A mine machinery diagnosis method and system

Abstract

A mine machinery diagnosis method comprises detecting sensor vibration data, judging whether the accumulated detected vibration data exceeds a set time length; if not, executing a first algorithm to analyze equipment bearing failure; if yes, judging whether a second analysis is performed for the sensor; if the second analysis is performed, classifying and analyzing current vibration data by using analyzed centroid point coordinates; if the second analysis is not performed, executing the second analysis to obtain centroid point coordinates; the second analysis is performed by calculating characteristic parameters of each cluster centroid point coordinate and current vibration data to classify current vibration data, the second analysis is performed at intervals to perform clustering analysis and update centroid point coordinates; judging the equipment running state at this time, if the equipment is in a running state, executing the first algorithm to analyze equipment bearing failure, if the equipment is in a static state, not executing the first algorithm to analyze; the above diagnosis method and system improve the overall operation efficiency of the system by using the second analysis algorithm to identify the start-stop state of the equipment.

Description

Technical Field

[0001] This invention relates to fault diagnosis of mining machinery, and particularly to a method and system for diagnosing mining machinery. Background Technology

[0002] Most systems on the market that claim to diagnose mechanical faults do not take into account the importance of real-time transmission and the limited computing resources in practical applications. To achieve these two goals and make the system truly usable, in addition to meticulously optimizing the hyperparameters of various algorithms, it is even more important to deeply integrate with the business environment.

[0003] Factors affecting the operational load and stability of a mechanical diagnostic system include: the amount of data collected at a single point, the number of data points, server-side hardware, and system optimization. Furthermore, the execution time of various algorithms is quite long, especially under single-machine operation, which severely tests the system's operational capacity. Summary of the Invention

[0004] Therefore, it is necessary to provide a diagnostic method for mining machinery that can improve efficiency.

[0005] At the same time, a mining machinery diagnostic system that improves efficiency is provided.

[0006] A method for diagnosing mining machinery, comprising:

[0007] Detection duration determination: Collect vibration data from the sensor and determine whether the cumulative collected vibration data exceeds the set duration.

[0008] Execute the first algorithm: If the result is negative, then execute the first algorithm to analyze the equipment bearing fault;

[0009] Second analysis and judgment: If the judgment is yes, then determine whether to perform a second analysis for the sensor;

[0010] Classification Analysis: If the second analysis has already been performed, the current vibration data is classified using the analyzed centroid coordinates; if the second analysis has not been performed, it is executed to obtain the centroid coordinates.

[0011] Update centroid points: The second analysis has been performed. The current vibration data is classified by calculating the coordinates of the centroid points of each cluster and the characteristic parameters of the current vibration data. The second analysis is performed at set intervals to perform cluster analysis and update the centroid point coordinates.

[0012] Status determination: Based on the second analysis, the operating status of the equipment is determined using the coordinates of the centroid and the characteristic parameters of the current vibration data. If the equipment is in operation, the first algorithm is executed to analyze the bearing fault. If the equipment is stationary, the first algorithm is not executed.

[0013] The second analysis includes:

[0014] Division: Pre-observe a sample of N data points and divide the N objects into K classes;

[0015] Selecting initial center points: K points are randomly selected as the initial cluster centers, i.e., the initial center points, according to the classification.

[0016] Classification: According to the principle of "minimum distance from the initial center point", all observations are assigned to the class of each center point, and the number of initial center points is the same as the K value;

[0017] Update centroids: Calculate the mean of all sample points in each class to calculate the centroid of each cluster as the new centroid, and update the centroids.

[0018] Output: Repeat the classification and center point update steps based on the obtained center point positions until the center point positions no longer change or the specified number of iterations is reached. Output the classification results of each point and the coordinates of K centroids.

[0019] The partitioning module further includes: selecting kurtosis, peak value, and root mean square value as input indicators for the second analysis; each sampling yields a three-dimensional coordinate point composed of kurtosis, peak value, and root mean square value;

[0020] The centroid of each cluster is the average of all data points in that cluster. The coordinates of the centroid of a given cluster are (x...). c ,y c ,z c Then we have:

[0021]

[0022] Where N is the number of data points in the cluster, x n and y n z represents the x-coordinate and y-coordinate of the nth point in the cluster, respectively. n This represents the vertical coordinate of the nth point in the cluster.

