Method and device for analyzing a fault state of a mining equipment based on multiple sensors

Abstract

The application relates to a multi-sensor-based fault state analysis method and device for mining equipment. The method comprises the following steps: using multiple sensors arranged at different positions of the mining equipment to respectively collect multiple health state vibration signals of the mining equipment in a health state and multiple real-time working state vibration signals of the mining equipment in a real-time working state; processing the health state vibration signals and the real-time working state vibration signals collected by each sensor to construct a corresponding reference feature set of the sensor; processing the multiple reference feature sets corresponding to the multiple sensors to obtain a multi-sensor enhanced feature set of the multiple sensors; and determining a fault state of the mining equipment by using a classifier according to the multi-sensor enhanced feature set. According to the technical scheme, different faults of the mining equipment can be more accurately recognized, and the accuracy of fault analysis of the mining equipment is improved.

Description

Technical Field

[0001] This application generally relates to the field of modern mining engineering technology, and in particular to a method and apparatus for analyzing the fault status of mining equipment based on multiple sensors. Background Technology

[0002] In actual coal production, equipment is affected by operating conditions, environment, and power supply systems, resulting in nonlinear and unstable vibration signals. The installation location of vibration sensors is also constrained by equipment structure, harsh environments, and electrical wiring, making it impossible to place all sensors in ideal locations. Furthermore, predicting the extent to which a particular fault manifests at a specific measuring point or in which direction is difficult for such a complex mechanical system. This is especially true for underground coal mine equipment, where vibration signals are characterized by redundancy, high dimensionality, and strong nonlinearity. Therefore, single-sample feature extraction from a single sensor cannot fully address these challenges during fault identification. For large equipment like tunneling machines, faults are influenced by operational vibrations, and while there may be some differences, they are not significant. Single-sample feature extraction can easily miss effective features or distort target parameters, hindering fault identification of critical components. While healthy and unhealthy vibration signals are easily distinguished during the operation of large coal mine equipment, identifying unhealthy signals from healthy signals remains challenging. Summary of the Invention

[0003] To address at least one of the aforementioned technical problems, this application provides a multi-sensor-based method and apparatus for analyzing the fault status of mining equipment, aiming to more accurately identify different faults in mining equipment and improve the accuracy of fault analysis.

[0004] According to a first aspect of this application, a method for analyzing the fault status of mining equipment based on multiple sensors is provided, comprising: using multiple sensors arranged at different locations on the mining equipment to collect multiple healthy state vibration signals and multiple real-time operating state vibration signals of the mining equipment in a healthy state and in a real-time operating state, respectively; processing the healthy state vibration signals and the real-time operating state vibration signals collected by each sensor to construct a reference feature set corresponding to the sensor, wherein the reference feature set is formed by fusing the healthy state vibration signal feature set corresponding to the healthy state vibration signals and the real-time operating state vibration signal feature set corresponding to the real-time operating state vibration signals; processing the multiple reference feature sets corresponding to the multiple sensors to obtain a multi-sensor enhanced feature set of the multiple sensors; and using a classifier to determine the fault status of the mining equipment based on the multi-sensor enhanced feature set.

[0005] In one embodiment, processing the health state vibration signal and the real-time operating state vibration signal collected by each sensor to construct a reference feature set corresponding to the sensor includes: analyzing the health state vibration signal and the real-time operating state vibration signal to obtain sensitive feature parameters of the health state vibration signal and the real-time operating state vibration signal; constructing a health state vibration signal feature set and a real-time operating state vibration signal feature set corresponding to each sensor based on the sensitive feature parameters of the health state vibration signal and the real-time operating state vibration signal; and fusing the health state vibration signal feature set and the real-time operating state vibration signal feature set corresponding to each sensor to construct the reference feature set corresponding to the sensor.

[0006] In one embodiment, analyzing the healthy state vibration signal and the real-time operating state vibration signal to obtain sensitive feature parameters of the healthy state vibration signal and the real-time operating state vibration signal includes: performing time-domain analysis and wavelet packet analysis on the healthy state vibration signal and the real-time operating state vibration signal to obtain the sensitive feature parameters of the healthy state vibration signal and the real-time operating state vibration signal, wherein the sensitive feature parameters include time-domain parameters and wavelet parameters.

