understood that in the technical solution of the present application, the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points within a predetermined time period are first converted into the frequency domain through Fourier transformation to obtain multiple vibration frequency domain statistical features. It should be understood that compared with conventional vibration feature analysis methods, in the technical solution of the present application, a frequency domain analysis method based on Fourier transformation is adopted, and the deep neural network model can be used to extract high-dimensional implicit local correlation features for online monitoring from the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points within a predetermined time period.

[0066] In the above-mentioned online monitoring system 100 for the processing state of the grinding equipment, the frequency domain feature encoding module 130 is used to obtain a vibration frequency domain feature vector by passing the multiple vibration frequency domain statistical features through a sequence encoder of a joint encoder, and the sequence encoder is a context encoder including an embedding layer. It should be understood that encoding the multiple vibration frequency domain statistical features through a context encoder can extract a feature distribution based on global statistical correlation features, thereby obtaining a vibration frequency domain feature vector.

[0067] Figure 3 FIG. 4 is a block diagram of a frequency domain feature encoding module in an online monitoring system for the processing state of a grinding device according to an embodiment of the present application. Figure 3 As shown, in a specific embodiment of the present application, the frequency domain feature encoding module 130 includes a vector embedding unit 131, which is used to use the embedding layer of the sequence encoder to convert the multiple vibration frequency domain statistical features into input vectors respectively to obtain a sequence of multiple vibration frequency domain embedded vectors; a context encoding unit 132, which is used to use the converter-based Bert model of the sequence encoder to perform global semantic encoding based on a global mask structure on the sequence of the multiple vibration frequency domain embedded vectors to obtain multiple statistical feature vectors; a feature vector cascading unit 133, which is used to cascade the multiple statistical feature vectors to generate the vibration frequency domain feature vector.

[0068] Figure 4 FIG. 1 is a block diagram of a context encoding unit in a frequency domain feature encoding module according to an embodiment of the present application. Figure 4 As shown, in a specific embodiment of the present application, the context encoding unit 132 includes a query vector construction subunit 1321, which is used to arrange the sequence of the multiple vibration frequency domain embedding vectors in one dimension to obtain a vibration frequency domain sequence feature vector; a self-attention subunit 1322, which is used to calculate the product between the vibration frequency domain sequence feature vector and the transposed vector of each vibration frequency domain embedding vector in the sequence of the multiple vibration frequency domain embedding vectors to obtain multiple self-attention association matrices; a normalization subunit 1323, which is used to normalize each self-attention association matrix in the multiple self-attention association matrices to obtain multiple standardized self-attention association matrices; an attention calculation subunit 1324, which is used to pass each standardized self-attention association matrix in the multiple standardized self-attention association matrices through a Softmax classification function to obtain multiple probability values;

[0069] The attention applying subunit 1325 is used to weight each vibration frequency domain embedding vector in the sequence of the multiple vibration frequency domain embedding vectors using each probability value in the multiple probability values ​​as a weight to obtain the multiple statistical feature vectors.

[0070] In the above-mentioned online monitoring system 100 for the processing state of the grinding equipment, the temperature feature encoding module 140 is used to arrange the temperature values ​​of the grinding equipment collected by the temperature sensor at the plurality of predetermined time points as temperature input vectors and then pass through the multi-scale neighborhood feature extraction module to obtain the temperature time series correlation feature vector. It should be understood that, considering the use of a multi-scale approach, the temperature values ​​of the grinding equipment collected by the temperature sensor at the plurality of predetermined time points can be observed and analyzed from different perspectives, and the correlation features in the time domain can be fully explored. Among them, the large convolution kernel will extract features from the large-scale time series neighborhood, and the influence of each numerical value in its neighborhood is smaller, thereby reducing the fluctuation of the input data and reducing the influence of the noise on the output features. However, the large-scale convolution kernel reduces the difference in numerical changes, which is easy to cause the problem of excessive smoothing, so that the output features lose the ability to distinguish. In contrast, the small-scale convolution kernel can better retain the information in the input data, but it is also more susceptible to the interference of the noise. Taking into account the characteristics of convolutions of different scales, convolution units of different sizes are combined to extract features of different time series scales, and then feature splicing is used to complete feature fusion, thereby obtaining multi-scale neighborhood features. In this way, the output features include both smoothed features and original input features. Specifically, the temperature values ​​of different grinding equipment collected by temperature sensors at the multiple predetermined time points may produce different temperature change patterns. The multi-scale approach can better adapt to the temperature changes of different grinding equipment processing states. By using information at multiple scales, more temperature features can be obtained to improve the accuracy and stability of the classifier. Therefore, after arranging the temperature values ​​of the grinding equipment collected by the temperature sensor at the multiple predetermined time points as a temperature input vector, a multi-scale neighborhood feature extraction module is used to obtain a temperature time series correlation feature vector.