[0023] In a preferred embodiment, the classification step involves assigning data points to the nearest cluster center.

[0024] In a preferred embodiment, the division step divides the target state of the device into two states: running and stopped, resulting in a K value of 2.

[0025] In a preferred embodiment, the partitioning step further includes: selecting kurtosis, peak value, and root mean square value as input indicators; selecting all reported data from the sensor within M hours to obtain N sets of time-domain statistical indicators of sensor acceleration; and using the N sets of indicators as input for the second analysis.

[0026] Input=[[beta1,amp1,rms1],[beta2,amp2,rms2],…,[beta N ,amp N ,rms N ]],

[0027] The vibration data includes: vibration frequency.

[0028] In a preferred embodiment, the centroid of each cluster is the average value of all data points in that cluster, and the coordinates of the centroid of a cluster are (x... c ,y c ,z c Then we have:

[0029]

[0030] Where N is the number of all data points in the cluster, x n and y n z represents the x-coordinate and y-coordinate of the nth point in the cluster, respectively. n This represents the vertical coordinate of the nth point in the cluster;

[0031] In a preferred embodiment, in the state determination step: each sampling yields a three-dimensional coordinate point composed of kurtosis, peak value, and root mean square value. The Euclidean distance from this point to the centroid of the running state and the centroid of the stationary state are calculated respectively. The closer distance indicates the current operating state of the sensor. The coordinates of the centroid of the running state are PA:(x a ,y a ,z a The coordinates of the centroid at rest are PS:(x s ,y s ,z s The coordinates of the target point are P-1:(x1,y1,z1), and the Euclidean distance from the target point to the centroid of the running state is d. a The Euclidean distance from the target point to the stationary centroid is d. s Then we have:

[0032]

[0033] In a preferred embodiment, the first algorithm analysis includes:

[0034] High-frequency filtering: Performing high-frequency filtering on the original signal to remove low-frequency signal components.

[0035] Decomposition: EMD mode decomposition, the mode with the highest similarity to the original signal after decomposition is taken as the decomposed signal;

[0036] Low-frequency filtering: Low-frequency filtering removes high-frequency components from the decomposed signal;

[0037] Find the envelope signal: Perform a Hilbert transform on the decomposed low-frequency signal to obtain the envelope signal;

[0038] Obtain the envelope spectrum signal: Perform a Fourier transform on the envelope signal to obtain the envelope spectrum signal.

[0039] A mining machinery diagnostic system, comprising:

[0040] Detection duration judgment module: Detects vibration data from the sensor and determines whether the cumulative detected vibration data exceeds a set duration.

[0041] Execute the first algorithm module: If the determination is negative, then execute the first algorithm to analyze the equipment bearing fault;

[0042] Second analysis and judgment module: If the judgment is yes, then determine whether to perform a second analysis for the sensor;

[0043] Classification Analysis Module: If the second analysis has already been performed, the current vibration data is classified using the analyzed centroid coordinates; if the second analysis has not been performed, it is executed to obtain the centroid coordinates.

[0044] Update centroid module: The second analysis has been performed. The current vibration data is classified by calculating the coordinates of the centroids of each cluster and the characteristic parameters of the current vibration data. The second analysis is performed at set intervals to perform cluster analysis and update the centroid coordinates.

[0045] Status judgment module: Based on the second analysis, the system uses the coordinates of the centroid and the characteristic parameters of the current vibration data to determine the current operating status of the equipment. If the equipment is in operation, the first algorithm is executed to analyze the bearing fault. If the equipment is stationary, the first algorithm is not executed.

[0046] The second analysis includes:

[0047] The partitioning module pre-observes a sample of N data points and divides the N objects into K classes using a locally optimal approach.

[0048] Initial center point selection module: K points are randomly selected as the initial cluster centers, i.e., the initial center points, based on the classification.

[0049] Classification module: According to the principle of "minimum distance from the initial center point", all observations are assigned to the class of each center point. The number of initial center points is the same as the K value.

[0050] Update centroid module: Calculate the mean of all sample points in each class to calculate the centroid of each cluster as the new centroid, and update the centroid.

