CN116560895A

CN116560895A - Fault diagnosis method for mechanical equipment

Info

Publication number: CN116560895A
Application number: CN202310839479.7A
Authority: CN
Inventors: 王宏伟; 姚林虎; 刘峰; 李炙芫; 王浩然; 付翔; 曹文艳; 王洪利; 陶磊; 李永安; 耿毅德; 闫志蕊
Original assignee: Taiyuan University of Technology
Current assignee: Taiyuan University of Technology
Priority date: 2023-07-10
Filing date: 2023-07-10
Publication date: 2023-08-08
Anticipated expiration: 2043-07-10
Also published as: CN116560895B

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

Background

The traditional fault diagnosis method lays a foundation for extracting obvious fault characteristics in signals, but in the intelligent development process, the complexity of the mechanical equipment structure and the running state is obviously improved, and the research on intelligent fault diagnosis technology becomes an important task of mechanical equipment fault diagnosis. Artificial intelligence is a common tool for intelligent fault diagnosis, and includes a number of machine learning algorithms that can effectively refine the salient features in the data set, and that have been successfully applied to fault diagnosis of various types of rotating equipment with higher accuracy than other conventional fault diagnosis techniques.

At present, the data volume of machine equipment operation data is exponentially increased, a machine learning algorithm for machine equipment fault diagnosis can be influenced by the data volume, and physical information of the same data and different spaces is not fused in a traditional machine learning algorithm, so that the data is lost, and the accuracy of fault classification is reduced.

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

Disclosure of Invention

The embodiment of the disclosure aims to provide a fault diagnosis method, so that the influence of super parameters on a machine learning algorithm can be reduced, and the characteristic information of multiple spaces can be fused, so that the loss of data is reduced, and the accuracy of fault diagnosis by a fault classifier is improved.

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

acquiring measurement information of vibration signals of mechanical equipment, and determining a measurement matrix through the measurement information;

constructing a true neighbor point neighbor graph according to the self-adaptive neighbor strategy, and determining distance information in the measurement information through the true neighbor point 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 spaces;

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 pre-trained fault classifier to obtain a fault classification result.

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

the measurement information acquisition module is used for acquiring measurement information of the vibration signal of the mechanical equipment and determining a measurement matrix through the measurement information;

the real neighbor point neighbor map construction module is used for constructing a real neighbor point neighbor map according to the self-adaptive neighbor strategy and determining distance information in the measurement information through the real neighbor point neighbor map;

the kernel space construction module is used for determining at least two kernel spaces according to a preset exponential linear kernel function;

the feature extraction module is used for 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 spaces;

the measurement matrix reconstruction module is used for reconstructing the measurement matrix according to the nuclear matrix and the nuclear space pheromone fusion tag discrimination information;

the dimension reduction module is used for carrying out dimension reduction processing on the characteristics of the reconstructed measurement matrix, and inputting the dimension reduction processed result and the label discrimination information into the pre-trained fault classifier to obtain a fault classification result.

The technical scheme provided by the embodiment of the disclosure can comprise the following beneficial effects:

According to the embodiment of the disclosure, measurement information of vibration signals is obtained, and a measurement matrix is determined; determining distance information in the measurement information by constructing a nearest neighbor point neighbor graph; mapping different information represented by the distance information into different nuclear spaces, 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. On one hand, by constructing a true neighbor point neighbor graph according to the characteristic data of the vibration signal, the true neighbor point neighbor graph can be determined by screening the information of the vibration signal, the traditional manifold learning method generally depends on setting super parameters to determine the number of neighbor points, and the traditional super parameters need to be manually set, so that the performance and the result of an algorithm are greatly influenced, and the algorithm can adaptively select proper neighbor points without depending on manually set super parameters based on the neighbor point fusion distance information and the angle information to dynamically screen the true neighbor points, so that the influence of the super parameters on a machine learning algorithm is reduced; on the other hand, the kernel space is determined based on an exponential linear kernel function, the same features in a plurality of high-dimensional spaces are fused together to carry out reconstruction measurement matrix, the kernel space is determined by the kernel function, complex interaction between a nonlinear mode and the features can be better captured, so that richer data representation is provided, the same features in the plurality of high-dimensional spaces are fused together to carry out reconstruction measurement matrix, the essential features of data can be reflected, and the structure and the relation of original data are better reserved, so that loss and distortion of information can be avoided, data loss is reduced, effectiveness and accuracy of feature data are guaranteed, and fault classification accuracy is improved.

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 disclosure.

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. It will be apparent to those of ordinary skill in the art that the drawings in the following description are merely examples of the disclosure and that other drawings may be derived from them without undue effort.

Fig. 1 schematically shows a flow chart of a fault diagnosis method for a machine arrangement.

Fig. 2 schematically illustrates a flow chart for determining a fault category according to some embodiments of the present disclosure.

Fig. 3 schematically illustrates a flow chart of determining a true neighbor point neighbor map according to some embodiments of the present disclosure.

Fig. 4 schematically illustrates a flow chart of determining mapping different information of a distance information representation to different kernel spaces according to some embodiments of the disclosure.

Fig. 5 schematically shows a schematic diagram of a fault diagnosis apparatus for machine equipment, which may be applied to an embodiment of the present disclosure.

Fig. 6 schematically illustrates a structural schematic 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 indicate the same or corresponding parts.

Detailed Description

Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers 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 the present specification. Rather, they are merely examples of methods consistent with some aspects of the present description as detailed in the accompanying claims.

The terminology used in the description presented herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the description. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any or 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, these information should not be limited to these terms. These terms are only used to distinguish one type of information from another. For example, the first information may also be referred to as second information, and similarly, the second information may also be referred to as first information, without departing from the scope of the present description. The word "if" as used herein may be interpreted as "at … …" or "at … …" or "responsive to a determination", depending on the context.

In the related art, there are the following technical problems:

in the existing fault diagnosis technology, in order to ensure the accuracy of diagnosis, a machine learning algorithm is generally used for determining learning parameters, and the parameters are input into a pre-trained fault classifier for classification, but because the data volume of machine equipment operation data is exponentially increased, the machine learning algorithm can be influenced by the super parameters, and when the machine learning algorithm determines the learning parameters, the phenomenon of under-fitting or over-fitting of the data can occur, so that the data is lost, and the accuracy of fault classification is reduced.