[0007] In one embodiment, processing the multiple reference feature sets corresponding to the multiple sensors to obtain a multi-sensor enhanced feature set of the multiple sensors includes: performing a first-layer dimensionality reduction process on the reference feature set of each sensor based on manifold learning to obtain low-dimensional features of each sensor; fusing the low-dimensional features of the multiple sensors to obtain a multi-sensor feature set of the multiple sensors; and performing a second-layer dimensionality reduction process on the multi-sensor feature set based on manifold learning to obtain the multi-sensor enhanced feature set of the multiple sensors.

[0008] In one embodiment, the first-layer dimensionality reduction processing of the reference feature set for each sensor based on manifold learning to obtain low-dimensional features for each sensor includes: establishing a nearest neighbor space based on the reference feature set; constructing a local weight matrix based on the nearest neighbor space; and embedding coordinate projection based on the local weight matrix to obtain the low-dimensional clustering space of the reference feature set.

[0009] In one embodiment, the second-layer dimensionality reduction processing of the multi-sensor feature set based on manifold learning to obtain the multi-sensor enhanced feature set of the plurality of sensors includes: calculating the multi-sensor enhanced feature set using a local preservation projection algorithm based on the multi-sensor feature set.

[0010] In one embodiment, determining the fault state of the mining equipment using a classifier based on the multi-sensor enhanced feature set includes: dividing the multi-sensor enhanced feature set into a test dataset and a trial dataset; inputting the test dataset into the classifier for training to construct a supervised classifier; and inputting the trial dataset into the constructed supervised classifier to determine the fault state of the mining equipment.

[0011] In one embodiment, the classifier is a KNN classifier.

[0012] In one embodiment, the method further includes noise reduction processing on the acquired health state vibration signal and the real-time working state vibration signal to reduce noise interference.

[0013] According to a second aspect of this application, a multi-sensor-based mining equipment fault status analysis device is provided. The device includes a processor and a memory, the memory storing computer program instructions. When the computer program instructions are executed by the processor, the multi-sensor-based mining equipment fault status analysis method according to the first aspect of this application is implemented.

[0014] The technical solution of this application has the following beneficial technical effects:

[0015] According to the technical solution of this application, a reference feature set of health and fault status of sensors at different measuring points of mining equipment is first constructed. Then, referenced manifold learning is performed on the reference feature set to initially distinguish different faults. Further, low-dimensional feature parameters from multiple sensors within the same time period are fused to construct a spatiotemporal multi-sensor pseudo-manifold network architecture, completing the dimensionality upscaling of the low-dimensional feature parameters. A secondary dimensionality reduction process is then performed on the dimensionality-upgraded manifold space to extract the enhanced low-dimensional feature parameters, and a classifier is used to achieve fault identification. This multi-sensor and multi-level enhanced deep learning method can more accurately identify different faults in mining equipment and improve the accuracy of fault analysis. Attached Figure Description

[0016] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of this application are illustrated by way of example and not limitation, and the same or corresponding reference numerals denote the same or corresponding parts, wherein:

[0017] Figure 1 This is a flowchart of a multi-sensor-based fault state analysis method for mining equipment according to an embodiment of this application;

[0018] Figure 2This is a schematic diagram of the structure of a multi-sensor-based fault status analysis device for mining equipment according to an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] It should be understood that when the terms "first," "second," etc., are used in the claims, description, and drawings of this application, they are only used to distinguish different objects and not to describe a specific order. The terms "comprising" and "including" used in the description and claims of this application indicate the presence of the described features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.

[0021] According to the first aspect of this application, this application provides a method for fault state analysis of mining equipment based on multiple sensors. Figure 1 This is a flowchart of a multi-sensor-based fault state analysis method 100 for mining equipment according to an embodiment of this application. Figure 1 As shown, the method includes steps S101 to S104, as detailed below:

[0022] S101, using multiple sensors arranged at different locations on the mining equipment, multiple health state vibration signals of the mining equipment in a healthy state and multiple real-time working state vibration signals of the mining equipment in a real-time working state are collected respectively.