[0051] Output module: Based on the obtained center point position, repeat the classification step and the center point update step until the center point position no longer changes or the specified number of iterations is reached, and output the classification result of each point and the coordinates of K centroids;

[0052] The partitioning module further includes: selecting kurtosis, peak value, and root mean square value as input indicators for the second analysis; each sampling yields a three-dimensional coordinate point composed of kurtosis, peak value, and root mean square value;

[0053] The centroid of each cluster is the average of all data points in that cluster. The coordinates of the centroid of a given cluster are (x...). c ,y c ,z c Then we have:

[0054]

[0055] Where N is the number of data points in the cluster, x n and y n z represents the x-coordinate and y-coordinate of the nth point in the cluster, respectively. n This represents the vertical coordinate of the nth point in the cluster.

[0056] In a preferred embodiment, in the state determination module: each sampling obtains a three-dimensional coordinate point composed of kurtosis, peak value, and root mean square value; the Euclidean distance from this point to the centroid of the running state and the centroid of the stationary state are calculated respectively; the closer distance indicates the current working state of the sensor. The coordinates of the centroid of the running state are PA:(x a ,y a ,z a The coordinates of the centroid at rest are PS:(x s ,y s ,z s The coordinates of the target point are P-1:(x1,y1,z1), and the Euclidean distance from the target point to the centroid of the running state is d. a The Euclidean distance from the target point to the stationary centroid is d. s Then we have:

[0057]

[0058] In the classification module, data points are assigned to the nearest cluster center;

[0059] In the partitioning module, the target state of the device is divided into two states: running and stopped, so the K value is 2;

[0060] The partitioning module further includes: selecting kurtosis, peak value, and root mean square value as input indicators; selecting all reported data from the sensor within M hours to obtain N sets of time-domain statistical indicators of sensor acceleration; and using these N sets of indicators as input for the second analysis.

[0061] Input=[[beta1,amp1,rms1],[beta2,amp2,rms2],…,[beta N ,amp N ,rms N ]];

[0062] The centroid of each cluster is the average value of all data points in that cluster. The coordinates of the centroid of a cluster are (x...). c ,y c ,z c Then we have:

[0063]

[0064] Where N is the number of data points in the cluster, x n and y n z represents the x-coordinate and y-coordinate of the nth point in the cluster, respectively. n The vertical axis coordinate represents the nth point in the cluster, and the vibration data includes the vibration frequency.

[0065] In a preferred embodiment, the first algorithm analysis includes:

[0066] High-frequency filtering module: Performs high-frequency filtering on the original signal, removing low-frequency signal components from the original signal.

[0067] Decomposition module: EMD mode decomposition, taking the mode with the highest similarity to the original signal after decomposition as the decomposed signal;

[0068] Low-frequency filtering module: Low-frequency filtering removes high-frequency components from the decomposed signal;

[0069] Envelope signal calculation module: Performs Hilbert transform on the decomposed low-frequency signal to obtain the envelope signal;

[0070] Envelope spectrum signal acquisition module: Perform Fourier transform on the envelope signal to obtain the envelope spectrum signal.

[0071] The above-mentioned mining machinery diagnostic method and system improves the overall operating efficiency of the system by using a second analysis algorithm to identify the start-up and shutdown status of equipment. The second analysis algorithm identifies the stop and start status of different equipment, and then no complex algorithm or the first algorithm is used for equipment in the stop state, thereby reducing the system load. Attached Figure Description

[0072] Figure 1 This is a flowchart of a mining machinery diagnostic method according to an embodiment of the present invention;

[0073] Figure 2 This is a flowchart of the first algorithm analysis according to an embodiment of the present invention;

[0074] Figure 3 This diagram illustrates the effect of the second analysis method according to an embodiment of the present invention. Detailed Implementation

[0075] The following examples are provided to help better understand the present invention, but are not intended to limit the invention.

[0076] like Figures 1 to 2 As shown, a method for diagnosing mining machinery according to the present invention includes:

[0077] Step S201, detection duration judgment: Collect sensor vibration data and determine whether the cumulative collected vibration data exceeds the set duration, such as 7 days;

[0078] Step S203: Execute the first algorithm: If the determination is negative, then execute the first algorithm to analyze the equipment bearing fault;

[0079] Step S205, Second analysis and judgment: If the judgment is yes, then determine whether to perform a second analysis for the sensor;

[0080] Step S207, Classification Analysis: If the second analysis has been performed, the current vibration data is classified and analyzed using the analyzed centroid coordinates; if the second analysis has not been performed, the second analysis is executed to obtain the centroid coordinates.