Based on one or more problems in the related art, the embodiments of the present disclosure first propose a fault diagnosis method for a mechanical equipment, which may be performed by a computing model on the mechanical equipment, or may be performed by a terminal device or a server that is independent of the mechanical equipment, and the fault diagnosis method for the mechanical equipment in the present embodiment will be described below by taking a fault analysis of vibration signals collected from the mechanical equipment by the server as an example.

As described in fig. 1, fig. 1 is a flowchart illustrating a fault diagnosis method for mechanical equipment according to an exemplary embodiment of the present disclosure, including the steps of:

in step S110, obtaining measurement information of vibration signals of the mechanical equipment, and determining a measurement matrix according to the measurement information;

in step S120, a true neighbor point neighbor graph is constructed according to an adaptive neighbor policy, and distance information in the metric information is determined through the true neighbor point neighbor graph;

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

in step S140, mapping different information represented by the distance information into different kernel spaces, and determining a kernel matrix and a kernel space pheromone corresponding to the kernel spaces;

in step S150, reconstructing the metric matrix according to the kernel matrix and the kernel space pheromone fusion tag discrimination information;

in step S160, the features of the reconstructed metric matrix are subjected to dimension reduction, and the result of the dimension reduction and the label discrimination information are input into a pre-trained fault classifier to obtain a fault classification result.

According to the fault diagnosis method, the characteristic data of the vibration signal of the mechanical equipment can be obtained, and a true neighbor point neighbor graph is determined through the characteristic data; determining at least two kernel spaces according to a preset exponential linear kernel function; mapping the distance information into a nuclear space, and determining a nuclear matrix and a nuclear space pheromone of a measurement matrix corresponding to the nuclear space; and reconstructing the measurement matrix according to the nuclear matrix and the nuclear space pheromone fusion tag discrimination information. On one hand, the real neighbor point neighbor map is constructed according to the characteristic data of the vibration signals, and the real neighbor point neighbor map can be determined by screening the information of the vibration signals, so that the influence of the super parameters on a machine learning algorithm is reduced; on the other hand, the kernel space is determined based on the exponential linear kernel function, and the same features in a plurality of high-dimensional spaces are fused together to reconstruct the metric matrix, so that data loss is reduced, and the accuracy of fault classification is improved.

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

in an example embodiment of the present disclosure, the machine equipment vibration signal refers to a representation of vibrations and physical quantities of vibrations generated by the machine equipment during operation for analyzing and assessing the operational status, performance, and health of the machine equipment. Acquiring a vibration signal of the mechanical equipment may monitor an operational state of the mechanical equipment, e.g., acquiring a vibration amplitude of the vibration signal of the mechanical equipment may reflect an operational mass and balance of the mechanical equipment; the vibration frequency components of the mechanical equipment can be obtained by analysis through the acquisition of the vibration signals of the mechanical equipment, so that the information such as the rotation speed and the resonance point of the mechanical equipment can be determined; of course, it is also possible to determine whether the equipment has problems of unbalance, bearing failure, gear wear, looseness, etc. by analyzing the frequency spectrum and frequency components of the vibration signal of the mechanical equipment. The present exemplary embodiment is not particularly limited in the function of acquiring the mechanical equipment vibration signal.

The metric information refers to all characteristic data related to the vibration signal, and may be used to analyze the working state of the mechanical equipment and perform fault diagnosis on the mechanical equipment, for example, the metric information may be time domain data representing sampling time and time variation of the vibration signal, may be frequency domain data for analyzing energy distribution of the vibration signal at different frequencies, may be a spectrogram for observing spectral characteristics of the vibration signal, or may be a time domain index for describing vibration characteristics of the vibration signal at the time domain, and the different metric information may provide information of different levels and angles.

Alternatively, the measurement information may be information indicating an angle and a distance between data points in the vibration signal, the angle information is an angle relation between data points in the vibration signal in the high-dimensional euclidean space, the distance information is a distance between data points in the vibration signal in the high-dimensional euclidean space, and the relation between each data point in the vibration signal may be determined by the angle information and the distance information.

The metric matrix refers to a matrix for measuring the distance between any two feature vectors in the feature metric space, and is used for describing the distance data between the feature vectors in the metric space. The distance between two feature vectors may be determined from each element in the metric matrix, e.g., for an n-dimensional metric space, the dimension of the metric matrix is n, where the elements of the ith row and jth column represent the distance between the ith and jth vectors in the metric space. The diagonal elements of the metric matrix are typically 0, representing a distance of 0 from one vector to itself, while the non-diagonal elements represent distances between different vectors. By measuring the matrix, we can measure the distances in the vector space, and thus calculate the similarity and difference between the vectors. The distance measurement mode can be defined according to the requirements of specific tasks, so that data can be better analyzed and modeled, and the selection of the measurement matrix is not particularly limited in the present exemplary embodiment.

In step S120, a true neighbor point neighbor graph is constructed according to the adaptive neighbor policy, and distance information in the metric information is determined by the true neighbor point neighbor graph.

In an example embodiment of the present disclosure, a true neighbor point neighbor map refers to a map formed by a sample point and a true neighbor point, where the map is used to map part of feature data in a vibration signal Gao Weiou space into a low-dimensional space to represent the feature data, and the true neighbor point neighbor map may be a true neighbor point neighbor map with a fixed true neighbor point number, or a true neighbor point neighbor map that dynamically changes according to the true neighbor point number, and the specific state of the true neighbor point neighbor map is not particularly limited in this example embodiment.

For example, angle information and distance information of a vibration signal sample point can be selected to construct a dynamic nearest neighbor point nearest neighbor map, the nearest neighbor point nearest neighbor map can reduce the dimensions of the distance information and the angle information of the sample point and the nearest neighbor point in a high-dimensional European space to a low-dimensional space for representation, and the angle information and the distance information between the sample point and the nearest neighbor point obtained through the nearest neighbor point nearest neighbor map are consistent with the angle information and the distance information between the sample point and the nearest neighbor point in the high-dimensional space, so that the difficulty of acquiring the angle information and the distance information between the sample point and the nearest neighbor point in the high-dimensional space is reduced, and the stability of the relationship between the sample point and the nearest neighbor point is maintained.

The distance information in the true neighbor point neighbor map determination metric information refers to geodesic distance information between the sample point and the true neighbor point, which is used to calculate the distance between the sample point and the true neighbor point in the high-dimensional space of the vibration signal 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 spherical surface or the geodesic distance on the two-dimensional plane, and when the geodesic distance information is the geodesic distance on the two-dimensional plane, the geodesic distance is represented by Euclidean distance; of course, the geodesic distance information may also be a geodesic distance on a curved surface, and the plane on which the geodesic distance information is located in the present exemplary embodiment is not particularly limited.

