CN116560895A - Fault diagnosis method for mechanical equipment - Google Patents

Abstract

The disclosure provides a fault diagnosis method for mechanical equipment, relates to the technical field of mechanical equipment fault diagnosis, and can be used for a scene of fault diagnosis of the mechanical equipment based on rolling bearing vibration signals. The method comprises the following steps: determining a metric matrix according to the metric information of the vibration signal; determining distance information in the measurement information by constructing a nearest neighbor point neighbor graph; mapping different information represented by the distance information into a nuclear space and determining a nuclear matrix and a nuclear space pheromone; reconstructing the measurement matrix according to the nuclear matrix and the nuclear space pheromone fusion tag discrimination information; and performing dimension reduction processing on the characteristics of the reconstructed metric matrix, and inputting the dimension reduction processed result and label discrimination information into a classifier to obtain a fault classification result. The fault diagnosis method can reduce the influence of super parameters, reduce the loss of data in the characteristic extraction process and improve the accuracy of fault classification.

Description

Fault Diagnosis Method for Mechanical Equipment

technical field

The present disclosure relates to the technical field of fault diagnosis of mechanical equipment, in particular, to a fault diagnosis method for mechanical equipment.

Background technique

Traditional fault diagnosis methods have laid the foundation for extracting significant fault features in signals. However, in the process of intelligent development, the complexity of mechanical equipment structure and operating status has increased significantly. Research on intelligent fault diagnosis technology has become an important aspect of mechanical equipment fault diagnosis. Task. As a common tool for intelligent fault diagnosis, artificial intelligence contains many machine learning algorithms, which can effectively extract the salient features in the data set, and have been successfully applied to the fault diagnosis of various rotating equipment. Compared with other traditional fault diagnosis techniques higher accuracy than .

At present, the data volume of mechanical equipment operation data is increasing exponentially, and the machine learning algorithm used for mechanical equipment fault diagnosis will be affected by the data volume, and the traditional machine learning algorithm does not integrate the physical information of the same data in different spaces. It will cause the loss of data and reduce the accuracy of fault classification.

It should be noted that the information disclosed in the above background section is only for enhancing the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to those of ordinary skill in the art.

Contents of the invention

The purpose of the embodiments of the present disclosure is to provide a fault diagnosis method, which can reduce the influence of hyperparameters on machine learning algorithms, and can integrate multi-space feature information to reduce data loss, thereby improving the performance of fault classifiers. diagnostic accuracy.

According to the first aspect of the embodiments of the present disclosure, there is provided a fault diagnosis method for mechanical equipment, including:

Obtain the measurement information of the vibration signal of the mechanical equipment, and determine the measurement matrix through the measurement information;

Construct the true nearest neighbor neighbor graph according to the adaptive neighbor strategy, and determine the distance information in the measurement information through the true neighbor neighbor graph;

determining at least two kernel spaces according to a preset exponential linear kernel function;

Map different information represented by distance information to different nuclear spaces, and determine the nuclear matrix and nuclear space pheromone corresponding to the nuclear space;

Reconstruct the metric matrix according to the discriminant information of the nuclear matrix and nuclear space pheromone fusion labels;

The dimensionality reduction processing is performed on the features of the reconstructed measurement matrix, and the dimensionality reduction processing results and label discrimination information are input into the pre-trained fault classifier to obtain the fault classification results.

According to the second aspect of the embodiments of the present disclosure, there is provided a fault diagnosis device for mechanical equipment, including:

The measurement information acquisition module is used to obtain the measurement information of the vibration signal of the mechanical equipment, and determine the measurement matrix through the measurement information;

A true neighbor neighbor graph building module, used to construct a true neighbor neighbor graph according to an adaptive neighbor strategy, and determine distance information in the measurement information through the true neighbor neighbor graph;

A kernel space construction module, configured to determine at least two kernel spaces according to a preset exponential linear kernel function;

The feature extraction module is used to map different information represented by the distance information into different nuclear spaces, and determine the nuclear matrix and nuclear space pheromone corresponding to the nuclear space;

The metric matrix reconstruction module is used to reconstruct the metric matrix according to the kernel matrix and nuclear space pheromone fusion label discrimination information;

The dimensionality reduction module is used to perform dimensionality reduction processing on the features of the reconstructed measurement matrix, and input the dimensionality reduction processing results and label discrimination information into the pre-trained fault classifier to obtain fault classification results.

The technical solutions provided by the embodiments of the present disclosure may include the following beneficial effects:

Through the embodiments of the present disclosure, the metric information of the vibration signal is obtained, and the metric matrix is determined; the distance information in the metric information is determined by constructing a true neighbor point neighbor graph; and then different information represented by the distance information is mapped to different kernel spaces, and Determine the kernel matrix and nuclear space pheromone; then reconstruct the measurement matrix according to the fusion label discrimination information of the nuclear matrix and nuclear space pheromone; perform dimensionality reduction processing on the features of the reconstructed measurement matrix, and reduce the dimensionality The results and label discrimination information are input into the classifier to obtain the fault classification result. On the one hand, by constructing a true neighbor map based on the characteristic data of the vibration signal, the true neighbor map can be determined by screening the information of the vibration signal. Traditional manifold learning methods usually rely on setting hyperparameters to determine the true neighbor map. Quantity, and the traditional hyperparameters need to be set manually, which has a great impact on the performance and results of the algorithm. Based on the fusion of distance information and angle information of neighboring points, the real neighboring points are dynamically screened, and the algorithm can adaptively select the appropriate neighboring points. It does not rely on manually set hyperparameters, thereby reducing the influence of hyperparameters on machine learning algorithms; on the other hand, the kernel space is determined based on the exponential linear kernel function, and the same features in multiple high-dimensional spaces are fused together Reconstructing the metric matrix and determining the kernel space through the kernel function can better capture the complex interaction between nonlinear patterns and features, thereby providing a richer data representation and fusing the same features in multiple high-dimensional spaces together Reconstructing the metric matrix can reflect the essential characteristics of the data and better retain the structure and relationship of the original data, which can avoid the loss and distortion of information, thereby reducing the loss of data, ensuring the validity and accuracy of feature data, and improving Accuracy of fault classification.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure.

Description of drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description serve to explain the principles of the disclosure. Apparently, the drawings in the following description are only some embodiments of the present disclosure, and those skilled in the art can obtain other drawings according to these drawings without creative efforts.

FIG. 1 schematically shows a flow chart of a fault diagnosis method for mechanical equipment.

Fig. 2 schematically shows a flowchart of determining a fault type according to some embodiments of the present disclosure.

Fig. 3 schematically shows a flow chart of determining a neighbor graph of true neighbors according to some embodiments of the present disclosure.

Fig. 4 schematically shows a flow chart of determining to map different information represented by distance information to different kernel spaces according to some embodiments of the present disclosure.

Fig. 5 schematically shows a schematic diagram of a fault diagnosis device for mechanical equipment that can be applied to an embodiment of the present disclosure.

Fig. 6 schematically shows a structural diagram of a computer system of an electronic device according to some embodiments of the present disclosure.

In the drawings, the same or corresponding reference numerals denote the same or corresponding parts.

Detailed ways

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numerals in different drawings refer to the same or similar elements unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with this specification. Rather, they are merely examples of approaches consistent with some aspects of the specification as recited in the appended claims.