[0023] Specifically, multiple sensors are arranged at different locations of the mining equipment, and the multiple sensors are used to collect multiple vibration signals of the mining equipment when it is in a healthy state and multiple vibration signals of the mining equipment when it is in a real-time working state.

[0024] In some embodiments, after acquiring the health state vibration signal and the real-time operating state vibration signal, the method further includes performing noise reduction processing on the acquired health state vibration signal and the real-time operating state vibration signal to reduce noise interference.

[0025] S102, process the health state vibration signal and the real-time working state vibration signal collected by each sensor, and construct a reference feature set corresponding to the sensor, wherein the reference feature set is formed by fusing the health state vibration signal feature set corresponding to the health state vibration signal and the real-time working state vibration signal feature set corresponding to the real-time working state vibration signal.

[0026] Specifically, the process of processing the health state vibration signal and the real-time operating state vibration signal collected by each sensor to construct the reference feature set corresponding to the sensor includes: analyzing the health state vibration signal and the real-time operating state vibration signal to obtain sensitive feature parameters of the health state vibration signal and the real-time operating state vibration signal; constructing the health state vibration signal feature set and the real-time operating state vibration signal feature set corresponding to each sensor based on the sensitive feature parameters of the health state vibration signal and the real-time operating state vibration signal; and fusing the health state vibration signal feature set and the real-time operating state vibration signal feature set corresponding to each sensor to construct the reference feature set corresponding to the sensor.

[0027] Further, the step of analyzing the healthy state vibration signal and the real-time operating state vibration signal to obtain the sensitive feature parameters of the healthy state vibration signal and the real-time operating state vibration signal includes: performing time-domain analysis and wavelet packet analysis on the healthy state vibration signal and the real-time operating state vibration signal to obtain the sensitive feature parameters of the healthy state vibration signal and the real-time operating state vibration signal, wherein the sensitive feature parameters include time-domain parameters and wavelet parameters.

[0028] As an example, firstly, m sets of vibration signals from n vibration sensors under k operating states are extracted, and feature parameters for each set of vibration signals are extracted according to Table 1. Then, feature parameter samples z for each sensor under different operating states are constructed. i , z i =[pk,st,me,va,Kr,L,C,S,E1,...E8] T i = 1, 2, 3... m, and the characteristic parameters are shown in Table 1.

[0029] Table 1 Feature Parameter Table

[0030]

[0031]

[0032] Then, a feature sample set for each operating state of a single sensor is established.

[0033] W n =[Z n,1 Z n,2 Z n,3 ,...,Z n,k ]

[0034] Where n represents the number of sensors.

[0035] Z n,k =[z1,z2,...,z m ]

[0036] Ultimately, each sensor forms a W n ∈R 16×(mk) The reference parameter sample set.

[0037] S103, process the multiple reference feature sets corresponding to the multiple sensors to obtain the multi-sensor enhanced feature set of the multiple sensors.

[0038] Specifically, the process of processing the multiple reference feature sets corresponding to the multiple sensors to obtain the multi-sensor enhanced feature set of the multiple sensors includes: performing a first-layer dimensionality reduction process on the reference feature set of each sensor based on manifold learning to obtain low-dimensional features of each sensor; fusing the low-dimensional features of the multiple sensors to obtain the multi-sensor feature set of the multiple sensors; and performing a second-layer dimensionality reduction process on the multi-sensor feature set based on manifold learning to obtain the multi-sensor enhanced feature set of the multiple sensors.

[0039] Furthermore, the first-layer dimensionality reduction processing of the reference feature set of each sensor based on manifold learning to obtain the low-dimensional features of each sensor includes: establishing a nearest neighbor space based on the reference feature set; constructing a local weight matrix based on the nearest neighbor space; and embedding coordinate projection based on the local weight matrix to obtain the low-dimensional clustering space of the reference feature set.