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

In an example embodiment of the present disclosure, the exponential linear kernel refers to a kernel for a machine learning algorithm constructed based on exponential linear kernel theory for mapping high-dimensional data of vibration signals 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 parameters for constructing the kernel space according to experience of previous similar problems, and optimal parameters can be found by performing grid search within a certain range, and of course, the method for selecting parameters when constructing the kernel space by using the exponential linear kernel function and the method for constructing the kernel space are not particularly limited by using the bayesian optimization algorithm and adjusting parameters according to the last result in each iteration.

The determining at least two kernel spaces according to the preset exponential linear kernel function may be to store different feature data in the vibration signal, for example, two kernel spaces may be determined according to the exponential linear kernel function, two different feature data may be stored in the two kernel spaces, three or more kernel spaces may be determined according to the exponential linear kernel function, three or more feature data may be stored in the three or more kernel spaces, and the number of kernel spaces determined by the exponential linear kernel function is consistent with the number of types of feature data to be stored, where the number of kernel functions determined by the exponential linear kernel function is not particularly limited in this exemplary embodiment.

in an example embodiment of the present disclosure, the kernel matrix refers to a matrix constructed according to an exponential linear kernel function theory, and is the same as the exponential linear kernel function matrix, and is used to represent geodesic distance information of vibration signals, where the kernel matrix may be used for fusion of feature information in different kernel spaces, for example, by setting a linear weight in the exponential linear kernel function theory as related data in the kernel space, and fusing the feature information of the kernel space into the kernel matrix, so as to achieve fusion of feature information of different spaces through the kernel matrix.

The kernel space pheromone refers to data representing the characteristics in the corresponding kernel space, is used for reconstructing a metric matrix and constructing the kernel matrix, and the number of the kernel space pheromones which can be determined is generally the same as the number of the kernel spaces, for example, when two kernel spaces exist, the two kernel space pheromones can be determined according to the kernel spaces; when there are three or more nuclear spaces, three or more nuclear space pheromones may be determined by the three or more nuclear spaces, and the number of nuclear space pheromones is not particularly limited in the present exemplary embodiment, and the number of nuclear space pheromones that can be determined is generally the same as the number of nuclear spaces.

in an example embodiment of the present disclosure, the label discrimination information refers to a method for accurately classifying data points into different classes or labels defined in advance by analyzing characteristics and attributes of the data points in a vibration signal, wherein, the method comprises the steps of merging label discrimination information to reconstruct a measurement matrix, merging some data points which cannot be classified into different classes in the measurement matrix with the label discrimination information, and considering differences among different classes when reconstructing the measurement matrix, so as to improve discrimination capability core and classification performance of the measurement matrix.

The method has the advantages that the expression capability of the features in the original data can be enhanced by reconstructing the measurement matrix through the nuclear matrix and the nuclear space pheromone fusion tag discrimination information, and the reconstructed measurement matrix can better capture the key features of the input data, so that the accuracy and the performance of the subsequent processing task are improved; in addition, the reconstruction process can enable the measurement matrix to be focused on the characteristic with higher degree of distinction, reduce the influence of redundancy and useless information, be beneficial to improving the selectivity of data, better distinguish the difference between different categories and reduce the variability inside the categories; meanwhile, by fusing label discrimination information, the reconstructed measurement matrix can better combine semantic information and category relation of data, so that more category differences can be reserved in the reconstructed measurement matrix, and the performance of classification or recognition tasks can be improved; by reconstructing the metric matrix, the expressive power, the selectivity and the information integration of the data characteristics can be improved, so that better input is provided for fault classification, and more accurate results and higher performance are obtained.

In an example embodiment of the present disclosure, the dimension reduction processing refers to converting and compressing features in the reconstructed metric matrix to reduce dimensions of the features, where the dimension reduction processing may be performed on the reconstructed metric matrix by a linear or nonlinear manner, for example, the dimension reduction processing may be performed on the reconstructed metric matrix by a nonlinear processing method of feature value decomposition, or the dimension reduction processing may be performed on the reconstructed metric matrix by a linear processing method of Linear Discriminant Analysis (LDA), and of course, the dimension reduction processing may be performed on the reconstructed metric matrix by other methods.

The pre-trained fault classifier is a classifier which learns the fault mode and the fault characteristics through training of large-scale fault data, so that new input data can be subjected to fault classification, and the classifier is used for classifying input feature vectors and labels. The pre-trained fault classifier can be a variety of machine learning models and deep learning models, for example, can be an SVM classifier supporting a vector machine, and can also be a classifier supporting a random forest, of course, the pre-trained fault classifier can be selected according to actual requirements, and the type of the pre-trained fault classifier is not particularly limited in this example embodiment.

Determining a real neighboring point neighboring graph according to angle information and distance information of the vibration signal, calculating by utilizing a shortest path algorithm based on the real neighboring point neighboring graph to obtain a geodesic distance matrix and measurement information expressed by the geodesic distance, and determining the geodesic distance matrix and the measurement information expressed by the geodesic distance matrix by the real neighboring point neighboring graph can reduce the influence of super parameters on a machine algorithm model; the method comprises the steps of mapping a geodesic distance matrix and measurement information expressed by the geodesic distance to different nuclear spaces respectively, fusing a discrimination tag with the nuclear matrix and the nuclear space pheromone determined in the mapping process, reconstructing the measurement space according to an exponential linear kernel function theory, further determining another nuclear space pheromone by using the reconstructed measurement matrix, constructing a nuclear matrix according to the nuclear space pheromone, performing dimension reduction processing through characteristic information expressed in the nuclear matrix, and finally inputting a dimension reduction processed result and the tag discrimination information into a pre-trained machine classifier to obtain a classification result, so that the loss of data in the extraction process can be reduced, and the accuracy of fault diagnosis of the classifier is improved.

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

In an example embodiment of the present disclosure, the determination of the true neighbor map in step S110 may be achieved by the following steps and with reference to fig. 3:

step S310, obtaining a sample point and a primary adjacent point of a vibration signal of the mechanical equipment, wherein the primary adjacent point is determined by an adjacent value set by the sample point; step S320, determining the true neighbor point number and the primary neighbor point number according to the distance between the sample point and the primary neighbor point; and step S330, constructing a true neighbor point neighbor graph according to the true neighbor point number and the initial neighbor point number.