The terms used in this specification are for the purpose of describing particular embodiments only, and are not intended to limit the specification. As used in this specification and the appended claims, the singular forms "a", "the", and "the" are intended to include the plural forms as well, unless the context clearly dictates otherwise. It should also be understood that the term "and/or" as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

It should be understood that although the terms first, second, third, etc. may be used in this specification to describe various information, the information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this specification, first information may also be called second information, and similarly, second information may also be called first information. Depending on the context, the word "if" as used herein may be interpreted as "at" or "when" or "in response to a determination."

In related technologies, there are the following technical problems:

In the existing fault diagnosis technology, in order to ensure the accuracy of diagnosis, machine learning algorithms are usually used to determine the learning parameters, and the parameters are input into the pre-trained fault classifier for classification. Exponential growth, the machine learning algorithm will be affected by hyperparameters, and when the machine learning algorithm determines the learning parameters, data underfitting or overfitting will occur, resulting in data loss and reducing the accuracy of fault classification.

Based on one or more problems in related technologies, the embodiments of the present disclosure firstly propose a fault diagnosis method for mechanical equipment, which can be executed by a calculation model on the mechanical equipment, or by a terminal independent of the mechanical equipment The device or the server executes the fault diagnosis method of the mechanical equipment in this embodiment by taking the server's fault analysis of the vibration signal collected from the mechanical equipment as an example.

As shown in FIG. 1, FIG. 1 is a flow chart of a fault diagnosis method for mechanical equipment according to an exemplary embodiment of the present disclosure, including the following steps:

In step S110, the metric information of the vibration signal of the mechanical equipment is acquired, and a metric matrix is determined through the metric information;

In step S120, constructing a true neighbor graph according to an adaptive neighbor strategy, and determining distance information in the measurement information through the true neighbor graph;

In step S130, at least two kernel spaces are determined according to a preset exponential linear kernel function;

In step S140, different information represented by the distance information is mapped to different nuclear spaces, and a nuclear matrix and nuclear space pheromones corresponding to the nuclear spaces are determined;

In step S150, the metric matrix is reconstructed according to the kernel matrix and the nuclear space pheromone fusion label discrimination information;

In step S160, dimensionality reduction processing is performed on the features of the reconstructed measurement matrix, and the result after dimensionality reduction processing and label discrimination information are input into a pre-trained fault classifier to obtain a fault classification result.

According to the fault diagnosis method in the present disclosure, the characteristic data of the vibration signal of the mechanical equipment can be obtained, and the true neighbor graph can be determined through the characteristic data; at least two kernel spaces can be determined according to the preset exponential linear kernel function; the distance information can be mapped to In the kernel space, determine the kernel matrix and the kernel space pheromone corresponding to the metric matrix and the kernel space; reconstruct the metric matrix according to the fusion label discrimination information of the kernel matrix and the kernel space pheromone. On the one hand, by constructing the true neighbor graph based on the characteristic data of the vibration signal, the true neighbor graph can be determined by screening the information of the vibration signal, which reduces the influence of hyperparameters on the machine learning algorithm; on the other hand, based on the index The linear kernel function determines the kernel space, and fuses the same features in multiple high-dimensional spaces together to reconstruct the metric matrix to reduce data loss and improve the accuracy of fault classification.

Next, step S110 to step S160 will be described in detail.

In step S110, the metric information of the vibration signal of the mechanical equipment is acquired, and a metric matrix is determined through the metric information;

In an exemplary embodiment of the present disclosure, the mechanical equipment vibration signal refers to the vibration generated by the mechanical equipment during operation and the representation of the physical quantity of the vibration, which is used to analyze and evaluate the operating state, performance and health status of the mechanical equipment. Obtaining vibration signals of mechanical equipment can monitor the operating status of mechanical equipment. For example, obtaining the vibration amplitude of mechanical equipment vibration signals can reflect the operating quality and balance of mechanical equipment; obtaining mechanical equipment vibration signals can analyze the vibration frequency components of mechanical equipment, thereby Determine the rotation speed, resonance point and other information of mechanical equipment; of course, by analyzing the frequency spectrum and frequency components of the vibration signal of mechanical equipment, it is also possible to determine whether the equipment has imbalance, bearing failure, gear wear, looseness and other failure problems. In this example embodiment, there is no special limitation on the effect of acquiring vibration signals of mechanical equipment.

Metric information refers to all characteristic data related to vibration signals, which can be used to analyze the working status of mechanical equipment and perform fault diagnosis on mechanical equipment. For example, metric information can be the time domain representing the sampling time and vibration signal over time The data can also be the frequency domain data used to analyze the energy distribution of the vibration signal at different frequencies, of course it can also be the spectrogram used to observe the frequency spectrum characteristics of the vibration signal, or it can be used to describe the vibration characteristics of the vibration signal in the time domain Different metric information can provide information of different levels and angles, and this example embodiment does not specifically limit the type of metric information that characterizes the vibration signal.

Optionally, the metric information can be the information representing the angle and distance between the data points in the vibration signal, the angle information is the angle relationship of the data points in the vibration signal in the high-dimensional Euclidean space, and the distance information is the vibration signal in the high-dimensional Euclidean space The distance of the data points in the vibration signal can determine the relationship between the data points in the vibration signal through the angle information and distance information.

Wherein, the metric matrix refers to a matrix that measures the distance between any two eigenvectors in the feature metric space, and is used to describe the distance data between the eigenvectors in the metric space. The distance between two eigenvectors can be determined according to each element in the metric matrix. For example, for an n-dimensional metric space, the dimension of the metric matrix is n×n, where the element in row i and column j represents the metric space The distance between the i-th and j-th vectors. The diagonal elements of the metric matrix are usually 0, indicating that a vector has a distance of 0 from itself, while the off-diagonal elements indicate the distance between different vectors. Through the metric matrix, we can measure the distance in the vector space to calculate the similarity and difference between the vectors. The distance measurement method can be defined according to the requirement of a specific task, so as to better analyze and model the data, and this example embodiment does not specifically limit the selection of the measurement matrix.

In step S120, the true neighbor neighbor graph is constructed according to the adaptive neighbor strategy, and the distance information in the measurement information is determined through the true neighbor neighbor graph.

In an exemplary embodiment of the present disclosure, the true neighbor point neighbor graph refers to a graph with a certain angle and distance between the sample point and the true neighbor point, which is used to map part of the characteristic data in the high-dimensional Euclidean space of the vibration signal to the low-dimensional dimensional space, the true neighbor graph can be a true neighbor graph with a fixed number of true neighbors, or a true neighbor graph that dynamically changes according to the number of true neighbors. This example embodiment is for true neighbors The specific state of the point neighbor graph is not particularly limited.

For example, the angle information and distance information of the vibration signal sample points can be selected to construct a dynamic true neighbor neighbor map. The true neighbor neighbor map can combine the distance information and angle information of sample points and true neighbor points in high-dimensional Euclidean space Dimensionality reduction is expressed in a low-dimensional space, and the angle information and distance information between the sample point and the true neighbor point obtained through the true neighbor point neighbor graph and the angle information between the sample point and the true neighbor point in the high-dimensional space and The distance information is consistent, which reduces the difficulty of obtaining the angle information and distance information between the sample point and the true neighbor point in the high-dimensional space, and maintains the stability of the relationship between the sample point and the true neighbor point.