[0081] Step S209, update centroid points: The second analysis has been performed. The current vibration data is classified by calculating the coordinates of the centroid points of each cluster and the characteristic parameters of the current vibration data. The second analysis is performed once at a set time interval to perform cluster analysis and update the coordinates of the centroid points.

[0082] Step S211, Status Judgment: Determine the current operating status of the equipment. If it is in the operating state, execute the first algorithm to analyze the bearing fault of the equipment. If it is in the stationary state, do not execute the first algorithm analysis.

[0083] The second analysis includes:

[0084] Division: Pre-observe a sample of N data points and divide the N objects into K classes;

[0085] Selecting initial center points: K points are randomly selected as the initial cluster centers, i.e., the initial center points, according to the classification.

[0086] Classification: According to the principle of "minimum distance from the initial center point", all observations are assigned to the class of each center point, and the number of initial center points is the same as the K value;

[0087] Update centroids: Calculate the mean of all sample points in each class to calculate the centroid of each cluster as the new centroid, and update the centroids.

[0088] Output: Based on the obtained center point position, repeat the classification step and the center point update step until the center point position no longer changes or the specified number of iterations is reached, and output the classification result of each point and the coordinates of K centroids.

[0089] Furthermore, in the classification step of this embodiment, data points are assigned to the nearest cluster center.

[0090] Furthermore, in the division step of this embodiment, the target state of the device is divided into two states: running and stopped, so the value of K is 2.

[0091] Furthermore, the partitioning step in this embodiment also includes: selecting kurtosis, peak value, and root mean square value as input indicators; selecting all reported data from the sensor within M hours to obtain N sets of sensor acceleration time-domain statistical indicators; and using these N sets of indicators as input for the second analysis.

[0092] Input=[[beta1,amp1,rms1],[beta2,amp2,rms2],…,[beta N ,amp N ,rms N ]).

[0093] Furthermore, in this embodiment, the centroid of each cluster is the average value of all data points in that cluster, and the coordinates of the centroid of a certain cluster are (x... c ,y c ,z c Then we have:

[0094]

[0095] Where N is the number of all data points in the cluster, x n and y n These represent the x-coordinate and y-coordinate of the nth point in the cluster, respectively.

[0096] Furthermore, in the state determination step of this embodiment: each sampling yields a three-dimensional coordinate point composed of kurtosis, peak value, and root mean square value. The Euclidean distance from this point to the centroid of the running state and the centroid of the stationary state are calculated respectively. The closer distance indicates the current working state of the sensor. The coordinates of the centroid of the running state are PA:(x a ,y a,z a The coordinates of the centroid at rest are PS:(x s ,y s ,z s The coordinates of the target point are P-1:(x1,y1,z1), and the Euclidean distance from the target point to the centroid of the running state is d. a The Euclidean distance from the target point to the stationary centroid is d. s Then we have:

[0097]

[0098] Furthermore, the first algorithm analysis in this embodiment includes:

[0099] Step S301, High-frequency filtering: Perform high-frequency filtering on the original signal to remove low-frequency signal components.

[0100] Step S303, Decomposition: EMD mode decomposition, taking the mode with the highest similarity to the original signal after decomposition as the decomposed signal;

[0101] Step S305, Low-frequency filtering: Low-frequency filtering removes high-frequency components from the decomposed signal;

[0102] Step S307, Calculate the envelope signal: Perform Hilbert transform on the decomposed low-frequency signal to obtain the envelope signal;

[0103] Step S309, Obtain the envelope spectrum signal: Perform a Fourier transform on the envelope signal to obtain the envelope spectrum signal.

[0104] A mining machinery diagnostic system, comprising:

[0105] Detection duration judgment module: Collects sensor vibration data and determines whether the cumulative collected vibration data exceeds the set duration;

[0106] Execute the first algorithm module: If the determination is negative, then execute the first algorithm to analyze the equipment bearing fault;

[0107] Second analysis and judgment module: If the judgment is yes, then determine whether to perform a second analysis for the sensor;

[0108] Classification Analysis Module: If the second analysis has been performed, the current vibration data is classified and analyzed using the analyzed centroid coordinates; if the second analysis has not been performed, the second analysis is executed to obtain the centroid coordinates.