The sample points of the vibration signal of the mechanical equipment refer to a group of data points selected from the vibration signal, wherein the data points comprise angle characteristic information and distance characteristic information of the vibration signal in a time or frequency domain and are used for characteristic extraction, fault detection and diagnosis of the vibration signal. The sample points may be selected with a sufficient number of data points, for example, the sample points may be selected with data points that are sufficiently representative of the entire vibration signal and are uniformly distributed in the time or frequency domain, and the sample points may be selected according to the processing method and application scenario of the vibration signal, and the selection of the sample points is not particularly limited in this exemplary embodiment.

The primary neighboring point is a data point having a certain distance from the sample point, and is used for selecting a suitable data point near the sample point to construct a primary neighboring diagram, the primary neighboring point can be determined by presetting a neighboring value K, or the primary neighboring point can be determined by meshing, or of course, a fixed range can be set, the data point within the fixed range from the sample point is taken as the primary neighboring point, the primary neighboring point selection method can be selected according to requirements, and the primary neighboring point selection method is not particularly limited in this example embodiment.

Optionally, the primary neighboring point is determined by a preset neighboring value K, where the neighboring point of each sample point is determined by the preset neighboring value K and defined as the primary neighboring point. For example, if the neighbor value K is set to 5, the shortest distance between the data and the sample point is calculated according to the shortest path algorithm, and 5 data points with a relatively close distance to the sample point are selected to form a neighboring point, namely, a primary neighboring point.

Optionally, when constructing a nearest neighbor point neighbor graph, pre-processing is required to be performed on the initial nearest neighbor points to obtain a Euclidean distance matrix, and normalization processing is performed on the Euclidean distance matrix to eliminate differences in distance scales between different sample points, so that the distance between each sample point is ensured to have comparability, the performance and accuracy of the algorithms are improved, and the condition of under fitting of data is reduced; and in some cases, the contributions of different features to the distance calculation may be unbalanced, and by normalizing the distance matrix, the weights between different features can be balanced, so that the influence of some features on the calculation of the Euclidean distance is avoided.

Wherein the Euclidean distance matrix refers to a matrix for measuring the linear distance between the primary neighboring point and the sample point, and is used for representing the primary neighboring point The Euclidean distance between a point and a sample point is similar to the Euclidean distance matrix for data points in a high dimensional space, and the calculation of the square sum of the coordinate differences is applied to all dimensions. For example, the initial neighbor graph G may be first constructed according to the determined initial neighbor points; then calculating the cosine similarity matrix of the initial neighborWherein: />，/>The kth dimension characteristic representing the ith sample point,representing a set of initial neighbor points for an ith sample point; then, reconstruct neighbor cosine similarity adjustment matrix +.>Wherein->Finally, calculating Euclidean distance matrix between the initial adjacent point and the central pointWherein->。

Normalization refers to Scaling data to scale the processed data within a specific range or relative size, and in data processing and machine learning, normalization generally refers to converting data to a range between 0 and 1, also known as Min-Max Scaling, to eliminate differences between features. Of course, the normalization processing may be performed by other methods, for example, the normalization processing may be performed by a method of Z-score normalization, or the normalization processing may be performed by normal distribution, and the specific method of the normalization processing is not particularly limited in this example embodiment.

For example, the initial neighboring Euclidean distance matrix of all sample points is normalized by a min-max scaling method to obtain N neighboring point normalized Euclidean distance matrices: wherein the method comprises the steps ofData in the euclidean distance matrix is converted into a range between 0 and 1.

The true neighbor points refer to sample points, wherein the weight of each sample point is smaller than or equal to the average weight of the sample points, and the sample points are used for determining the number of the true neighbor points, so that a true neighbor point neighbor graph is constructed according to the number of the true neighbor points. For example, a weight matrix between all sample points and their initial neighbors may be constructed:wherein->The method comprises the steps of carrying out a first treatment on the surface of the Calculating average weight of all sample points and initial neighbor point set>Wherein->The method comprises the steps of carrying out a first treatment on the surface of the Then the sample point x _i Is +.>Average weight to the sample point +.>Comparing if->Then consider x _j Is x _i Pseudo-neighbor of (a)Point of->Then consider x _j Is x _i Is a true neighbor of (a); according to the number of the true neighbor points corresponding to each sample point, dynamically adjusting the neighbor point number of the sample point to obtain a dynamic true neighbor value K ⁺ Constructing a dynamic nearest neighbor point nearest neighbor graph G ⁺ 。

In an example embodiment of the present disclosure, mapping different information represented by distance information into different kernel spaces in step S140 may be achieved by:

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

Alternatively, there are several ways to map distance information to different kernel spaces, for example, metric information represented by geodesic distances may be mapped into kernel spaces by a double-centering method; the geodesic distance matrix can be mapped into another kernel space by a kernel function mapping method; of course, the distance information may also be mapped into the core space by the feature value extraction method, and the present exemplary embodiment is not particularly limited as to the method of mapping the distance information into the core space.

For example, the exponential linear kernel function may be determined based on an exponential linear kernel function theory, where the exponential linear kernel function is a kernel function used to map data in an input space into a high-dimensional feature space, a similarity between two sample points may be measured, and a kernel matrix determined by the exponential linear kernel function has a semi-positive property, where the semi-positive property of the kernel matrix may ensure stability and correctness of a machine learning algorithm, and where the kernel matrix determined by the exponential linear kernel function has a semi-positive property may be demonstrated using geodesic distance information and filling conditions of the kernel function. For example, an exponential linear kernel function matrix is determined based on geodesic distance information and exponential linear kernel function theory Wherein->Is a linear weight, b (0<b.ltoreq.1) is the offset coefficient, +.>For sample x _i And x _j A geodesic distance between; and according to the filling conditions of the kernel function: for any data sampleIts corresponding Gram matrix->Is a semi-positive definite matrix, N is the number of samples. The semipositive nature of the exponential linear kernel function matrix is demonstrated as follows:

the exponential linear kernel function structure can know that it utilizes geodetic matrix information, then it is obvious that matrixIs a symmetric matrix, so that only +.>A semi-positive definite matrix; deriving a mapping of the exponential linear function:if sample x _j Is x _i Is a neighbor point of->ThenWherein the mapping->From this, it can be seen that when sample x _j Is x _i The exponential linear function satisfies the kernel function filling condition when adjacent points of (a).