[0071] Figure 5 FIG. 1 is a block diagram of a temperature feature encoding module in an online monitoring system for the processing state of a grinding device according to an embodiment of the present application. Figure 5 As shown, in a specific embodiment of the present application, the temperature feature encoding module 140 includes a first scale convolution encoding unit 141, which is used to use the first convolution layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the temperature input vector respectively using the following first convolution formula to obtain the first scale feature vector;

[0072] Among them, the first convolution formula is:

[0073]

[0074] Wherein, a is the width of the first convolution kernel in the x direction, F(a) is the first convolution kernel parameter vector, G(xa) is the local vector matrix operated with the first convolution kernel function, W is the size of the first convolution kernel, X represents the temperature input vector, and Cov(X) represents one-dimensional convolution encoding of the temperature input vector respectively;

[0075] A second scale convolutional coding unit 142, configured to use the second convolutional layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolutional coding on the temperature input vector using the following second convolutional formula to obtain the second scale feature vector;

[0076] Among them, the second convolution formula is:

[0077]

[0078] Wherein, b is the width of the second convolution kernel in the x direction, F(b) is the second convolution kernel parameter vector, G(xb) is the local vector matrix operated with the second convolution kernel function, m is the size of the second convolution kernel, X represents the temperature input vector, cov(X) represents one-dimensional convolution encoding of the temperature input vector respectively; the cascade unit 143 is used to cascade the first scale feature vector and the second scale feature vector to obtain the temperature time series correlation feature vector.

[0079] In the above-mentioned online monitoring system 100 for the processing state of the grinding equipment, the fusion module 150 is used to fuse the vibration frequency domain feature vector and the temperature time series associated feature vector to obtain a classification feature matrix. It should be understood that considering that the vibration frequency domain feature vector and the temperature time series associated feature vector usually have a high dimension, this will lead to the dimensionality disaster problem. The data points in the high-dimensional feature space are sparsely distributed, and the distance between samples is difficult to measure accurately, and it is also easy to cause overfitting problems. By reducing the dimension of the feature space, the dimensionality disaster problem can be alleviated. In addition, each dimension in the high-dimensional feature vector may represent a different feature or feature combination, but not all features contribute to the classification task. In the low-dimensional feature space, it is easier to understand the meaning and function of each feature, and improve the interpretability of the feature. There may be redundant or noisy features in the high-dimensional feature vector, which may not be of substantial help to the classification task, and may even interfere with the learning process of the model. By reducing the dimension of the feature space, the impact of redundant and noisy features can be reduced and the robustness of the feature can be improved.

[0080] In a specific embodiment of the present technical solution, the fusion module 150 includes: respectively calculating the covariance matrix of the vibration frequency domain eigenvector and the temperature time series associated eigenvector to obtain a first covariance matrix and a second covariance matrix; performing singular value decomposition on the first covariance matrix to obtain multiple first singular values ​​and multiple first singular vectors corresponding to the multiple first singular values; performing singular value decomposition on the second covariance matrix to obtain multiple second singular values ​​and multiple second singular vectors corresponding to the multiple second singular values; extracting the first singular vectors corresponding to the first k first singular values ​​from the multiple first singular vectors and extracting the second singular vectors corresponding to the first k second singular values ​​from the multiple second singular vectors as the basis of the low-dimensional sparse intermediate space; mapping the vibration frequency domain eigenvector and the temperature time series associated eigenvector to the low-dimensional sparse intermediate space respectively to obtain a first type of homography mapping eigenvector and a second type of homography mapping eigenvector; multiplying the transposed vectors of the first type of homography mapping eigenvector and the second type of homography mapping eigenvector to obtain the classification feature matrix.

[0081] Here, the vibration frequency domain feature vectors and temperature time series associated feature vectors with different feature manifold modes are mapped to a common low-dimensional sparse intermediate space through homography-like transformation to linearly embed and fuse the semantic features of different modes in the low-dimensional sparse space to obtain a classification feature matrix with more expressive and discriminative capabilities. In other words, the feature fusion problem of high-dimensional feature vectors is transformed into the problem of homography-like mapping and linear fusion of low-dimensional feature space, and in this way, the interpretability and robustness of the classification feature matrix are increased.

[0082] In the above-mentioned online monitoring system 100 for the processing state of the grinding equipment, the classification module 160 is used to obtain a classification result by passing the classification feature matrix through a classifier, and the classification result indicates whether the processing state of the grinding equipment is abnormal.