[0109] Update centroid module: The second analysis has been performed. The current vibration data is classified by calculating the coordinates of the centroids of each cluster and the characteristic parameters of the current vibration data. The second analysis is performed at set intervals to perform cluster analysis and update the centroid coordinates.

[0110] Status determination module: Determines the current operating status of the equipment. If it is in the operating state, the first algorithm is executed to analyze the bearing fault of the equipment. If it is in the stationary state, the first algorithm is not executed.

[0111] Furthermore, the second analysis in this embodiment includes:

[0112] The partitioning module pre-observes a sample of N data points and divides the N objects into K classes using a locally optimal approach.

[0113] Initial center point selection module: K points are randomly selected as the initial cluster centers, i.e., the initial center points, based on the classification.

[0114] Classification module: According to the principle of "minimum distance from the initial center point", all observations are assigned to the class of each center point. The number of initial center points is the same as the K value.

[0115] Update centroid module: Calculate the mean of all sample points in each class to calculate the centroid of each cluster as the new centroid, and update the centroid.

[0116] Output module: Based on the obtained center point position, repeat the classification step and the center point update step until the center point position no longer changes or the specified number of iterations is reached, and output the classification result of each point and the coordinates of K centroids.

[0117] Furthermore, in the state determination module of this embodiment: each sampling yields a three-dimensional coordinate point composed of kurtosis, peak value, and root mean square value. The Euclidean distance from this point to the centroid of the running state and the Euclidean distance to the centroid of the stationary state are calculated respectively. The closer distance indicates the current working state of the sensor. The coordinates of the centroid of the running state are PA:(x a ,y a ,z a The coordinates of the centroid at rest are PS:(x s ,y s ,z s The coordinates of the target point are P-1:(x1,y1,z1), and the Euclidean distance from the target point to the centroid of the running state is d. a The Euclidean distance from the target point to the stationary centroid is d. s Then we have:

[0118]

[0119] Furthermore, in the classification module of this embodiment, data points are grouped to the nearest cluster center.

[0120] Furthermore, in the partitioning module of this embodiment, the target state of the device is divided into two states: running and stopped, thus the K value is 2.

[0121] Furthermore, the partitioning module in this embodiment also includes: selecting kurtosis, peak value, and root mean square value as input indicators; selecting all reported data from the sensor within M hours to obtain N sets of sensor acceleration time-domain statistical indicators; and using these N sets of indicators as input for the second analysis.

[0122] Input=[[beta1,amp1,rms1],[beta2,amp2,rms2],…,[beta N ,amp N ,rms N ]];

[0123] The centroid of each cluster is the average value of all data points in that cluster. The coordinates of the centroid of a cluster are (x...). c ,y c ,z c Then we have:

[0124]

[0125] Where N is the number of data points in the cluster, x n and y n z represents the x-coordinate and y-coordinate of the nth point in the cluster, respectively. n This represents the vertical coordinate of the nth point in the cluster.

[0126] The first algorithm analysis includes:

[0127] High-frequency filtering module: performs high-frequency filtering on the original signal, filtering out low-frequency signal components in the original signal;

[0128] Decomposition module: EMD (Empirical Mode Decomposition) mode decomposition, which takes the mode with the highest similarity to the original signal after decomposition as the decomposed signal;

[0129] Low-frequency filtering module: Low-frequency filtering removes high-frequency components from the decomposed signal;

[0130] Envelope signal finding module: Performs Hilbert transform on the decomposed low-frequency signal to obtain the envelope signal;

[0131] Envelope spectrum signal acquisition module: Perform Fourier transform on the envelope signal to obtain the envelope spectrum signal.

[0132] The sensor samples acceleration data every ten minutes. Each sampling yields 32,000 points. Various time-domain statistical indicators (as shown in Table 1) can be calculated from these 32,000 points. This invention selects kurtosis (beta), peak value (amp), and root mean square value (rms) as input indicators.