If sample x _j Not x _i Is adjacent to (a) then there is，/>Is the finite term d ^E And, therefore, do not let->With z term d ^E And->The following steps are: />，

From this, it can be seen that when sample x _j Not x _i The exponential linear function also satisfies the kernel function filling condition when adjacent points of (a).

From the above-mentioned evidence, it can be known that the symmetric matrixHas semi-positive properties, satisfying the filling conditions of the kernel function, and thus the kernel matrix determined by the exponential linear kernel function has semi-positive properties.

The similarity between the distance information is calculated by using an exponential linear kernel function, which can be calculated by calculating the euclidean distance between the sample points x and y by using the exponential linear kernel function, and then the euclidean distance is used as an index of the exponential linear kernel function, so that a smaller distance generates a larger similarity value and a larger distance generates a smaller similarity value; the inner product between distance information calculated using an exponential linear kernel function may be calculated using the definition of the exponential linear kernel function.

By calculating the similarity and inner product between the distance information, the similarity value between the sample points and the inner product value in the kernel space can be obtained. These similarity values and inner product values of the kernel space may be used for feature extraction, mapping the distance information of the sample points into the kernel space. The similarity and inner product between the distance information are calculated through the exponential linear kernel function, and different distance information are mapped into different kernel spaces, so that an effective mode is provided for processing the nonlinear separable sample point data, and an important role is played in a pre-training machine learning task.

In an example embodiment of the present disclosure, mapping metric information of the geodesic distance representation into the first kernel space may be accomplished by:

step S410, determining a geodesic distance matrix and a square geodesic distance matrix according to the nearest neighbor point neighbor map; step S420, a preset fixed kernel matrix is obtained; and step S430, performing multiplication operation on the two sides of the square geodesic distance matrix and the fixed kernel matrix respectively according to a double-centering method, and mapping measurement information represented by the geodesic distance into a first kernel space.

The geodesic distance matrix is a distance matrix determined based on a nearest neighbor point neighbor graph and is used for measuring similarity and difference between data points. The geodesic distance matrix may be determined by using a shortest path algorithm based on the true neighbor map, for example, the geodesic distance between each data point in the true neighbor map may be calculated by Dijkstra's algorithm, or Floyd's algorithm, then the geodesic distance matrix is determined according to the determined geodesic distance, and a suitable algorithm may be selected according to the need.

The metric information represented by the geodetic distance refers to the geodetic distance represented in the geodetic distance matrix for measuring the distance of the shortest path between data points. The measurement 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 measurement information represented by the geodesic distance matrix in this example embodiment is not described in detail.

A fixed core matrix refers to a matrix used to non-linearly map data and map the data into core space. The mapping of data into the kernel space by the fixed kernel matrix is achieved by multiplying the two sides of the square geodesic matrix by the fixed kernel matrix, which can be determined by a custom kernel function, for example by selecting an appropriate kernel width parameter sigma from an exponential linear kernel function, calculating the kernel values between each pair of samples and combining them into one kernel matrix, thus determining the fixed kernel matrix. Of course, the fixed kernel matrix may also be determined according to other kernel functions, such as gaussian kernels or polynomial kernels, and the determination of the kernel matrix is not particularly limited in this exemplary embodiment.

The first nuclear space nuclear matrix is a matrix representing the geodesic distance calculated according to the geodesic distance matrix and the square geodesic distance matrix, and is used for mapping the measurement information represented in the geodesic distance matrix into the first nuclear space. The first kernel space kernel matrix may be determined by a geodesic distance matrix, for example, a true neighbor graph G may be obtained by an adaptive neighbor strategy ⁺ Calculating to obtain a geodesic distance matrix D _G Sum-square geodesic distance matrixSimultaneously calculate the matrix operator +.>I.e. a first core-space core matrix.

Mapping metric information represented by geodesic distances into the first kernel space may be based on a geodesic distance matrix D _G Calculating to obtain a square geodesic distance matrix, wherein the square geodesic matrix isThen multiplying the square geodesic distance matrix with the H matrix by double-centering method to obtain +.>The metric information represented by the geodetic matrix is mapped into the first kernel space by an operation between the matrices. Where the H matrix is a known fixed kernel matrix during the spatial mapping process, this disclosure will not be described in detail.

In an example embodiment of the present disclosure, mapping the geodesic distance matrix into the second kernel space may be accomplished by:

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 into the kernel space by an exponential linear kernel function, wherein the exponential linear kernel function is one of the kernel functions and has a semi-positive property, and the semi-positive property of the exponential linear kernel function can ensure the stability of the data points in the mapping process. The geodesic distance matrix can be obtained by first calculating, mapping each element in the geodesic distance matrix into an exponential linear kernel function, and determining a second kernel space according to the exponential linear kernel function, namely mapping each element in the geodesic distance matrix into the second kernel space.

For example, an exponential linear kernel function matrix isAnd (2) anddistance to ground matrix D _G Element->Substituting into elements in an exponential linear kernel function matrix to thereby obtain a geodesic distance matrix D _G Mapping the index linear kernel function into a second kernel space to finish feature extraction and obtain a second kernel space kernel matrix, namely +.>。

According to the nonlinear mapping of the exponential linear kernel function, the nonlinear relation between samples can be captured, more effective feature representation and learning can be performed in the second kernel space, the feature dimension can be reduced while key information is maintained, the feature expression capability is improved, and the machine learning task can be better performed.

In an example embodiment of the present disclosure, determining a feature representation in a first kernel space by a first kernel space pheromone may be accomplished by:

and determining a first nuclear space matrix according to the measurement information of the geodesic distance representation, setting a first nuclear space matrix spectrum radius according to the first nuclear space matrix, determining a first nuclear space pheromone according to the first nuclear space matrix spectrum radius, and determining the characteristic representation in the first nuclear space according to the first nuclear space pheromone.

The first core space pheromone refers to data representing features in the first core space and is used for reconstructing a metric matrix, and the first core space pheromone is determined by the first core space core matrix, for example, by the first core Hilbert space core matrix Constructing a blocking matrix->And calculating the first nuclear space matrix spectral radius +.>And let Hilbert spatial pheromone +.>A first nuclear space pheromone is determined.

The feature representation in the first kernel space is determined through the first kernel space pheromone, so that the mode of operating the feature in the first kernel space in subsequent operation can be simplified, and the accuracy of data is ensured while the operation efficiency is improved.

In an example embodiment of the present disclosure, the reconstruction of the metric matrix according to the nuclear matrix and the nuclear space pheromone fusion tag discrimination information in step 150 may be implemented by:

and carrying out nonlinear operation on the measurement matrix through the core matrix and the core space pheromone, and reconstructing the measurement matrix.