The distance information in the measurement information of the true neighbor point neighbor map is the geodesic distance information between the sample point and the true neighbor point. The geodesic distance information is used to calculate the sample point and the true neighbor point in the high-dimensional space of the vibration signal. The distance between and is obtained by calculating the length of the shortest path between the sample point and the true neighbor point. The geodesic distance information can be the geodesic distance on the sphere or the geodesic distance on the two-dimensional plane. When the geodesic distance information is the geodesic distance on the two-dimensional plane, it is represented by the Euclidean distance Geodesic distance; of course, the geodesic distance information may also be a geodesic distance on a curved surface, and this example embodiment does not specifically limit the plane where the geodesic distance information is located.

In step S130, at least two kernel spaces are determined according to a preset exponential linear kernel function.

In an exemplary embodiment of the present disclosure, the exponential linear kernel function refers to a kernel function used in a machine learning algorithm constructed based on the exponential linear kernel function theory, and is used for mapping high-dimensional data of a vibration signal into a kernel space. The exponential linear kernel function determines the kernel space by selecting appropriate parameters. For example, common values can be selected as the parameters for constructing the kernel space based on previous experience of similar problems, or the best parameters can be found by performing grid search within a certain range. , of course, the Bayesian optimization algorithm can also be used to adjust the parameters according to the previous result in each iteration. This example embodiment does not discuss the method of selecting parameters when constructing the kernel space of the exponential linear kernel function and the construction method of the kernel space. special limited.

Wherein, at least two kernel spaces are determined according to the preset exponential linear kernel function in order to store different characteristic data in the vibration signal, for example, two kernel spaces may be determined according to the exponential linear kernel function, respectively To store two different feature data, three or more kernel spaces can also be determined according to the exponential linear kernel function, and three or more feature data are stored in three or more kernel spaces, and the exponential linear The number of kernel spaces determined by the kernel function is consistent with the number of types of feature data to be stored, and this example embodiment does not specifically limit the number of kernel functions determined by the exponential linear kernel function.

In step S140, different information represented by the distance information is mapped to different nuclear spaces, and a nuclear matrix and nuclear space pheromones corresponding to the nuclear spaces are determined;

In an exemplary embodiment of the present disclosure, the kernel matrix refers to a matrix constructed according to the exponential linear kernel function theory, which is the same as the exponential linear kernel function matrix, and is used to represent the geodesic distance information of the vibration signal. The kernel matrix can be used for different kernels The fusion of feature information in the space, for example, can be done by setting the linear weight in the exponential linear kernel function theory as the relevant data in the kernel space, and fusing the feature information of the kernel space into the kernel matrix. Fusion of feature information from different spaces.

The nuclear space pheromone refers to the data representing the characteristics in the corresponding nuclear space, which is used for the reconstruction of the measurement matrix and the construction of the nuclear matrix. The number of nuclear space pheromones that can be determined is usually the same as the number of nuclear spaces. For example, when the nuclear space When there are two spaces, two nuclear space pheromones can be determined according to the nuclear space; when there are three or more nuclear spaces, three or more nuclear spaces can be determined through three or more nuclear spaces For pheromones, the number of nuclear space pheromones in this example embodiment is not particularly limited, and generally the number of nuclear space pheromones that can be determined is the same as the number of nuclear spaces.

In step S150, the metric matrix is reconstructed according to the kernel matrix and the nuclear space pheromone fusion label discrimination information;

In an exemplary embodiment of the present disclosure, label discrimination information refers to accurately classifying data points into different categories or labels defined in advance by analyzing the characteristics and attributes of data points in vibration signals, wherein the fusion label The discriminant information reconstructs the metric matrix, and some data points in the metric matrix that cannot be distinguished into different categories can be fused with the label discriminant information. When reconstructing the metric matrix, the differences between different categories are considered to improve the discriminative ability of the metric matrix. Kernel and classification performance.

The reconstruction of the metric matrix through the fusion of the kernel matrix and the nuclear space pheromone label discriminant information can enhance the expressive ability of the features in the original data, and the reconstructed metric matrix can better capture the key features of the input data, thereby improving the subsequent processing tasks. accuracy and performance; in addition, the reconstruction process can make the measurement matrix more focused on features with higher discrimination, reduce the influence of redundant and useless information, help to improve the selectivity of data, and can better distinguish different The difference between categories can reduce the variability within the category; at the same time, by fusing the label discriminant information, the reconstructed metric matrix can better combine the semantic information of the data and the category relationship, which is helpful in the reconstructed metric matrix. Retain more differences between categories to improve the performance of classification or recognition tasks; by reconstructing the measurement matrix, the expressiveness, selectivity and information integration of data features can be improved, thereby providing better input for fault classification and obtaining more accurate results and higher performance.

In step S160, dimensionality reduction processing is performed on the features of the reconstructed measurement matrix, and the result after dimensionality reduction processing and label discrimination information are input into a pre-trained fault classifier to obtain a fault classification result.

In an example embodiment of the present disclosure, the dimensionality reduction process refers to transforming and compressing the features in the reconstructed metric matrix to reduce the dimension of the features, and the reconstructed metric matrix can be linearly or nonlinearly Dimensionality reduction processing of the measurement matrix, for example, the dimensionality reduction processing of the reconstructed measurement matrix can be performed by the nonlinear processing method of eigenvalue decomposition, or the linear processing method of linear discriminant analysis (LDA) can be used to reduce the dimensionality of the reconstructed measurement matrix Dimensionality reduction processing is performed on the matrix. Of course, other methods may also be used to perform dimensionality reduction processing on the reconstructed measurement matrix. This example embodiment does not specifically limit the dimensionality reduction method of the reconstructed measurement matrix.

The pre-trained fault classifier refers to a classifier that learns fault patterns and characteristics through training on large-scale fault data, so that it can classify faults on new input data, and is used to classify input feature vectors and labels. The pre-trained fault classifier can be a variety of machine learning models and deep learning models, for example, it can be the SVM classifier of the support vector machine, or it can be a classifier that supports random forests. Of course, the pre-trained fault classifier can be It is selected according to actual requirements, and this example embodiment does not specifically limit the types of pre-trained fault classifiers.

By determining the true neighbor graph based on the angle information and distance information of the vibration signal, and using the shortest path algorithm to calculate the true neighbor graph, the geodesic distance matrix and the metric information represented by the geodesic distance are obtained. Through the true neighbor graph Determining the metric information represented by the geodesic distance matrix and the geodesic distance matrix can reduce the influence of hyperparameters on the machine algorithm model; and by mapping the metric information expressed by the geodesic distance matrix and the geodesic distance matrix to different kernel spaces, and then The kernel matrix determined in the mapping process and the nuclear space pheromone are fused to distinguish labels, and the metric space is reconstructed according to the exponential linear kernel function theory, and the reconstructed metric matrix is further used to determine another nuclear space pheromone. Spatial pheromones construct a kernel matrix, and perform dimensionality reduction processing through the feature information represented in the kernel matrix, and finally input the result of dimensionality reduction processing and label discrimination information into the pre-trained machine classifier to obtain classification results, which can reduce The data is lost during the extraction process, thereby improving the accuracy of the classifier for fault diagnosis.

The technical solutions involved in steps S110 to S160 are explained in detail below.