[0040] The step of performing a second-layer dimensionality reduction process on the multi-sensor feature set based on manifold learning to obtain the multi-sensor enhanced feature set of the multiple sensors includes: calculating the multi-sensor enhanced feature set using a local preservation projection algorithm based on the multi-sensor feature set.

[0041] In the example above, manifold learning is used to reduce the dimensionality of the referenced analysis sample set, extracting low-dimensional feature parameters of the vibration signal. Manifold learning is a unique machine learning method. The goal of the referenced manifold learning algorithm is to obtain the low-dimensional embedding subspace Y = {y1, y2, y3, ..., y...} for each sensor. i},y i∈R d×m The manifold learning process consists of three steps.

[0042] (1) Constructing a neighborhood space

[0043] Based on the high-dimensional features and each sample point n in N i Find the k nearest neighbors of each sample point using the Euclidean distance between them.

[0044] N i =knn(n i ,k),N i =[n 1i ,...n ik (1)

[0045] (2) Calculate local weights

[0046] By calculating the nonlinear relationship between each sample and its nearest neighbor subspace, the local error function ε(W) is minimized. This completes the construction of the local weight matrix W.

[0047]

[0048] Where: N is the high-dimensional feature set; N i For n i The k nearest neighbors are the nearest neighbor space constructed in Equation 1. ij For sample n i With sample n j The weight between them, if two are not neighbors, then w ji =0, and satisfies condition n i The weights of the nearest neighbor subspace satisfy:

[0049]

[0050] According to the expression in Equation 4.18, let:

[0051] S i =(NN) i ) T (NN i (4)

[0052] so:

[0053]

[0054] To obtain the optimal weight matrix, this application combines population dimensionality reduction analysis with the Lagrange multiplier method.

[0055]

[0056] Differentiating Equation 6 yields:

[0057]

[0058] so:

[0059]

[0060] Among them 1 k It is a k×1 column vector consisting entirely of 1s.

[0061] (3) Embedded coordinate projection

[0062] Embedded coordinate projection is essentially solving for the mapping in low-dimensional space. First, we define an n×n sparse matrix W to represent w.

[0063] therefore:

[0064] W = {w i =[w i1 ,w i2 ,...w in ]} T (9)

[0065] so

[0066]

[0067] The above calculations yield the following results:

[0068]

[0069] According to the matrix calculation formula:

[0070]

[0071] Simplifying Formula 12 from Formula 11, we get:

[0072] ψ(Y)=tr(Y(IW)(IW) T Y T )

[0073] =tr(YMY) T (13)

[0074] Where: M = (IW)(IW) T

[0075] Using the Lagrange multiplier method, we can simplify Formula 13 to obtain:

[0076] L(Y) = tr(YMY) T )+λ(YMY T -nI) (14)

[0077] Differentiation yields:

[0078]

[0079] therefore

[0080] M T Y T =-λY T (16)

[0081] Because: M T =M

[0082] so

[0083] MY T =-λY T (17)

[0084] Therefore Y T The matrix is ​​composed of eigenvectors of matrix M. To reduce the data dimensionality to d dimensions, we only need to obtain the eigenvectors corresponding to the d smallest non-zero eigenvalues ​​of M. In LLE analysis, the smallest eigenvalue is generally discarded because it is too close to 0. Therefore, we select eigenvectors with eigenvalues ​​in the range [2, ..., d+1] from smallest to largest. Through manifold learning, we will ultimately obtain the low-dimensional feature space LDM for each sensor. n ∈R (d×m) .

[0085] The first step of dimensionality reduction yields a low-dimensional mapping space for each sensor. However, a single low-dimensional feature extraction from a single sensor is insufficient to achieve state separation and recognition, and the performance varies between sensors. Therefore, this invention fuses the low-dimensional spaces of n sensors to form a higher-dimensional multi-sensor feature set, Mapping.

[0086] Mapping = [LDM1] T LDM2 T LDM3 T ,...,LDM n T ] T

[0087] This feature set is a spatially and temporally pseudomanifold composed of low-dimensional features from n sensors in k states at m different time points. Further dimensionality reduction requires preserving both the local spatial features of the manifold and the overall manifold characteristics across different states. Therefore, Locality Preserving Projection (LPP) is chosen for dimensionality reduction analysis of the feature set.