[0133] Table 1

[0134]

[0135] Selecting all sensor-reported data within M hours will yield N sets of sensor acceleration time-domain statistical indices. These N sets of indices are used as input to a clustering algorithm, as expressed by the following formula (1):

[0136] Input=[[beta1,amp1,rms1],[beta2,amp2,rms2],…,[beta N ,amp N ,rms N (1).

[0137] The target states are running and non-running, so we set one of the initial centers to the mean center of the input, which represents "the centroid of the device if it is always running during this period"; the other initial center is set to the zero point [0,0,0], which represents "the centroid of the device if it is always stationary during this period".

[0138] The classification result of this invention is the classification result for each point. The classification method is that each point is assigned to the cluster that is closest to it. The means of defining the closest distance is to determine the Euclidean distance from the point to the centroid of each cluster.

[0139] The principle of the second analysis algorithm is to classify each data point into the cluster corresponding to the nearest initial centroid. That is, calculate the distance of each data point to each initial centroid, and then classify the data point into the cluster corresponding to the nearest initial centroid.

[0140] In the second analysis algorithm, K refers to the number of clusters to be created. Therefore, the number of initial centroids is the same as K because each initial centroid will become the center of a cluster.

[0141] In the second analysis algorithm, N represents the total number of samples, and K in the problem represents the number of classes that all samples need to be divided into. If N samples are divided into two classes, the value of K is 2; if N samples need to be divided into three classes, the value of K is 3.

[0142] The second analysis algorithm is a clustering algorithm that iteratively finds local optima for k clusters. First, the algorithm randomly selects K points as initial cluster centers, and then assigns all data points to the nearest cluster center. Next, the algorithm calculates the centroid of each cluster (i.e., the average value or other characteristic of all data points within the cluster) and updates the center of each cluster to its centroid. Then, the algorithm again assigns all data points to the nearest cluster center and repeats the above steps until convergence (convergence: meaning the cluster center does not move in the next iteration or the movement is less than a very small threshold). In short, the second analysis algorithm determines the local optimum of each cluster through iterative processing. It determines the centroid of the cluster by calculating the average value of all data points within the cluster and updates the center of the cluster to its centroid. Then, the algorithm again assigns all data points to the nearest cluster center and repeats the above steps until convergence.

[0143] The method of this patent invention significantly optimizes the system's data processing time, improves the utilization efficiency of the server's CPU and memory, and greatly saves server computing resources.

[0144] Based on existing wireless sensor standards, with a sampling frequency of 32000Hz and 32000 data points, running the envelope spectrum algorithm 10000 times in a MATLAB environment, the average time per run after weighting is 0.127986s. However, in practice, the number of monitoring points is typically around 1500, and the data reporting cycle is 10 minutes. Therefore, it can be seen that executing just one bearing fault diagnosis algorithm accounts for approximately 32% of the system's average time.

[0145] like Figure 3 As shown in the diagram, the effect of the second analysis algorithm is illustrated. Orange dots represent data when the device is stationary, while blue dots represent data when the device is running.

[0146] Bearing failure characteristic frequencies refer to the specific frequencies exhibited by a bearing when it fails. These characteristic frequencies are determined by the bearing's structure and materials and can be used to diagnose whether the bearing is in a faulty state. For example, when a bearing experiences wear or cracks, it will generate specific frequencies, which can be detected using vibration analysis algorithms.

[0147] In practical applications, some devices are not always running, and real-time monitoring is unnecessary when devices are not running. Therefore, to address the issue of excessive system downtime, this invention proposes an algorithm that uses a second analysis algorithm to identify the start / stop status of devices, thereby improving the overall operating efficiency of the system.

[0148] This invention, taking into account practical situations, uses artificial intelligence to determine the operating status of equipment. When the equipment is stationary, various diagnostic algorithms no longer need to be computed, thus significantly reducing system load. This invention optimizes the diagnostic system, reduces the waste of computing resources, and improves system efficiency.

[0149] This invention has been applied in coal mines and has successfully detected dozens of alarms during production. During observation, the peak CPU utilization of the fault diagnosis system was 6.3%, and the average CPU utilization over 24 hours was 0.2%.