In an exemplary embodiment of the present disclosure, the determination of the neighbor map of true neighbors in step S110 can be realized by the following steps with reference to FIG. 3 :

Step S310, obtain the sample point and the initial neighbor point of the mechanical equipment vibration signal, the initial neighbor point is determined by the neighbor value set by the sample point; Step S320, determine the number of true neighbor points and the initial neighbor point according to the distance between the sample point and the initial neighbor point Number of points; step S330 , constructing a true neighbor point neighbor graph according to the true neighbor point number and the initial neighbor point number.

Among them, the sample points of vibration signals of mechanical equipment refer to a group of data points selected from the vibration signals. extraction and fault detection and diagnosis. A sufficient number of data points can be selected for the sample point. For example, the sample point can be selected to fully represent the characteristics of the entire vibration signal and uniformly distributed in the time or frequency domain. The sample point can also be selected according to the vibration signal processing method and application scenario. Selection, this example embodiment does not specifically limit the selection of sample points.

The initial neighbor point refers to the data point with a certain distance from the sample point. It is used to select the appropriate data point near the sample point to construct the initial neighbor graph. The initial neighbor point can be determined by preset neighbor value K, or through the grid Determining the initial neighbor points by dividing, of course, by setting a fixed range, the data points within the fixed range from the sample point can be used as the initial neighbor points. The selection method of the initial neighbor points can be selected according to the requirements. The selection of the neighbor points is not particularly limited.

Optionally, the initial neighbor point is determined by a preset neighbor value K, wherein the neighbor point of each sample point is determined by the preset neighbor value K and defined as the initial neighbor point. For example, if the neighbor value K is set to 5, then the shortest distance between the data and the sample point is calculated according to the shortest path algorithm, and 5 data points that are closer to the sample point are selected to form the neighbor point, that is, the initial neighbor point.

Optionally, when constructing the neighbor map of true neighbor points, it is necessary to preprocess the initial neighbor points to obtain the Euclidean distance matrix, and normalize the Euclidean distance matrix to eliminate the difference between different sample points. The difference on the distance scale ensures that the distance between each sample point is comparable, thereby improving the performance and accuracy of these algorithms and reducing the underfitting of the data; and in some cases, different features have a significant impact on the distance calculation. The contribution of may be unbalanced. By normalizing the distance matrix, the weights between different features can be balanced, and certain features can be prevented from having too much influence on the calculation of Euclidean distance.

Among them, the Euclidean distance matrix refers to the matrix that measures the straight-line distance between the initial neighbor point and the sample point, and is used to represent the Euclidean distance between the initial neighbor point and the sample point. For data points in high-dimensional space , the calculation method of the Euclidean distance matrix is similar, and the calculation of the square and root of the coordinate difference is applied to all dimensions. For example, the initial neighbor graph G can be constructed based on the determined initial neighbor points; then the initial neighbor cosine similarity matrix can be calculated , where: /> , /> Represents the k-th dimension feature of the i-th sample point, Represents the initial neighbor point set of the i-th sample point; then, construct the neighbor cosine similarity adjustment matrix/> , where /> , and finally calculate the Euclidean distance matrix between the initial neighbor point and the center point , where /> .

Normalization processing refers to scaling the data so that the processed data is scaled within a specific range or relative size. In data processing and machine learning, normalization usually refers to converting data between 0 and 1 The range of , also known as Min-Max Scaling, is used to eliminate differences between features. Of course, normalization processing can also be performed by other methods, for example, normalization processing can be performed by Z-score normalization method, or normalization processing can be performed by normal distribution. The specific method is not particularly limited.

For example, the initial neighbor Euclidean distance matrix of all sample points is normalized by the minimum-maximum scaling method to obtain the normalized Euclidean distance matrix of N nearest neighbors :in , which converts the data in the Euclidean distance matrix to a range between 0 and 1.

The true neighbor point refers to the sample point whose weight value is less than or equal to the average weight value of the sample point obtained by comparing each weight value of the sample point with the average weight value of the sample point, and is used to determine the number of true neighbor points. According to the number of true neighbors, the neighbor graph of true neighbors is constructed. For example, it is possible to construct a weight matrix between all sample points and their initial neighbors: , where /> ; Then calculate the average weight of all sample points and their initial neighbor point set /> , where /> ; Then each weight of the sample point x _i /> and the average weight of the sample point /> for comparison, if /> Then it is considered that x _j is the pseudo neighbor point of x _i , if /> Then it is considered that x _j is the true neighbor point of x _i ; according to the number of true neighbor points corresponding to each sample point, dynamically adjust the number of neighbor points of the sample point, obtain the dynamic true neighbor value K ⁺ , and construct the dynamic true neighbor point neighbor Figure G ⁺ .

In an exemplary embodiment of the present disclosure, the following steps may be implemented to map different information represented by the distance information in step S140 into different kernel spaces:

Calculate the similarity and inner product between the distance information, and map different information represented by the distance information into different kernel spaces according to the similarity and inner product.

Optionally, there are many ways to map the distance information to different kernel spaces. For example, the metric information represented by the geodesic distance can be mapped to the kernel space through the method of dual centering; the geodesic distance can also be mapped to the kernel space through the kernel function mapping method. The distance matrix is mapped to another kernel space; of course, the distance information can also be mapped to the kernel space by means of eigenvalue extraction, and the method of mapping the distance information to the kernel space is not particularly limited in this example embodiment.

For example, the exponential linear kernel function can be determined based on the exponential linear kernel function theory, where the exponential linear kernel function is a kernel function used to map the data in the input space to a high-dimensional feature space, which can measure two samples The similarity between points, and the kernel matrix determined by the exponential linear kernel function has a positive semi-definite property. The semi-positive definite property of the kernel matrix can ensure the stability and correctness of the machine learning algorithm. The kernel matrix determined by the exponential linear kernel function has The positive semi-definite property can be proved by using the geodesic distance information and the necessary and sufficient conditions of the kernel function. For example, the exponential linear kernel function matrix is determined according to the geodesic distance information and the exponential linear kernel function theory , where /> is the linear weight, b (0<b≤1) is the offset coefficient, /> is the geodesic distance between samples x _i and x _j ; and according to the necessary and sufficient conditions of the kernel function: for any data sample , and its corresponding Gram matrix /> is a positive semi-definite matrix, and N is the number of samples. The proof of the positive semi-definite property of the exponential linear kernel function matrix is as follows:

The structure of the exponential linear kernel function shows that it utilizes the information of the geodesic matrix, then the matrix is a symmetric matrix, so it is only necessary to prove that /> is a positive semidefinite matrix; derive the mapping for exponential linear functions: If the sample x _j is the neighbor point of x _i , then there is /> ,but which maps /> , it can be seen that when the sample x _j is the neighbor point of x _i , the exponential linear function satisfies the necessary and sufficient conditions of the kernel function.

If the sample x _j is not the neighbor point of x _i , then there is , /> is the sum of finite terms d ^E , so let /> There are z items d ^E and /> , then there is: /> ,

It can be seen that when the sample x _j is not the neighbor point of x _i , the exponential linear function also satisfies the necessary and sufficient conditions of the kernel function.

According to the above proof results, we know that the symmetric matrix It has a positive semi-definite property and satisfies the necessary and sufficient conditions of the kernel function, so the kernel matrix determined by the exponential linear kernel function has a positive semi-definite property.