[142] .

[0088] Local Preserving Projection (LPP) is a manifold learning algorithm proposed by He based on the Laplacian Eigenmaps (LE) algorithm. This algorithm retains the advantages of both nonlinear and linear manifold learning, its core idea being an approximate linear learning of LE. LPP can extract the most discriminative features for dimensionality reduction, preserving local information and reducing multi-layered external interference factors in the signal. While essentially a linear dimensionality reduction method, it cleverly combines the nonlinear analysis techniques of the Laplacian Eigenmaps (LE) algorithm, preserving local nonlinear relationships when analyzing high-dimensional original data. Compared to other nonlinear dimensionality reduction methods, LPP can add new analysis samples within the original manifold, finding the corresponding mapping in the dimensionality-reduced subspace through computation. Therefore, LPP can evaluate and analyze new sample data. Other nonlinear methods can only define training data points and cannot evaluate new test data.

[0089] The LPP algorithm process is similar to other manifold learning algorithms, mainly consisting of three steps. [142,146,147] (1) Find the nearest neighbor; (2) Calculate the relation matrix; (3) Construct the objective function.

[0090] The derivation process of the LPP dimensionality reduction algorithm is shown below.

[0091] Suppose the original sample set is X = {x1, x2, x3...x} n}. Where the spatial dimension of X is D, x i ∈R D Let i = 0, 1, 2, ..., n. Also, assume Y = {y1, y2, y3...y}. n Y is a low-dimensional mapping of the original sample set onto the manifold space, where the spatial dimension of Y is d, and d << n.

[0092] When the LE algorithm finds a low-dimensional mapping, it needs to minimize the following objective function to ensure that the original nearest neighbor structure remains unchanged after the high-dimensional data points of the nearest neighbors are mapped to the low-dimensional space inherent in the manifold.

[0093]

[0094] Among them, W ij For the nearest neighbor graph G, x i With x j The edge weights between vertices. For a nearest neighbor graph G, if two vertices j and i are neighbors, then G... i,j =1, otherwise G i,j =0. The formula for calculating the edge weight of the nearest neighbor graph is as follows:

[0095]

[0096] Where t is a constant. The calculation is performed using heat kernel theory. The edge weight matrix can be simplified to:

[0097]

[0098] Transform Equation 19 using the Laplace transform:

[0099]

[0100] Among them, D ii =∑ j W ij Let L be the Laplacian matrix, and L = DW. Therefore, the objective function can be transformed into the following equation:

[0101]

[0102] Using the Lagrange algorithm, equation 22 can be transformed into a problem of solving the eigenvalues ​​of the following equation:

[0103] LY=λDY (23)

[0104] By solving Equation 23, the low-dimensional spatial characteristics of the original data can be obtained.

[0105] The LPP algorithm assumes that the manifold is linear, therefore

[0106] y i =K T x i (twenty four)

[0107] Substituting formula 23 into formula 21, we can obtain the following result:

[0108]

[0109] Similar to the LE transformation, the constraint Y in the Laplacian eigenmap... T When DY = 1, the transformation is performed using the LPP algorithm:

[0110] K T XDX T K = 1 (26)

[0111] Therefore, the objective function of LPP is:

[0112]

[0113] Therefore, the optimal mapping vector K of the LPP algorithm can be obtained by finding the eigenvalues ​​and eigenvectors of the following formula.

[0114] XLX T K = λDXT K (28)

[0115] LPP calculates the distance when finding nearest neighbors by directly calculating x. i Point and x i The method involves calculating the Euclidean distances between all points except the k closest points, sorting them, and then finding the k closest points. This method is relatively simple and practical.

[0116] The LPP algorithm is used to reduce the dimensionality of the multi-sensor feature set mapping after dimensionality enhancement. Further refinement and manifold representation between samples are then performed on the enhanced samples to obtain sensitive information from the sample data. This enables the enhancement of the differences between vibration signals and various faults.