[0150] Based on the above-described preferred embodiments according to this application, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the technical concept of this application. The technical scope of this application is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for diagnosing mining machinery, characterized in that, include: Detection duration determination: Collect sensor vibration data and determine whether the cumulative collected vibration data exceeds the set duration; Execute the first algorithm: If the result is negative, then execute the first algorithm to analyze the equipment bearing fault; Second analysis and judgment: If the judgment is yes, then determine whether to perform a second analysis for the sensor; Classification Analysis: If the second analysis has already been performed, use the analyzed centroid coordinates to perform classification analysis on the current vibration data; If the second analysis is not performed, then the second analysis will be performed to obtain the coordinates of the centroid. Update centroid points: The second analysis has been performed. The current vibration data is classified by calculating the coordinates of the centroid points of each cluster and the characteristic parameters of the current vibration data. The second analysis is performed at set intervals to perform cluster analysis and update the centroid point coordinates. Status judgment: Based on the second analysis, the operating status of the equipment is determined using the coordinates of the centroid and the characteristic parameters of the current vibration data. If the equipment is in operation, the first algorithm is executed to analyze the bearing fault. If the equipment is stationary, the first algorithm is not executed. The second analysis includes: Division: Pre-observe a sample of N data points and divide the N objects into K classes; Selecting initial center points: K points are randomly selected as initial cluster centers, i.e., initial center points, according to the classification; Classification: According to the principle of "minimum distance from the initial center point", all observations are classified into the class to which each center point belongs, and the number of initial center points is the same as the K value; Update centroids: Calculate the mean of all sample points in each class to calculate the centroid of each cluster as the new centroid, and update the centroids. Output: Repeat the classification and center point update steps based on the obtained center point positions until the center point positions no longer change or the specified number of iterations is reached. Output the classification results of each point and the coordinates of K centroids. The partitioning step further includes: selecting kurtosis, peak value, and root mean square value as input indicators for the second analysis; and obtaining a three-dimensional coordinate point composed of kurtosis, peak value, and root mean square value for each sampling. The centroid of each cluster is the average of all data points in that cluster. The coordinates of the centroid of a given cluster are (x...). c ,y c ,z c Then we have: Where N is the number of all data points in the cluster, x n and y n z represents the x-coordinate and y-coordinate of the nth point in the cluster, respectively. n This represents the vertical coordinate of the nth point in the cluster.

2. The method for diagnosing mining machinery according to claim 1, characterized in that, In the classification step, data points are assigned to the nearest cluster center.

3. The method for diagnosing mining machinery according to claim 1, characterized in that, In the division step, the target state of the device is divided into two states: running and stopped, so the value of K is 2.

4. The method for diagnosing mining machinery according to any one of claims 1 to 3, characterized in that, The partitioning step further includes: selecting all reported data from the sensors within M hours to obtain N sets of time-domain statistical indices for sensor acceleration, and using these N sets of indices as input for the second analysis. Input=[[beta1,amp1,rms1],[beta2,amp2,rms2],…,[beta N ,amp N ,rms N ]], The vibration data includes: vibration frequency.

5. The method for diagnosing mining machinery according to any one of claims 1 to 3, characterized in that, In the state determination step: each sampling yields a three-dimensional coordinate point composed of kurtosis, peak value, and root mean square value. The Euclidean distance from this point to the centroid of the running state and the centroid of the stationary state are calculated respectively. The closer distance indicates the sensor's current operating state. The coordinates of the centroid of the running state are PA:(x a ,y a ,z a The coordinates of the stationary centroid are PS:(x s ,y s ,z s The coordinates of the target point are P-1:(x1,y1,z1), and the Euclidean distance from the target point to the centroid of the running state is d. a The Euclidean distance from the target point to the stationary centroid is d. s Then we have:

6. The method for diagnosing mining machinery according to any one of claims 1 to 3, characterized in that, The first algorithm analysis includes: High-frequency filtering: High-frequency filtering is applied to the original signal to remove low-frequency signal components. Decomposition: EMD mode decomposition, taking the mode with the highest similarity to the original signal after decomposition as the decomposed signal. Low-frequency filtering: Low-frequency filtering removes high-frequency components from the decomposed signal; Find the envelope signal: Perform a Hilbert transform on the decomposed low-frequency signal to obtain the envelope signal; Obtain the envelope spectrum signal: Perform a Fourier transform on the envelope signal to obtain the envelope spectrum signal.