Among them, using the exponential linear kernel function to calculate the similarity between distance information can be to use the exponential linear kernel function to calculate the similarity between two sample points x and y, which can be calculated by calculating the Euclidean distance between the sample points , and then implement the Euclidean distance as the exponent of the exponential linear kernel function, so that a smaller distance produces a larger similarity value, and a larger distance produces a smaller similarity value; using the exponential linear kernel function to calculate The inner product between distance information can be calculated using the definition of exponential linear kernel function.

By calculating the similarity and inner product between distance information, the similarity value between sample points and the inner product value in the kernel space can be obtained. These similarity values and the inner product value of the kernel space can be used for feature extraction, and the distance information of the sample points is mapped to the kernel space. Calculate the similarity and inner product between distance information through exponential linear kernel function, and map different distance information into different kernel spaces, which provides an effective way to deal with non-linearly separable sample point data, and Play an important role in pre-training machine learning tasks.

In an exemplary embodiment of the present disclosure, the mapping of the metric information represented by the geodesic distance into the first kernel space may be implemented by referring to FIG. 4 through the following steps:

Step S410, determine the geodesic distance matrix and the square geodesic distance matrix according to the true neighbor point neighbor graph; step S420, obtain the preset fixed kernel matrix; The two sides are respectively multiplied by the fixed kernel matrix, and the metric information represented by the geodesic distance is mapped into the first kernel space.

Among them, the geodesic distance matrix refers to a distance matrix determined based on the true nearest neighbor graph, which is used to measure the similarity and difference between data points. The geodesic distance matrix can be determined using the shortest path algorithm based on the true nearest neighbor graph, for example, the Dijkstra algorithm or the Floyd algorithm can be used to calculate the geodesic distance between each data point in the true neighbor graph, and then according to the determined geodesic distance The geodesic distance matrix is determined by the geodesic distance, and an appropriate algorithm may be selected according to requirements to determine the geodesic distance matrix. In this exemplary embodiment, the algorithm for determining the geodesic distance matrix is not particularly limited.

The metric information represented by the geodesic distance refers to the geodesic distance represented in the geodesic distance matrix, which is used to measure the distance of the shortest path between data points. The metric information represented by the geodesic distance is the geodesic distance calculated by the geodesic distance matrix according to the shortest path algorithm, and the method for determining the metric information represented by the measurement matrix will not be repeated in this example embodiment.

The fixed kernel matrix refers to the matrix used to nonlinearly map the data and map the data into the kernel space. The fixed kernel matrix maps the data into the kernel space by multiplying both sides of the square geodesic distance matrix with the fixed kernel matrix, which can be determined by a custom kernel function, for example, according to the exponential linear The kernel function selects the appropriate kernel width parameter σ, calculates the kernel values between each pair of samples and composes them into a kernel matrix, thus determining the fixed kernel matrix. Certainly, the fixed kernel matrix may also be determined according to other kernel functions, such as a Gaussian kernel or a polynomial kernel, and the determination of the kernel matrix is not particularly limited in this example embodiment.

The first kernel space kernel matrix refers to the matrix representing the geodesic distance calculated according to the geodesic distance matrix and the square geodesic distance matrix, which is used to map the metric information represented in the geodesic distance matrix to the first kernel space . Among them, the kernel matrix of the first kernel space can be determined by the geodesic distance matrix. For example, the true neighbor point neighbor graph G ⁺ can be obtained through the adaptive neighbor strategy, and the geodesic distance matrix D _G and the squared geodesic distance matrix can be obtained by calculation , while computing the matrix operator /> That is, the first kernel space kernel matrix.

Mapping the metric information expressed by the geodesic distance into the first kernel space can calculate the square geodesic distance matrix according to the geodesic distance matrix D _G , where the square geodesic matrix is Then multiply the left and right sides of the square geodesic distance matrix with the H matrix by the method of double centering, that is, /> , through the operation between the matrices, the metric information represented by the geodesic matrix is mapped to the first kernel space. The H matrix is a known and fixed kernel matrix in the space mapping process, which will not be described in detail in this disclosure.

In an example embodiment of the present disclosure, the mapping of the geodesic distance matrix into the second kernel space may be implemented by referring to FIG. 4 through the following steps:

Step S440, setting the elements in the geodesic distance matrix as the geodesic distance information of the elements of the exponential linear kernel function matrix, and mapping the geodesic distance matrix into the second kernel space.

Optionally, the geodesic distance matrix is mapped to the kernel space by an exponential linear kernel function, wherein the exponential linear kernel function is a kind of kernel function and has a positive semi-definite property. The semi-positive definite property of the exponential linear kernel function can ensure that the data points in the stability of the mapping process. The geodesic distance matrix can be obtained by first calculating, and then each element in the geodesic distance matrix is mapped to the exponential linear kernel function, and then the second kernel space is determined according to the exponential linear kernel function, that is, each element in the geodesic distance matrix Elements are mapped into the second kernel space.

For example, the exponential linear kernel function matrix is ,and , the elements in the geodesic distance matrix D _G /> Substitute into the elements in the exponential linear kernel function matrix, so that the geodesic distance matrix D _G is mapped to the second kernel space through the exponential linear kernel function, and the feature extraction is completed, and the second kernel space kernel matrix is obtained, that is, .

Non-linear mapping according to the exponential linear kernel function can capture the nonlinear relationship between samples, and perform more effective feature representation and learning in the second kernel space, which can reduce the dimension of features while retaining key information, and improve The expressive ability of features helps to better perform machine learning tasks.

In an exemplary embodiment of the present disclosure, the following steps may be used to determine the feature representation in the first kernel space through the first kernel space pheromone:

Determine the first kernel space matrix according to the metric information represented by the geodesic distance, then set the spectral radius of the first kernel space matrix according to the first kernel space matrix, and determine the first kernel space pheromone through the spectrum radius of the first kernel space matrix, and then according to The first kernel space pheromone determines a feature representation in the first kernel space.

The first nuclear space pheromone refers to the data representing the characteristics in the first nuclear space, which is used to reconstruct the metric matrix. The first nuclear space pheromone is determined by the first nuclear space kernel matrix. For example, it can be obtained through the first kernel Hilbert space kernel matrix Construct block matrix /> and calculate the first kernel space matrix spectral radius /> , and let the first kernel Hilbert space pheromone /> , to determine the first nuclear spatial pheromone.

Determining the feature representation in the first kernel space by using the first kernel space pheromone can simplify the way of operating the features in the first kernel space in subsequent operations, thereby ensuring data accuracy while improving operation efficiency.

In an exemplary embodiment of the present disclosure, the reconstruction of the metric matrix in step 150 according to the fusion label discrimination information of the kernel matrix and the kernel space pheromone can be realized through the following steps:

The measurement matrix is reconstructed by performing nonlinear operation on the measurement matrix through the kernel matrix and the nuclear space pheromone.

Among them, the reconstruction of the metric matrix can adopt the method of linear correction or non-linear correction. The method of linear correction can be to reconstruct the metric matrix through linear transformation, for example, principal component analysis (PCA) and linear Discriminant analysis (LDA); the method of using nonlinear correction can be to use a nonlinear function to reconstruct the measurement matrix, such as an exponential linear kernel function, using an exponential linear kernel function to reconstruct the measurement matrix, and the extracted features can be mapped to a In the high-dimensional feature space, and then perform linear correction in the high-dimensional feature space; of course, the nonlinear correction method of popular learning can also be used to reconstruct the metric matrix by using the local geometric structure of the sample, which can better retain the data. Manifold structure, the method for reconstructing the metric matrix is not particularly limited in this example embodiment.