[0117] S104, Based on the multi-sensor enhanced feature set, a classifier is used to determine the fault state of the mining equipment.

[0118] Specifically, determining the fault state of the mining equipment using a classifier based on the multi-sensor enhanced feature set includes: dividing the multi-sensor enhanced feature set into a test dataset and a trial dataset; inputting the test dataset into the classifier for training to construct a supervised classifier; and inputting the trial dataset into the constructed supervised classifier to determine the fault state of the mining equipment.

[0119] Specifically, in the first and second steps of processing the vibration signal, the data is divided into two sets: test data and experimental data. The experimental data is used for training the classifier, while the test data is used for classification and identification.

[0120] The results of enhanced feature analysis are input into the classifier for classification and recognition. Classification generally consists of two steps: the first step is to train and learn using known experimental data to build a supervised classifier; the second step is to use the classifier built in the previous step to classify and recognize unknown test data, thus completing the classification of the samples. The classifier works by learning classification rules using given categories and known training data, and then classifying unknown data.

[0121] As an example, KNN classification is a theoretically mature method. This method is simple and intuitive, and relatively easy to implement. The basic idea is to select the k nearest neighbors of the sample to be classified based on the distance between the sample to be classified and each training sample, and finally determine the category of the sample based on these nearest neighbors. The calculation steps of KNN are: (1) Calculate the distance between the object to be classified and each object in the training set; (2) Identify the k nearest objects as the nearest neighbors of the test object; (3) Classify the test object based on the attribute affiliation of these nearest neighbors. The most critical issue in the KNN classification method is the calculation of distance.

[0122] For example, the distance calculation method could be the Manhattan distance method.

[0123]

[0124] The above specific embodiments illustrate the multi-sensor-based fault state analysis method for mining equipment of this application. According to the technical solution of this application, firstly, a reference feature set of health and fault states of sensors at different measuring points of the mining equipment is constructed. Then, referenced manifold learning is performed on the reference feature set to initially distinguish different faults. Further, low-dimensional feature parameters from multiple sensors within the same time period are fused to construct a spatiotemporal multi-sensor pseudo-manifold network architecture, completing the dimensionality increase of the low-dimensional feature parameters. A secondary dimensionality reduction process is then performed on the dimensionality-increased manifold space to extract the enhanced low-dimensional feature parameters, and a classifier is used to achieve fault identification. This multi-sensor and multi-level enhanced deep learning method can more accurately identify different faults in mining equipment and improve the accuracy of fault analysis for mining equipment.

[0125] According to a second aspect of this application, this application also provides a multi-sensor-based fault status analysis device for mining equipment.

[0126] Figure 2 This is a schematic diagram of the structure of a multi-sensor-based mining equipment fault state analysis 20 according to an embodiment of this application. The device includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the multi-sensor-based mining equipment fault state analysis method according to the first aspect of this application. The device also includes other components well-known to those skilled in the art, such as a communication bus and a communication interface. Their configuration and functions are known in the art and will not be described in detail here.

[0127] In this application, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this application can be implemented using computer-readable / executable instructions that can be stored or otherwise retained by such a computer-readable medium.

[0128] Based on the above description in this specification, those skilled in the art will also understand that the terms used, such as "upper" and "lower," which indicate orientation or positional relationship, are based on the orientation or positional relationship shown in the accompanying drawings of this specification. They are only for the purpose of facilitating the explanation of the present application and simplifying the description, and do not imply that the device or element involved must have the specific orientation, or be constructed and operated in a specific orientation. Therefore, the above-mentioned orientation or positional relationship terms should not be understood or interpreted as a limitation on the present application.

[0129] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0130] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for fault state analysis of mining equipment based on multiple sensors, characterized in that, include: Multiple sensors arranged at different locations on the mining equipment are used to collect multiple vibration signals of the mining equipment in a healthy state and multiple vibration signals of the mining equipment in a real-time operating state. The health state vibration signal and the real-time working state vibration signal collected by each sensor are processed to construct a reference feature set corresponding to the sensor, wherein the reference feature set is formed by fusing the health state vibration signal feature set corresponding to the health state vibration signal and the real-time working state vibration signal feature set corresponding to the real-time working state vibration signal. Process the multiple reference feature sets corresponding to the multiple sensors to obtain the multi-sensor enhanced feature set of the multiple sensors; Based on the multi-sensor enhanced feature set, a classifier is used to determine the fault state of the mining equipment.