The method for reconstructing the metric matrix can adopt a linear correction method or a nonlinear correction method, and the method for adopting the linear correction can reconstruct the metric matrix through linear transformation, such as Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA); the method of nonlinear correction may be to reconstruct a metric matrix using a nonlinear function, such as an exponential linear kernel function, reconstruct the metric matrix using an exponential linear kernel function, map the extracted features into a high-dimensional feature space, and then perform linear correction in the high-dimensional feature space; of course, a nonlinear correction method of popular learning can also be adopted, the local geometry of the sample is utilized to reconstruct the measurement matrix, the manifold structure of the data can be better reserved, and the method for reconstructing the measurement matrix is not particularly limited in the present exemplary embodiment.

The reconstruction of the metric matrix can improve the data characterization capability, reduce dimensionality, enhance classification performance, facilitate data visualization, association analysis and the like, thereby greatly helping machine learning tasks.

In an example embodiment of the present disclosure, the non-linear correction of the metric matrix may be implemented by:

and carrying out nonlinear operation on the measurement matrix, fusing label discrimination information, a first nuclear space pheromone, a second nuclear space nuclear matrix and an exponential linear kernel function theory, carrying out nonlinear correction on the measurement matrix, and determining the reconstructed measurement matrix.

The method comprises the steps of reconstructing a measurement matrix, carrying out nonlinear operation on the measurement matrix through a first nuclear space pheromone, label discrimination information, a second nuclear space nuclear matrix and an exponential linear kernel function theory, and carrying out nonlinear correction on the measurement matrix through changing parameters in the measurement matrix through partial matrix or data obtained in the feature extraction process, so that the reconstruction of the measurement matrix is realized. For example, the metric matrix has the following expression:nonlinear correction is carried out on the measurement matrix by fusing label discrimination information, a second core Hilbert space core matrix and an exponential linear core function theory: Wherein: />Is a standoff coefficient.

In an example embodiment of the present disclosure, fusing the first core-space pheromone into the second core-space core matrix may be accomplished by:

wherein, the linear weight refers to a parameter for determining an exponential linear kernel function matrix based on an exponential linear kernel function theory, and the parameter is used for constructing 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 isI.e., linear weights, features in the exponential linear kernel function matrix may be changed by changing the values of the linear weights.

Alternatively, let the linear weights in the exponential linear kernel theoryThe first core Hilbert spatial pheromone can be fused into the exponential linear kernel function matrix, and the second core spatial kernel matrix is determined by the exponential linear kernel function matrix, so that the features of the first core space can be fused into the second core spatial kernel matrix by setting the linear weight in the exponential linear kernel function as the first core spatial pheromone, and the fusion of the features in different high-dimensional spaces is completed.

In an example embodiment of the present disclosure, the Mercer kernel matrix may be constructed by:

And determining a second nuclear space matrix spectrum radius according to the reconstructed metric matrix, determining a second nuclear space pheromone according to the second nuclear space matrix spectrum radius, and constructing a Mercer nuclear matrix according to the reconstructed metric matrix and the second nuclear space pheromone.

The second kernel space pheromone refers to data representing the features in the metric matrix after reconstruction and is used for constructing a Mercer kernel matrix, the second kernel space pheromoneThe kernel space pheromone is determined by the reconstructed metric matrix, for example, the partitioned matrix is reconstructed by using the nonlinear correction matrix of the metric matrixAnd calculating a second kernel space matrix spectral radius, < >>Let the second core Hilbert spatial pheromone +.>。

The Mercer kernel matrix refers to a kernel matrix with symmetrical positive definite property, and is used for conveniently reducing the dimension of features in the reconstructed metric matrix. The Mercer Kernel matrix may be used for various pre-training machine learning tasks, for example, the Mercer Kernel matrix may Support Vector Machine (SVM) classifier learning, and may also support Kernel principal component analysis (Kernel PCA), and the pre-training machine used for learning by the Mercer Kernel matrix is not particularly limited in this example embodiment.

Mercer core matrix can be based on the second core Hilbert spatial pheromoneAnd a reconstructed squaring metric matrix +. >And the corresponding kernel space mapped by the reconstructed metric matrix through the double-centering methodThe specific construction method is as follows: />。

Determining a second nuclear space pheromone by using the reconstructed metric matrix, and constructing a Mercer nuclear matrix,

explicit calculation in a high-dimensional feature space can be avoided, so that the calculation complexity is reduced, and the calculation accuracy is improved.

In an example embodiment of the present disclosure, the feature of the reconstructed metric matrix may be subjected to dimension reduction processing, and the processed result and the tag discrimination information are input into a pre-trained fault classifier, so as to obtain a fault classification result:

and decomposing the characteristic values in the Mercury core matrix to obtain a covariance matrix of the Mercury core matrix, determining the corresponding characteristic values and characteristic vectors in the Mercury core matrix according to the covariance matrix, reducing the dimensions of the characteristics of the Mercury core matrix according to the characteristic values and the characteristic vectors to obtain a low-dimensional space embedded vector corresponding to the Mercury core matrix, and inputting the low-dimensional space embedded vector and label discrimination information into a pre-trained fault classifier to obtain a fault classification result.

The method for decomposing the feature values of the features in the Mercer kernel matrix is the prior art, and is not described herein, and the feature value decomposition method can be used to obtain the feature value Is embedded with coordinates +.>And (5) finishing the dimension reduction process of the feature extraction.

The method has the advantages that the dimension of the features of the Mercer core matrix is reduced, the low-dimension space embedded vector obtained by dimension reduction of the Mercer core matrix and the label discrimination information are input into the pre-trained fault classifier, the fault classification result is finally obtained, a technician can be helped to quickly and accurately find the fault cause, corresponding measures can be timely taken to solve the problem, and the efficiency and the accuracy of fault diagnosis can be improved by carrying out fault diagnosis through the pre-trained fault classifier.

Fig. 2 schematically illustrates a flow chart for fault diagnosis of mechanical equipment according to some embodiments of the present disclosure.

Referring to fig. 2, the fault diagnosis of the mechanical equipment may be implemented through steps S201 to S207, wherein:

step S201, angle information and distance information of a vibration signal are obtained;

step S202, obtaining a true neighbor point neighbor graph according to angle information and distance information; the method comprises the steps of determining a primary neighbor point of a sample point by setting a primary neighbor value, comparing a weight value of a distance between the sample point and the primary neighbor point with an average weight value to obtain a true neighbor point and a false neighbor point, constructing a true neighbor point neighbor graph according to the determined true neighbor points around the sample point, and dynamically changing the true neighbor point neighbor graph according to the number of the true neighbor points.