Reconstructing the measurement matrix can improve the representation ability of data, reduce dimensionality, enhance classification performance, facilitate data visualization and association analysis, etc., which is of great help to machine learning tasks.

In an exemplary embodiment of the present disclosure, the non-linear correction to the metric matrix may be implemented through the following steps, so as to reconstruct the metric matrix:

The reconstructed metric matrix is determined by performing nonlinear operation on the metric matrix and integrating the label discriminant information, the first kernel space pheromone, the second kernel space kernel matrix and the exponential linear kernel function theory to modify the metric matrix.

Reconstructing the metric matrix can perform nonlinear operations on the metric matrix through the first kernel space pheromone, label discrimination information, second kernel space kernel matrix and exponential linear kernel function theory, and through the partial matrix obtained in the feature extraction process or The data changes the parameters in the measurement matrix to perform nonlinear correction on the measurement matrix, so as to realize the reconstruction of the measurement matrix. For example, where the expression of the metric matrix is: , the metric matrix is nonlinearly corrected by fusing the label discriminant information, the second kernel Hilbert space kernel matrix and the exponential linear kernel function theory: where: /> is the offset coefficient.

In an exemplary embodiment of the present disclosure, the fusion of the first nuclear space pheromone into the second nuclear space kernel matrix can be achieved through the following steps:

Wherein, the linear weight refers to a parameter of the exponential linear kernel function matrix determined based on the exponential linear kernel function theory, and is used to construct the exponential linear kernel function matrix. The expression of the exponential linear kernel function matrix is: , each item of data in the exponential linear kernel function matrix is It is a linear weight, and the features in the exponential linear kernel function matrix can be changed by changing the value of the linear weight.

Optionally, let the linear weights in exponential linear kernel function theory , the first kernel Hilbert space pheromone can be fused into the exponential linear kernel function matrix, and the second kernel space kernel matrix is determined by the exponential linear kernel function matrix, so by setting the linear weight in the exponential linear kernel function to the first kernel Spatial pheromones can fuse the features of the first kernel space into the kernel matrix of the second kernel space to complete the fusion of features in different high-dimensional spaces.

In an exemplary embodiment of the present disclosure, the Mercer kernel matrix can be constructed through the following steps:

According to the reconstructed metric matrix, determine the spectral radius of the second nuclear space matrix, and determine the second nuclear space pheromone according to the spectral radius of the second nuclear space matrix, and then construct the Mercer kernel according to the reconstructed metric matrix and the second nuclear space pheromone matrix.

The second nuclear spatial pheromone refers to the data representing the features in the reconstructed metric matrix, which is used to construct the Mercer kernel matrix, and the second nuclear spatial pheromone is determined by the reconstructed metric matrix. For example, using the metric matrix The linear correction matrix again constructs the block matrix , and calculate the second kernel space matrix spectral radius, /> , let the second kernel Hilbert space pheromone /> .

The Mercer kernel matrix refers to a kernel matrix with symmetric positive definite properties, which is used to facilitate dimensionality reduction of the features in the reconstructed metric matrix. The Mercer kernel matrix can be used for various pre-training machine learning tasks. For example, the Mercer kernel matrix can support vector machine (SVM) classifier learning, and can also support kernel principal component analysis (Kernel PCA). This example embodiment is used for the Mercer kernel matrix The pre-training machine for learning is not particularly limited.

The Mercer kernel matrix can be based on the second kernel Hilbert space pheromone , and the reconstructed squared metric matrix /> And the corresponding kernel space mapped by the reconstructed metric matrix by the dual center method To construct, the specific construction method is: /> .

By using the reconstructed metric matrix to determine the second kernel spatial pheromone and constructing the Mercer kernel matrix,

It can avoid explicit calculation in high-dimensional feature space, thereby reducing computational complexity and improving calculation accuracy.

In an exemplary embodiment of the present disclosure, the following steps can be used to perform dimensionality reduction processing on the features of the reconstructed metric matrix, and input the processing result and label discrimination information into the pre-trained fault classifier to obtain the fault classification result :

The features in the Mercer kernel matrix are decomposed into eigenvalues to obtain the covariance matrix of the Mercer kernel matrix, and then the corresponding eigenvalues and eigenvectors in the Mercer kernel matrix are determined according to the covariance matrix, and the Mercer kernel matrix is calculated according to the eigenvalues and eigenvectors Dimensionality reduction is performed on the features of the Mercer kernel matrix to obtain the low-dimensional space embedding vector corresponding to the Mercer kernel matrix, and then the low-dimensional space embedding vector and label discrimination information are input into the pre-trained fault classifier to obtain the fault classification result.

Among them, the method of performing eigenvalue decomposition of the features in the Mercer kernel matrix is an existing technology, and will not be repeated here, and can be obtained by the eigenvalue decomposition method The low-dimensional space embedding coordinates of /> , to complete the dimensionality reduction process of feature extraction.

By reducing the dimensionality of the features of the Mercer kernel matrix, and inputting the low-dimensional space embedding vector and label discrimination information obtained by reducing the dimensionality of the Mercer kernel matrix into the pre-trained fault classifier, the fault classification result is finally obtained, which can help technicians Find the cause of the fault quickly and accurately, and take corresponding measures in time to solve the problem, and the fault diagnosis through the pre-trained fault classifier can improve the efficiency and accuracy of fault diagnosis.

Fig. 2 schematically shows a flow chart of performing fault diagnosis on mechanical equipment according to some embodiments of the present disclosure.

Referring to Fig. 2, the fault diagnosis of mechanical equipment can be realized through steps S201 to S207, wherein:

Step S201, acquiring angle information and distance information of the vibration signal;

Step S202, according to the angle information and the distance information to obtain the neighbor map of the true neighbor points; wherein, the initial neighbor point of the sample point is determined by setting the initial neighbor value, and the weight and the average weight of the distance between the sample point and the initial neighbor point Compare to get true neighbors and false neighbors, and build a true neighbor graph based on the real neighbors around the determined sample points, and the true neighbor graph can be dynamically changed according to the number of true neighbors.

Step S203, determine the geodesic distance matrix according to the true neighbor point neighbor map; wherein, according to the true neighbor point neighbor map to determine the geodesic distance matrix is obtained by the shortest path algorithm, the shortest path of the elements in the true neighbor point neighbor map can be calculated to determine the measure elements of the geodesic distance, thereby determining the geodesic distance matrix.

Step S204, map the geodesic distance matrix and the metric information represented by the geodesic distance into the kernel space respectively; wherein, map the metric information represented by the geodesic distance matrix into the first kernel space through the method of dual centralization, and use the index The linear kernel function maps the geodesic distance matrix into the second kernel space.

Step S205, according to the kernel matrix and kernel space pheromone determined in the feature mapping process, and fusing label discrimination information and exponential linear kernel function theory to reconstruct the measurement matrix; wherein the kernel matrix and kernel space determined in the feature mapping process The pheromones are the second kernel space kernel matrix and the first kernel space pheromone respectively, the second kernel space kernel matrix is constructed according to the exponential linear kernel function, by changing the linear weight of the second kernel space matrix in the determination process, the The pheromone of the first kernel space is fused into the kernel matrix of the second kernel space, so as to realize the fusion of the information of the first kernel space and the second kernel space, and perform nonlinear operation on the metric matrix by fusing the label discriminant information, so as to obtain the reconstructed metric matrix.