2. The method for fault state analysis of mining equipment based on multiple sensors according to claim 1, characterized in that, The process of processing the health state vibration signal and the real-time operating state vibration signal collected by each sensor to construct the reference feature set corresponding to the sensor includes: The vibration signal in the healthy state and the vibration signal in the real-time working state are analyzed to obtain the sensitive characteristic parameters of the vibration signal in the healthy state and the vibration signal in the real-time working state. Based on the sensitive characteristic parameters of the healthy state vibration signal and the sensitive characteristic parameters of the real-time working state vibration signal, construct the healthy state vibration signal feature set and the real-time working state vibration signal feature set corresponding to each sensor; The health state vibration signal feature set and the real-time working state vibration signal feature set corresponding to each sensor are fused to construct the reference feature set corresponding to the sensor.

3. The method for fault status analysis of mining equipment based on multiple sensors according to claim 2, characterized in that, The step of analyzing the healthy state vibration signal and the real-time working state vibration signal to obtain the sensitive feature parameters of the healthy state vibration signal and the real-time working state vibration signal includes: performing time-domain analysis and wavelet packet analysis on the healthy state vibration signal and the real-time working state vibration signal to obtain the sensitive feature parameters of the healthy state vibration signal and the real-time working state vibration signal, wherein the sensitive feature parameters include time-domain parameters and wavelet parameters.

4. The fault status analysis method for mining equipment based on multiple sensors according to claim 3, characterized in that, The step of processing the multiple reference feature sets corresponding to the multiple sensors to obtain the multi-sensor enhanced feature set of the multiple sensors includes: Based on manifold learning, the reference feature set of each sensor is subjected to a first-level dimensionality reduction process to obtain low-dimensional features of each sensor. By fusing the low-dimensional features of the multiple sensors, a multi-sensor feature set of the multiple sensors is obtained; The multi-sensor feature set is subjected to a second-layer dimensionality reduction process based on manifold learning to obtain the multi-sensor enhanced feature set of the multiple sensors.

5. The fault status analysis method for mining equipment based on multiple sensors according to claim 4, characterized in that, The first-level dimensionality reduction process based on manifold learning for the reference feature set of each sensor to obtain low-dimensional features for each sensor includes: A nearest neighbor space is established based on the reference feature set; Based on the nearest neighbor space, a local weight matrix is ​​constructed; Based on the local weight matrix, the coordinate projection is embedded to obtain the low-dimensional clustering space of the reference feature set.

6. The fault status analysis method for mining equipment based on multiple sensors according to claim 4, characterized in that, The step of performing a second-layer dimensionality reduction process on the multi-sensor feature set based on manifold learning to obtain the multi-sensor enhanced feature set of the multiple sensors includes: calculating the multi-sensor enhanced feature set using a local preservation projection algorithm based on the multi-sensor feature set.

7. The method for fault status analysis of mining equipment based on multiple sensors according to claim 1, characterized in that, Determining the fault state of the mining equipment using a classifier based on the multi-sensor enhanced feature set includes: The multi-sensor enhanced feature set is divided into an experimental dataset and a test dataset; The test dataset is input into the classifier for training to build a tutored classifier; The test dataset is input into the constructed supervised classifier to determine the fault state of the mining equipment.

8. The method for fault state analysis of mining equipment based on multiple sensors according to claim 7, characterized in that, The classifier is the KNN classifier.

9. The method for fault state analysis of mining equipment based on multiple sensors according to claim 8, characterized in that, The method further includes noise reduction processing of the collected health state vibration signal and the real-time working state vibration signal to reduce noise interference.

10. A multi-sensor-based fault status analysis device for mining equipment, characterized in that, The device includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the multi-sensor-based fault state analysis method for mining equipment according to any one of claims 1 to 9.

Images