Step S203, determining a geodesic distance matrix according to the nearest neighbor point nearest neighbor map; the method comprises the steps of determining a geodesic distance matrix according to a nearest neighbor point neighbor diagram, wherein the geodesic distance matrix is obtained through a shortest path algorithm, and the shortest path of elements in the nearest neighbor point neighbor diagram can be calculated to determine the elements of the geodesic distance, so that the geodesic distance matrix is determined.

Step S204, mapping the geodesic distance matrix and the measurement information expressed by the geodesic distance into the nuclear space respectively; the method comprises the steps of mapping measurement information represented by a geodesic distance matrix into a first kernel space through a double-centering method, and mapping the geodesic distance matrix into a second kernel space through an exponential linear kernel function.

Step S205, reconstructing a measurement matrix according to the nuclear matrix and the nuclear space pheromone determined in the characteristic mapping process and by fusing label discrimination information and an exponential linear kernel function theory; the method comprises the steps that a first kernel space kernel matrix and a second kernel space kernel matrix are respectively determined in a feature mapping process, the first kernel space kernel matrix is fused into a second kernel space kernel matrix by changing the linear weight of the second kernel space matrix in a determining process, fusion of the first kernel space information and the second kernel space information is achieved, nonlinear operation is carried out on the measurement matrix through fusion tag discrimination information, and therefore the measurement matrix after reconstruction is obtained.

Step S206, expressing the characteristics of the reconstructed metric matrix by using a nuclear space pheromone, and constructing a Mercer nuclear matrix; the features of the reconstructed metric matrix are represented by a second kernel space pheromone, and the first kernel space pheromone is integrated into the second kernel space kernel matrix, so that the second kernel space pheromone can represent the features in the reconstructed metric matrix.

Step S207, performing dimension reduction processing on the features in the Mercer kernel matrix, and inputting the obtained dimension reduction processing result and label discrimination information into a pre-trained classifier to obtain a fault classification result; the method comprises the steps of performing dimension reduction processing on a low-dimensional space embedded vector, and inputting the low-dimensional space vector and label discrimination information into a pre-trained fault classifier to obtain a fault classification result.

Next, a fault diagnosis apparatus for mechanical equipment according to an exemplary embodiment of the present disclosure will be described with reference to fig. 5.

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

the feature extraction module is used for mapping different information represented by the distance information into different nuclear spaces to perform feature extraction and determining a nuclear matrix and a nuclear space pheromone corresponding to the nuclear spaces;

The method comprises the steps of obtaining measurement information of a vibration signal, determining a measurement matrix, constructing a nearest neighbor point neighbor graph according to angle information and distance information in the measurement information, determining a geodesic distance matrix and measurement information expressed by geodesic distances between data points of the vibration signal in a high-dimensional space through the nearest neighbor point neighbor graph, determining a nuclear space through a kernel function, mapping the measurement information expressed by the geodesic distances into a first nuclear space, mapping the geodesic distance matrix into a second nuclear space, and finally reconstructing the measurement matrix through the second nuclear space nuclear matrix and the first nuclear space pheromone determined in the process of mapping the measurement information into the nuclear space, fusing tag discrimination information and an exponential linear kernel function theory, thereby realizing fusion of information in multiple spaces, determining characteristic expression in the reconstructed measurement matrix through the second nuclear space pheromone, constructing a Merger nuclear matrix by utilizing the square of the reconstructed measurement matrix and the second nuclear space pheromone, obtaining a low-dimensional embedded vector corresponding to the characteristic in the Merger nuclear matrix according to a method of decomposing the Merger nuclear matrix, and finally inputting the low-dimensional embedded vector and the low-dimensional predictive tag discrimination information into a training result, thereby obtaining a classifier.

It should be noted that although the steps of the methods of the present disclosure are illustrated in the accompanying drawings in a particular order, this does not require or imply that the steps must be performed in that particular order or that all of the illustrated steps be performed in order to achieve desirable results. Additionally or alternatively, certain steps may be omitted, multiple steps combined into one step to perform, and/or one step decomposed into multiple steps to perform, etc.

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

Those skilled in the art will appreciate that the various aspects of the present disclosure may be implemented as a system, method, or program product. Accordingly, various aspects of the disclosure may be embodied in the following forms, namely: an entirely hardware embodiment, an entirely software embodiment (including firmware, micro-code, etc.) or an embodiment combining hardware and software aspects may be referred to herein as a "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 merely an example and should not be construed as limiting the functionality and scope of use of the disclosed embodiments.

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

Wherein the storage unit stores program code that is executable by the processing unit 610 such that the processing unit 610 performs steps according to various exemplary embodiments of the present disclosure described in the above-described "exemplary methods" section of the present specification. For example, the processing unit 610 may perform step S110 shown in fig. 1, acquire metric information of the mechanical equipment vibration signal, and determine a metric matrix from the metric information; step S120, constructing a true neighbor point neighbor graph according to the self-adaptive neighbor strategy, and determining distance information in the measurement information through the true neighbor point neighbor graph; step S130, determining at least two kernel spaces according to a preset exponential linear kernel function; step S140, 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 spaces; step S150, reconstructing the metric matrix according to the nuclear matrix and the nuclear space pheromone fusion tag discrimination information; and step S160, 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 pre-trained fault classifier to obtain a fault classification result.

The storage unit 620 may include readable media in the form of volatile storage units, such as Random Access Memory (RAM) 621 and/or cache memory 622, and may further include Read Only Memory (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, each or some combination of which may include an implementation of a network environment.

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

The electronic device 600 may also communicate with one or more external devices 670 (e.g., keyboard, pointing device, bluetooth device, etc.), one or more devices that enable a user to interact with the electronic device 600, and/or any devices (e.g., routers, modems, etc.) that enable the electronic device 600 to communicate with one or more other computing devices. Such communication may occur through an input/output (I/O) interface 650. Also, electronic device 600 may 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 network adapter 660. As shown, network adapter 660 communicates with other modules of electronic device 600 over bus 630. It should be appreciated that although not shown, other hardware and/or software modules may be used in connection with electronic device 600, including, but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, data backup storage systems, and the like.