In step S206, the features of the reconstructed metric matrix are represented by nuclear space pheromones, and a Mercer kernel matrix is constructed; wherein, the features of the reconstructed metric matrix are represented by the second nuclear space pheromones, and the first nuclear space The pheromone is integrated into the second kernel space kernel matrix, so the second kernel space pheromone can characterize the features in the reconstructed metric matrix.

Step S207, perform dimensionality reduction processing on the features in the Mercer kernel matrix, and input the obtained dimensionality reduction processing results and label discrimination information into the pre-trained classifier to obtain fault classification results; wherein, the dimensionality reduction processing results are low dimensional space embedding vector, by inputting the low-dimensional space vector and label discriminant information into the pre-trained fault classifier, the fault classification result is obtained.

Next, a fault diagnosis device for machinery equipment according to an exemplary embodiment of the present disclosure will be described with reference to FIG. 5 .

As shown in Figure 5, the fault diagnosis device 500 for mechanical equipment may include a metric information acquisition module 510, a true neighbor point neighbor graph construction module 520, a kernel space construction module 530 and a feature extraction module 540, a metric matrix reconstruction module 550, Dimensionality reduction module 560 .

The measurement information acquisition module is used to obtain the measurement information of the vibration signal of the mechanical equipment, and determine the measurement matrix through the measurement information;

A true neighbor neighbor graph building module, used to construct a true neighbor neighbor graph according to an adaptive neighbor strategy, and determine distance information in the measurement information through the true neighbor neighbor graph;

A kernel space construction module, configured to determine at least two kernel spaces according to a preset exponential linear kernel function;

The feature extraction module is used to map the different information represented by the distance information to different nuclear spaces for feature extraction, and determine the nuclear matrix and nuclear space pheromone corresponding to the nuclear space;

The metric matrix reconstruction module is used to reconstruct the metric matrix according to the kernel matrix and nuclear space pheromone fusion label discrimination information;

The dimensionality reduction module is used to perform dimensionality reduction processing on the features of the reconstructed measurement matrix, and input the dimensionality reduction processing results and label discrimination information into the pre-trained fault classifier to obtain fault classification results.

Obtain the metric information of the vibration signal to determine the metric matrix, and construct a true neighbor graph according to the angle information and distance information in the metric information, and determine the geodesic distance between the data points of the vibration signal in the high-dimensional space through the true neighbor graph Matrix and the metric information represented by the geodesic distance, and determine the kernel space through the kernel function, map the metric information represented by the geodesic distance into the first kernel space, and map the geodesic distance matrix into the second kernel space, and, by The second kernel space kernel matrix and the first kernel space pheromone determined in the process of mapping the metric information to the kernel space, and the metric matrix is reconstructed by fusing the label discriminant information and the exponential linear kernel function theory, so as to realize the information in the multi-space , and determine the feature representation in the reconstructed metric matrix through the second nuclear spatial pheromone, use the square of the reconstructed metric matrix and the second nuclear spatial pheromone to construct the Mercer kernel matrix, and use the eigenvalues according to the Mercer kernel matrix The decomposition method obtains the low-dimensional space embedding vector corresponding to the features in the Mercer kernel matrix, and finally inputs the low-dimensional space embedding vector and label discrimination information into the pre-trained classifier to obtain the fault classification result.

It should be noted that although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that these steps must be performed in this specific order, or that all shown steps must be performed to achieve achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step for execution, and/or one step may be decomposed into multiple steps for execution, etc.

In addition, in an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above fault diagnosis method is also provided.

Those skilled in the art can understand that various aspects of the present disclosure can be implemented as a system, method or program product. Therefore, various aspects of the present disclosure can be embodied in the following forms, namely: a complete hardware embodiment, a complete software embodiment (including firmware, microcode, etc.), or an embodiment combining hardware and software aspects, which may be collectively referred to herein as "circuit", "module" or "system".

An electronic device 600 according to such an embodiment of the present disclosure is described below with reference to FIG. 6 . The electronic device 600 shown in FIG. 6 is only an example, and should not limit the functions and application scope of the embodiments of the present disclosure.

As shown in FIG. 6, electronic device 600 takes the form of a general-purpose computing device. Components of the electronic device 600 may include, but are not limited to: at least one processing unit 610 , at least one storage unit 620 , a bus 630 connecting different system components (including the storage unit 620 and the processing unit 610 ), and a display unit 640 .

Wherein, the storage unit stores program codes, and the program codes can be executed by the processing unit 610, so that the processing unit 610 executes various exemplary methods according to the present disclosure described in the “Exemplary Methods” section above in this specification. Example steps. For example, the processing unit 610 may execute step S110 as shown in FIG. 1 to obtain the metric information of the mechanical equipment vibration signal, and determine the metric matrix through the metric information; step S120, construct the true neighbor point according to the adaptive neighbor strategy Neighbor graph, and determine the distance information in the metric information through the true neighbor point neighbor graph; Step S130, determine at least two kernel spaces according to the preset exponential linear kernel function; Step S140, map different information represented by the distance information to different In the nuclear space, determine the nuclear matrix and nuclear space pheromone corresponding to the nuclear space; step S150, reconstruct the measurement matrix according to the fusion label discrimination information of the nuclear matrix and nuclear space pheromone; step S160, reconstruct the measurement matrix Dimensionality reduction processing is carried out on the features, and the result of dimensionality reduction processing and label discrimination information are input into the pre-trained fault classifier to obtain the fault classification result.

The storage unit 620 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 621 and/or a cache storage unit 622 , and may further include a read-only storage unit (ROM) 623 .

The storage unit 620 may also include a program/utility 624 having a set (at least one) of program modules 625, such program modules 625 including but not limited to: an operating system, one or more application programs, other program modules, and program data, Implementations of networked environments may be included in each or some combination of these examples.

Bus 630 may represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local area using any of a variety of bus structures. bus.

The electronic device 600 can also communicate with one or more external devices 670 (such as keyboards, pointing devices, Bluetooth devices, etc.), and can also communicate with one or more devices that enable the user to interact with the electronic device 600, and/or communicate with Any device (eg, router, modem, etc.) that enables the electronic device 600 to communicate with one or more other computing devices. Such communication may occur through input/output (I/O) interface 650 . Moreover, the electronic device 600 can also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN) and/or a public network such as the Internet) through the network adapter 660 . As shown, the network adapter 660 communicates with other modules of the electronic device 600 through the bus 630 . It should be appreciated that although not shown, other hardware and/or software modules may be used in conjunction with electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives And data backup storage system, etc.

Through the description of the above embodiments, those skilled in the art can easily understand that the exemplary embodiments described here can be implemented by software, or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of the present disclosure can be embodied in the form of software products, and the software products can be stored in a non-volatile storage medium (which can be CD-ROM, U disk, mobile hard disk, etc.) or on the network , including several instructions to make a computing device (which may be a personal computer, a server, a terminal device, or a network device, etc.) execute the method according to the embodiment of the present disclosure.

In an exemplary embodiment of the present disclosure, there is also provided a computer-readable storage medium on which a program product capable of implementing the above-mentioned method in this specification is stored. In some possible embodiments, various aspects of the present disclosure may also be implemented in the form of a program product, which includes program code, and when the program product is run on a terminal device, the program code is used to make the The terminal device executes the steps according to various exemplary embodiments of the present disclosure described in the "Exemplary Method" section above in this specification.