[0083] In a specific embodiment of the present technical solution, the classification module 160 uses the classifier to process the classification feature matrix using the following classification formula to obtain the classification result;

[0084] The formula is: O = softmax{(Wc, Bc)│Project(M)}, where Project(M) represents converting the classification feature matrix into a vector, W c is the weight matrix, B c represents a bias vector, softmax represents a normalized exponential function, and O represents the classification result.

[0085] In summary, the embodiment of the present application first obtains the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points within a predetermined time period and the temperature values ​​of the grinding equipment collected by the temperature sensor at multiple predetermined time points, and then obtains the vibration value after Fourier transformation through the sequence encoder of the joint encoder to obtain the vibration frequency domain feature vector, and then arranges the temperature value into a temperature input vector and obtains the temperature time series associated feature vector through a multi-scale neighborhood feature extraction module, and further, the vibration frequency domain feature vector and the temperature time series associated feature vector are fused to obtain a classification feature matrix, and finally, the classification feature matrix is ​​passed through a classifier to obtain a classification result, and the classification result indicates whether the processing state of the grinding equipment is abnormal, so as to further improve the efficiency and accuracy of the grinding equipment processing.

[0086] As described above, the online monitoring system 100 for the processing state of the grinding equipment according to the embodiment of the present application can be implemented in various terminal devices, such as a server deployed with online monitoring of the processing state of the grinding equipment. In one example, the online monitoring system 100 for the processing state of the grinding equipment can be integrated into the terminal device as a software module and / or a hardware module. For example, the online monitoring system 100 for the processing state of the grinding equipment can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the online monitoring system 100 for the processing state of the grinding equipment can also be one of the many hardware modules of the terminal device.

[0087] Alternatively, in another example, the online monitoring system 100 for the processing status of the grinding equipment and the terminal device may also be separate devices, and the online monitoring system 100 for the processing status of the grinding equipment may be connected to the terminal device via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0088] Exemplary Methods

[0089] Figure 6 FIG. 1 is a flow chart of an online monitoring method for the processing status of a grinding device according to an embodiment of the present application. Figure 6As shown, according to the embodiment of the present application, the online monitoring method of the processing state of the grinding equipment includes: S110, obtaining the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points within a predetermined time period and the temperature values ​​of the grinding equipment collected by the temperature sensor at multiple predetermined time points; S120, based on Fourier transform, converting the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points within a predetermined time period into the frequency domain to obtain multiple vibration frequency domain statistical features; S130, passing the multiple vibration frequency domain statistical features through a sequence encoder of a joint encoder to obtain a vibration frequency domain feature vector, wherein the sequence encoder is a context encoder including an embedding layer; S140, arranging the temperature values ​​of the grinding equipment collected by the temperature sensor at the multiple predetermined time points as a temperature input vector and then passing through a multi-scale neighborhood feature extraction module to obtain a temperature time series correlation feature vector; S150, fusing the vibration frequency domain feature vector and the temperature time series correlation feature vector to obtain a classification feature matrix; S160, passing the classification feature matrix through a classifier to obtain a classification result, wherein the classification result indicates whether the processing state of the grinding equipment is abnormal.

[0090] Here, those skilled in the art will appreciate that the specific operations of each step in the above-mentioned method for online monitoring of the processing state of the grinding equipment have been described in the above reference. Figures 1 to 5 The online monitoring system for the processing status of the grinding equipment has been described in detail, and therefore, its repeated description will be omitted.

Claims

1. An online monitoring system for the processing status of a grinding equipment, characterized in that: include: A data acquisition module, used to acquire vibration values ​​of the grinding device at multiple predetermined time points within a predetermined time period acquired by the vibration sensor and temperature values ​​of the grinding device at multiple predetermined time points acquired by the temperature sensor; A frequency domain feature extraction module, used for converting the vibration values ​​of the grinding equipment collected by the vibration sensor at multiple predetermined time points within a predetermined time period into the frequency domain based on Fourier transformation to obtain multiple vibration frequency domain statistical features; A frequency domain feature encoding module, used for passing the plurality of vibration frequency domain statistical features through a sequence encoder of a joint encoder to obtain a vibration frequency domain feature vector, wherein the sequence encoder is a context encoder including an embedding layer; A temperature feature encoding module, used for arranging the temperature values ​​of the grinding equipment collected by the temperature sensor at the plurality of predetermined time points into a temperature input vector and then passing it through a multi-scale neighborhood feature extraction module to obtain a temperature time series correlation feature vector; A fusion module, used for fusing the vibration frequency domain feature vector and the temperature time series correlation feature vector to obtain a classification feature matrix; The classification module is used to obtain a classification result by passing the classification feature matrix through a classifier, and the classification result indicates whether the processing state of the grinding equipment is abnormal.