7. A diagnostic system for mining machinery, characterized in that, include: Detection duration judgment module: Collects sensor vibration data and determines whether the cumulative collected vibration data exceeds the set duration; Execute the first algorithm module: If the determination is negative, then execute the first algorithm to analyze the equipment bearing fault; Second analysis and judgment module: If the judgment is yes, then determine whether to perform a second analysis for the sensor; Classification Analysis Module: If the second analysis has already been performed, the current vibration data is classified and analyzed using the analyzed centroid coordinates; If the second analysis is not performed, then the second analysis will be performed to obtain the coordinates of the centroid. Update centroid module: The second analysis has been performed. The current vibration data is classified by calculating the coordinates of the centroids of each cluster and the characteristic parameters of the current vibration data. The second analysis is performed at set intervals to perform cluster analysis and update the centroid coordinates. Status judgment module: Based on the second analysis, the system uses the coordinates of the centroid point and the characteristic parameters of the current vibration data to determine the current operating status of the equipment. If the equipment is in operation, the first algorithm is executed to analyze the bearing fault. If the equipment is stationary, the first algorithm is not executed. The second analysis includes: The partitioning module pre-observes a sample of N data points and divides the N objects into K classes using a locally optimal approach. Initial center point selection module: K points are randomly selected as the initial cluster centers, i.e., the initial center points, based on the classification. Classification module: According to the principle of "minimum distance from the initial center point", all observations are assigned to the class of each center point. The number of initial center points is the same as the K value. Update centroid module: Calculate the mean of all sample points in each class to calculate the centroid of each cluster as the new centroid, and update the centroid. Output module: Based on the obtained center point position, repeat the classification step and the center point update step until the center point position no longer changes or the specified number of iterations is reached, and output the classification result of each point and the coordinates of K centroids; The partitioning module further includes: selecting kurtosis, peak value, and root mean square value as input indicators for the second analysis; each sampling yields a three-dimensional coordinate point composed of kurtosis, peak value, and root mean square value; The centroid of each cluster is the average of all data points in that cluster. The coordinates of the centroid of a given cluster are (x...). c ,y c ,z c Then we have: Where N is the number of all data points in the cluster, x n and y n z represents the x-coordinate and y-coordinate of the nth point in the cluster, respectively. n This represents the vertical coordinate of the nth point in the cluster.

8. The mining machinery diagnostic system according to claim 7, characterized in that, In the state determination module: each sampling yields a three-dimensional coordinate point composed of kurtosis, peak value, and root mean square value. The Euclidean distances from this point to the centroid in the running state and the centroid in the stationary state are calculated respectively. The closer distance indicates the sensor's current operating state. The coordinates of the centroid in the running state are PA:(x a ,y a ,z a The coordinates of the stationary centroid are PS:(x s ,y s ,z s The coordinates of the target point are P-1:(x1,y1,z1), and the Euclidean distance from the target point to the centroid of the running state is d. a The Euclidean distance from the target point to the stationary centroid is d. s Then we have: In the classification module, data points are assigned to the nearest cluster center; In the partitioning module, the target state of the device is divided into two states: running and stopped, so the K value is 2; The segmentation module further includes: selecting all reported data from the sensors within M hours to obtain N sets of time-domain statistical indices for sensor acceleration, and using these N sets of indices as input for the second analysis. Input=[[beta1,amp1,rms1],[beta2,amp2,rms2],…,[beta N ,amp N ,rms N ]]; The vibration data includes the vibration frequency.

9. The mining machinery diagnostic system according to claim 7 or 8, characterized in that, The first algorithm analysis includes: High-frequency filtering module: Performs high-frequency filtering on the original signal, removing low-frequency signal components from the original signal. Decomposition module: EMD mode decomposition, taking the mode with the highest similarity to the original signal after decomposition as the decomposed signal; Low-frequency filtering module: Low-frequency filtering removes high-frequency components from the decomposed signal; Envelope signal calculation module: Performs Hilbert transform on the decomposed low-frequency signal to obtain the envelope signal; Envelope spectrum signal acquisition module: Perform Fourier transform on the envelope signal to obtain the envelope spectrum signal.

Images