From the above description of embodiments, those skilled in the art will readily appreciate that the example embodiments described herein may be implemented in software, or in combination with the necessary hardware. Thus, the technical solution according to the embodiments of the present disclosure may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (may be a CD-ROM, a U-disk, a mobile hard disk, etc.) or on a network, and includes several instructions to cause a computing device (may be a personal computer, a server, a terminal device, or a network device, etc.) to perform the method according to the embodiments of the present disclosure.

In an exemplary embodiment of the present disclosure, a computer-readable storage medium having stored thereon a program product capable of implementing the method described above in the present specification is also provided. In some possible embodiments, the various aspects of the present disclosure may also be implemented in the form of a program product comprising program code for causing a terminal device to carry out the steps according to the various exemplary embodiments of the disclosure as described in the "exemplary methods" section of this specification, when the program product is run on the terminal device.

A program product for implementing the above-described fault diagnosis method for mechanical equipment according to an embodiment of the present disclosure may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto, and in this document, a readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

The program product may employ 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 can be, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium would include the following: an electrical connection having one or more wires, a portable disk, a 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), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The computer readable signal medium may include a data signal propagated in baseband or as part of a carrier wave with readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A readable signal medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection 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 carrying out operations of the present disclosure may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. 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, partly on a remote computing device, or entirely on the remote computing device or server. In the case of remote computing devices, 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 may be connected to an external computing device (e.g., connected via the Internet using an Internet service provider).

Furthermore, the above-described figures 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 will be readily appreciated that the processes shown in the above figures do not indicate or limit the temporal order of these processes. In addition, it is also readily understood that these processes may be performed synchronously or asynchronously, for example, among a plurality of modules.

From the above description of embodiments, those skilled in the art will readily appreciate that the example embodiments described herein may be implemented in software, or in combination with the necessary hardware. Thus, the technical solution according to the embodiments of the present disclosure may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (may be a CD-ROM, a U-disk, a mobile hard disk, etc.) or on a network, and includes several instructions to cause a computing device (may be a personal computer, a server, a touch terminal, or a network device, etc.) to perform the method according to the embodiments of the present disclosure.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any adaptations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It is to be understood that the present disclosure is not limited to the precise arrangements and instrumentalities shown in the drawings, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A fault diagnosis method for mechanical equipment, characterized in that the fault diagnosis method comprises:

constructing a true neighbor point neighbor graph according to a self-adaptive neighbor strategy, and determining distance information in the measurement information through the true neighbor point neighbor graph;

reconstructing the metric matrix according to the nuclear matrix and the nuclear space pheromone fusion tag discrimination information;

2. The fault diagnosis method according to claim 1, wherein constructing a true neighbor point neighbor map according to an adaptive neighbor policy comprises:

acquiring a sample point and a primary neighboring point of a vibration signal of mechanical equipment, wherein the primary neighboring point is determined by a neighboring value set by the sample point;

determining the true neighbor point number and the initial neighbor point number according to the distance between the sample point and the initial neighbor point;

and constructing the nearest neighbor point neighbor graph according to the nearest neighbor point number and the initial nearest neighbor point number.

3. The fault diagnosis method according to claim 1, wherein said mapping different information of the distance information representation into different ones of the kernel spaces comprises:

4. A fault diagnosis method according to claim 3, the core space comprising a first core space and a second core space, wherein said mapping different information represented by the distance information into the different core spaces comprises:

determining a geodesic distance matrix and a square geodesic distance matrix according to the nearest neighbor point neighbor map;

Acquiring a preset fixed core matrix;

multiplying the two sides of the square geodesic distance matrix with the fixed kernel matrix according to a double-centering method, and mapping measurement information represented by the geodesic distance into the first kernel space;

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

5. The fault diagnosis method according to claim 4, wherein the mapping of metric information represented by geodesic distances into the first kernel space further comprises:

determining a first nuclear space matrix according to the measurement information represented by the geodesic distance;

setting a first nuclear space matrix spectrum radius according to the first nuclear space matrix, and determining a first nuclear space pheromone according to the first nuclear space matrix spectrum radius;

and determining the characteristic representation in the first nuclear space according to the first nuclear space pheromone.

6. The method of claim 1, wherein reconstructing the metric matrix from the nuclear matrix and the nuclear space pheromone fusion tag discrimination information comprises:

And carrying out nonlinear operation on the measurement matrix through the nuclear matrix and the nuclear space pheromone, and reconstructing the measurement matrix.

7. The method of claim 6, wherein the kernel matrix is a second kernel space kernel matrix, the kernel space pheromone is a first kernel space pheromone, and the reconstructing the metric matrix by performing nonlinear operation on the metric matrix through the kernel matrix and the kernel space pheromone comprises:

and carrying out nonlinear operation on the measurement matrix, fusing label discrimination information, a first nuclear space information element, a second nuclear space nuclear matrix and an exponential linear nuclear function theory, and carrying out nonlinear correction on the measurement matrix to determine a reconstructed measurement matrix.

8. The method of claim 7, wherein said non-linearly operating said metric matrix with said kernel matrix and said kernel space pheromone, and reconstructing said metric matrix further comprises:

fusing features in the first kernel space into a second kernel space;

the fusing features in the first kernel space into a second kernel space includes:

And setting the first kernel space pheromone as the linear weight of the exponential linear kernel function, so that the characteristic of the first kernel space is fused into a second kernel space kernel matrix.

9. The fault diagnosis method according to claim 1, wherein the performing the dimension reduction processing on the features of the reconstructed metric matrix includes:

constructing a kernel matrix;

the core matrix includes:

determining a second nuclear space matrix spectral radius according to the reconstructed metric matrix, and determining a second nuclear space pheromone according to the second nuclear space matrix spectral radius;

and constructing a kernel matrix according to the reconstructed metric matrix and the second kernel space pheromone.

10. The fault diagnosis method according to claim 1, wherein the performing the dimension reduction processing on the features of the reconstructed metric matrix, and inputting the processed result and the tag discrimination information into a pre-trained fault classifier, to obtain a fault classification result, includes:

performing eigenvalue decomposition on the classified features to obtain a covariance matrix of the reconstruction metric matrix;

determining a characteristic value and a characteristic vector corresponding to the reconstruction measurement matrix according to the covariance matrix, and performing dimension reduction on classified characteristics in the reconstruction measurement matrix according to the characteristic value and the characteristic vector to obtain a low-dimensional space embedded vector corresponding to the reconstruction measurement matrix;

And inputting the low-dimensional space embedded vector and the label discrimination information into a pre-trained fault classifier to obtain a fault classification result.