2. The online monitoring system for the processing status of the grinding equipment according to claim 1 is characterized in that: The frequency domain feature encoding module comprises: A vector embedding unit, configured to use the embedding layer of the sequence encoder to convert the plurality of vibration frequency domain statistical features into input vectors respectively to obtain a sequence of a plurality of vibration frequency domain embedding vectors; a context encoding unit, configured to perform global semantic encoding based on a global mask structure on the sequence of the plurality of vibration frequency domain embedding vectors using a transformer-based Bert model of the sequence encoder to obtain a plurality of statistical feature vectors; The feature vector cascading unit is used to cascade the multiple statistical feature vectors to generate the vibration frequency domain feature vector.

3. The online monitoring system for the processing status of the grinding equipment according to claim 2 is characterized in that: The context encoding unit comprises: A query vector construction subunit, used for one-dimensionally arranging the sequences of the plurality of vibration frequency domain embedding vectors to obtain a vibration frequency domain sequence feature vector; A self-attention subunit, configured to calculate the product between the vibration frequency domain sequence feature vector and the transposed vector of each vibration frequency domain embedding vector in the sequence of the plurality of vibration frequency domain embedding vectors to obtain a plurality of self-attention association matrices; A normalization subunit, used for normalizing each of the plurality of self-attention association matrices to obtain a plurality of normalized self-attention association matrices; an attention calculation subunit, configured to obtain a plurality of probability values ​​by applying a Softmax classification function to each of the plurality of standardized self-attention association matrices; The attention applying subunit is used to weight each vibration frequency domain embedding vector in the sequence of the multiple vibration frequency domain embedding vectors using each probability value in the multiple probability values ​​as a weight to obtain the multiple statistical feature vectors.

4. The online monitoring system for the processing status of the grinding equipment according to claim 3 is characterized in that: The temperature characteristic encoding module comprises: A first scale convolution encoding unit, configured to use the first convolution layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the temperature input vectors respectively using the following first convolution formula to obtain the first scale feature vectors; Among them, the first convolution formula is: Wherein, a is the width of the first convolution kernel in the x direction, F(a) is the first convolution kernel parameter vector, G(xa) is the local vector matrix operated with the first convolution kernel function, W is the size of the first convolution kernel, X represents the temperature input vector, and Cov(X) represents one-dimensional convolution encoding of the temperature input vector respectively; A second-scale convolutional encoding unit, configured to use the second convolutional layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolutional encoding on the temperature input vector using the following second convolutional formula to obtain the second-scale feature vector; Among them, the second convolution formula is: Wherein, b is the width of the second convolution kernel in the x direction, F(b) is the second convolution kernel parameter vector, G(xb) is the local vector matrix operated with the second convolution kernel function, m is the size of the second convolution kernel, X represents the temperature input vector, and Cov(X) represents one-dimensional convolution encoding of the temperature input vector respectively; A cascading unit is used to cascade the first scale feature vector and the second scale feature vector to obtain the temperature time series correlation feature vector.

5. The online monitoring system for the processing status of the grinding equipment according to claim 4 is characterized in that: The fusion module comprises: Calculating the covariance matrix of the vibration frequency domain eigenvector and the temperature time series associated eigenvector respectively to obtain a first covariance matrix and a second covariance matrix; Performing singular value decomposition on the first covariance matrix to obtain a plurality of first singular values ​​and a plurality of first singular vectors corresponding to the plurality of first singular values; Performing singular value decomposition on the second covariance matrix to obtain a plurality of second singular values ​​and a plurality of second singular vectors corresponding to the plurality of second singular values; Extracting first singular vectors corresponding to first k first singular values ​​from the plurality of first singular vectors and extracting second singular vectors corresponding to first k second singular values ​​from the plurality of second singular vectors as a basis of a low-dimensional sparse intermediate space; Mapping the vibration frequency domain feature vector and the temperature time series associated feature vector to the low-dimensional sparse intermediate space to obtain a first type homography mapping feature vector and a second type homography mapping feature vector; The first type homography mapping feature vector and the transposed vector of the second type homography mapping feature vector are multiplied to obtain the classification feature matrix.