According to the embodiment of the present disclosure, the program product for implementing the above-mentioned fault diagnosis method for mechanical equipment may adopt a portable compact disk read-only memory (CD-ROM) and include program codes, and may be used in terminal equipment such as personal run on the computer. However, the program product of the present disclosure is not limited thereto. In this document, a readable storage medium may be any tangible medium containing or storing a program, and the program may be used by or in combination with an instruction execution system, apparatus or device.

The program product may reside on any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: electrical connection with one or more conductors, portable disk, hard disk, random access memory (RAM), read only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

A computer readable signal medium may include a data signal carrying readable program code in baseband or as part of a carrier wave. Such propagated data signals may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A readable signal medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

Program code embodied on a readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages—such as Java, C++, etc., as well as conventional procedural Programming language - such as "C" or a similar programming language. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server to execute. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it may be connected to an external computing device (for example, using an Internet service provider). business to connect via the Internet).

In addition, the above-mentioned drawings are only schematic illustrations of processes included in the method according to the exemplary embodiments of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above figures do not imply or limit the chronological order of these processes. In addition, it is also easy to understand that these processes may be executed synchronously or asynchronously in multiple modules, for example.

Through the description of the above embodiments, those skilled in the art can easily understand that the exemplary embodiments described here can be implemented by software, or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of the present disclosure can be embodied in the form of software products, and the software products can be stored in a non-volatile storage medium (which can be CD-ROM, U disk, mobile hard disk, etc.) or on the network , including several instructions to make a computing device (which may be a personal computer, a server, a touch terminal, or a network device, etc.) execute the method according to the embodiment of the present disclosure.

Other embodiments of the disclosure will be readily apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present disclosure, and these modifications, uses or adaptations follow the general principles of the present disclosure and include common knowledge or conventional technical means in the technical field not disclosed in the present disclosure . The specification and examples are to be considered exemplary only, with the true scope and spirit of the disclosure indicated by the appended claims.

It should be understood that the present disclosure is not limited to the precise constructions which have been described above and shown in the drawings, and various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims (10)

1. A fault diagnosis method for mechanical equipment, characterized in that, the fault diagnosis method comprises: Acquiring metric information of vibration signals of mechanical equipment, and determining a metric matrix through the metric information; Constructing a true neighbor neighbor graph according to an adaptive neighbor strategy, and determining distance information in the metric information through the true neighbor neighbor graph; determining at least two kernel spaces according to a preset exponential linear kernel function; Mapping different information represented by the distance information into different nuclear spaces, and determining a nuclear matrix and a nuclear space pheromone corresponding to the nuclear space; reconstructing the metric matrix according to the kernel matrix and the nuclear space pheromone fusion label discrimination information; The dimensionality reduction processing is performed on the features of the reconstructed measurement matrix, and the dimensionality reduction processing results and label discrimination information are input into the pre-trained fault classifier to obtain the fault classification results. 2. fault diagnosis method according to claim 1, is characterized in that, described according to self-adaptive neighbor strategy construction real neighbor point neighbor map comprises: Obtaining the sample point and the initial neighbor point of the vibration signal of the mechanical equipment, the initial neighbor point is determined by the neighbor value set by the sample point; Determine the number of true neighbor points and the number of initial neighbor points according to the distance between the sample point and the initial neighbor point; Constructing the neighbor graph of true neighbors according to the number of true neighbors and the number of initial neighbors. 3. The fault diagnosis method according to claim 1, wherein the mapping of different information represented by the distance information into different kernel spaces comprises: The similarity and the inner product between the distance information are calculated, and different information represented by the distance information is mapped into different kernel spaces according to the similarity and the inner product. 4. The fault diagnosis method according to claim 3, wherein the kernel space includes a first kernel space and a second kernel space, wherein the different information represented by the distance information is mapped to the different kernels space, including: determining a geodesic distance matrix and a squared geodesic distance matrix according to the true neighbor point neighbor graph; Obtain a pre-set fixed kernel matrix; performing multiplication with the fixed kernel matrix on both sides of the square geodesic distance matrix according to the dual centering method, and mapping the metric information represented by the geodesic distance into the first kernel space; The elements in the geodesic distance matrix are set as the geodesic distance information of the elements of the exponential linear kernel function matrix, and the geodesic distance matrix is mapped into the second kernel space. 5. The fault diagnosis method according to claim 4, wherein the mapping of the metric information represented by the geodesic distance into the first kernel space further comprises: determining a first kernel space matrix according to the metric information represented by the geodesic distance; Set the spectral radius of the first nuclear space matrix according to the first nuclear space matrix, and determine the first nuclear space pheromone through the spectral radius of the first nuclear space matrix; A feature representation in the first kernel space is determined according to the first kernel space pheromone. 6. The fault diagnosis method according to claim 1, wherein said metric matrix is reconstructed according to said kernel matrix and said kernel space pheromone fusion label discrimination information, comprising: The measurement matrix is reconstructed by performing nonlinear operation on the measurement matrix through the kernel matrix and the nuclear space pheromone. 7. fault diagnosis method according to claim 6, is characterized in that, described kernel matrix is the second nuclear space kernel matrix, and described nuclear space pheromone is the first nuclear space pheromone, and described through described kernel matrix And the nuclear spatial pheromone performs a nonlinear operation on the metric matrix, and reconstructing the metric matrix includes: By performing nonlinear operations on the metric matrix, and fusing the label discrimination information, the first nuclear space pheromone, the second nuclear space kernel matrix and the exponential linear kernel function, the metric matrix is nonlinearly corrected, and the reconstructed metric matrix. 8. fault diagnosis method according to claim 7, is characterized in that, described measurement matrix is carried out non-linear operation to described measurement matrix by described nuclear matrix and nuclear space pheromone, described measurement matrix is reconstructed and also comprises: fusing features in the first kernel space into a second kernel space; Said fusing the features in the first kernel space into the second kernel space includes: The first kernel space pheromone is set as the linear weight of the exponential linear kernel function, so that the features of the first kernel space are fused into the second kernel space kernel matrix. 9. The fault diagnosis method according to claim 1, wherein said dimensionality reduction process is carried out to the features of the reconstructed metric matrix, comprising: Construct the kernel matrix; The kernel matrix includes: Determine a second nuclear space matrix spectral radius according to the reconstructed metric matrix, and determine a second nuclear space pheromone according to the second nuclear space matrix spectral radius; A kernel matrix is constructed according to the reconstructed metric matrix and the second kernel space pheromone. 10. The fault diagnosis method according to claim 1, characterized in that, the dimensionality reduction process is carried out to the features of the reconstructed metric matrix, and the result of the processing and the label discrimination information are input to the pre-trained fault classifier , the fault classification results are obtained, including: Decomposing the classified features into eigenvalues to obtain the covariance matrix of the reconstructed metric matrix; Determine the eigenvalues and eigenvectors corresponding to the reconstructed metric matrix according to the covariance matrix, and perform dimensionality reduction on the classified features in the reconstructed metric matrix according to the eigenvalues and eigenvectors, to obtain the reconstructed The low-dimensional space embedding vector corresponding to the metric matrix; According to the low-dimensional space embedding vector and the label discriminant information, it is input into a pre-trained fault classifier to obtain a fault classification result.