6. The online monitoring system for the processing status of the grinding equipment according to claim 5, characterized in that: The classification module includes: using the classifier to process the classification feature matrix according to the following classification formula to obtain the classification result; Wherein, the formula is: O = softmax{(W c ,B c )│Project(M)}, where Project(M) represents converting the classification feature matrix into a vector, W c is the weight matrix, B c represents a bias vector, softmax represents a normalized exponential function, and O represents the classification result.

7. An online monitoring method for the processing status of a grinding device, characterized in that: include: Acquire the vibration values ​​of the grinding device collected by the vibration sensor at a plurality of predetermined time points within a predetermined time period and the temperature values ​​of the grinding device collected by the temperature sensor at a plurality of predetermined time points; Based on Fourier transformation, the vibration values ​​of the grinding equipment collected by the vibration sensor at a plurality of predetermined time points within a predetermined time period are converted into a frequency domain to obtain a plurality of vibration frequency domain statistical features; Passing the plurality of vibration frequency domain statistical features through a sequence encoder of a joint encoder to obtain a vibration frequency domain feature vector, wherein the sequence encoder is a context encoder including an embedding layer; Arranging the temperature values ​​of the grinding equipment collected by the temperature sensor at the plurality of predetermined time points as a temperature input vector and then performing the temperature extraction through a multi-scale neighborhood feature extraction module to obtain a temperature time series correlation feature vector; Fusion of the vibration frequency domain feature vector and the temperature time series correlation feature vector to obtain a classification feature matrix; The classification feature matrix is ​​passed through a classifier to obtain a classification result, and the classification result indicates whether the processing state of the grinding equipment is abnormal.

8. The method for online monitoring of the processing state of a grinding device according to claim 7, characterized in that: The plurality of vibration frequency domain statistical features are passed through a sequence encoder of a joint encoder to obtain a vibration frequency domain feature vector, wherein the sequence encoder is a context encoder including an embedding layer, including: Using the embedding layer of the sequence encoder, the plurality of vibration frequency domain statistical features are respectively converted into input vectors to obtain a sequence of a plurality of vibration frequency domain embedding vectors; Using the transformer-based Bert model of the sequence encoder, a global semantic encoding based on a global mask structure is performed on the sequence of the plurality of vibration frequency domain embedding vectors to obtain a plurality of statistical feature vectors; The plurality of statistical feature vectors are concatenated to generate the vibration frequency domain feature vector.

9. The method for online monitoring of the processing state of a grinding device according to claim 8, characterized in that: Using the converter-based Bert model of the sequence encoder, a global semantic encoding based on a global mask structure is performed on the sequence of the plurality of vibration frequency domain embedding vectors to obtain a plurality of statistical feature vectors, including: Arranging the sequences of the plurality of vibration frequency domain embedding vectors in one dimension to obtain a vibration frequency domain sequence feature vector; Calculating the product between the vibration frequency domain sequence feature vector and the transposed vector of each vibration frequency domain embedding vector in the sequence of multiple vibration frequency domain embedding vectors to obtain multiple self-attention association matrices; Normalizing each of the plurality of self-attention association matrices to obtain a plurality of standardized self-attention association matrices; Passing each of the plurality of standardized self-attention association matrices through a Softmax classification function to obtain a plurality of probability values; Each vibration frequency domain embedding vector in the sequence of the plurality of vibration frequency domain embedding vectors is weighted using each probability value in the plurality of probability values ​​as a weight to obtain the plurality of statistical feature vectors.

10. The method for online monitoring of the processing state of grinding equipment according to claim 9, characterized in that: Arranging the temperature values ​​of the grinding equipment collected by the temperature sensor at the plurality of predetermined time points as a temperature input vector and then extracting the temperature time series associated feature vector through a multi-scale neighborhood feature extraction module includes: Using the first convolution layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the temperature input vectors respectively using the following first convolution formula to obtain the first scale feature vector; Among them, the first convolution formula is: Wherein, a is the width of the first convolution kernel in the x direction, F(a) is the first convolution kernel parameter vector, G(xa) is the local vector matrix operated with the first convolution kernel function, w is the size of the first convolution kernel, X represents the temperature input vector, and Cov(X) represents one-dimensional convolution encoding of the temperature input vector respectively; Using the second convolution layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the temperature input vector using the following second convolution formula to obtain the second scale feature vector; Among them, the second convolution formula is: Wherein, b is the width of the second convolution kernel in the x direction, F(b) is the second convolution kernel parameter vector, G(xb) is the local vector matrix operated with the second convolution kernel function, m is the size of the second convolution kernel, X represents the temperature input vector, and Cov(X) represents one-dimensional convolution encoding of the temperature input vector respectively; The first scale feature vector and the second scale feature vector are cascaded to obtain the temperature time series correlation feature vector.

Images