Belt conveyor sudden failure early warning method and system

Abstract

The application provides a belt conveyor sudden failure early warning method and system, including: a health index acquisition step: after cleaning the equipment sudden failure data, time domain and frequency domain characteristics are calculated, the characteristics are sorted according to the order of the ability to represent the failure, the previous characteristics are subjected to principal component analysis dimension reduction fusion, and the health index representing the belt failure process during the sudden failure is obtained; a failure prediction step: the health index is predicted by using a long short-term memory neural network, the health state curve of the belt failure stage is predicted in the early stage of the sudden failure, the abnormal value in the health state curve is judged, and the belt scratch failure occurrence time is positioned according to the abnormal value. The application can achieve the prediction and alarm function in the early stage of the sudden failure, ensures the safety of the equipment, and improves the construction efficiency of the tunnel boring machine.

Description

Technical Field

[0001] This invention relates to the field of intelligent detection technology for belt conveyors, specifically to a method and system for early warning of sudden faults in belt conveyors. Background Technology

[0002] Belt conveyors, as a key component of the supporting system for tunnel boring machines (TBMs), are commonly used long-distance continuous transport equipment for excavated materials in complex geological environments. The health of the belt conveyor is crucial to the safe and stable operation of the TBM; a malfunction not only severely damages the equipment but also directly halts tunneling, significantly impacting construction efficiency. Therefore, establishing an effective belt conveyor fault early warning system is essential to ensure its safe and reliable operation.

[0003] Currently, the maintenance and fault detection of belt conveyors still mainly rely on manual inspections and periodic replacements, due to the simple structure and relatively low cost of belt conveyors. However, this method is not suitable for situations with harsh geological conditions and the transportation of construction waste. Complex and variable external working conditions can easily lead to sudden failures such as belt tearing, slippage, breakage, and deviation caused by foreign objects getting stuck.

[0004] Currently, data-driven fault early warning methods for belt conveyors have received widespread attention. For example, Chinese invention patent document with publication number CN114662834A discloses an online health status assessment method for mining belt conveyors. By utilizing the acquired feature parameters and employing the multidimensional time series clustering algorithm TICC, the trend of health status changes is obtained, and the health status level is classified based on field experience.

[0005] Chinese utility model patent document with publication number CN204549270U discloses a fault early warning system for a belt conveyor. The system calculates the spectral curves of vibration sensor data for each component, and issues an early warning signal if the deviation from the standard spectral curve exceeds a threshold.

[0006] In Chinese invention patent document with publication number CN112232366A, a method and system for early warning of electrical equipment faults based on RFID monitoring is disclosed. The method obtains first and second fault warning information by training a time-series temperature dataset through a denoising autoencoder network and a long short-term memory network, respectively, and then inputs it into a trained XGBoost model to obtain the fault warning level.

[0007] However, existing fault early warning methods cannot provide predictive alarms when fault trends are not yet significant. The specific reasons are as follows:

[0008] 1) For sudden failures, the early fault characteristic curve does not change significantly, but the gradient of the characteristic curve increases rapidly in a short period of time when the failure occurs. Existing fault early warning methods for belt conveyors cannot provide early warning for sudden failures, which can easily lead to untimely alarms. By the time the alarm signal is issued, the fault has often already caused serious consequences.

[0009] 2) Setting alarm thresholds based on field experience or historical fault curves can easily lead to missed or false alarms. Some early warning systems use deep learning methods for classification and diagnosis, which requires a large amount of fault training data and is computationally inefficient.

[0010] The belt conveyor system supporting tunnel boring machines operates continuously in complex geological environments, transporting excavated soil out of the tunnel. Falling foreign objects can easily cause sudden malfunctions such as belt tearing, slippage, breakage, and misalignment. Existing fault warning methods lack the ability to predict fault trends from historical data, failing to issue alarm signals in the early stages of a fault. Furthermore, relying on field experience or historical fault curves to determine fault occurrence can easily lead to missed or false alarms. Summary of the Invention

[0011] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for early warning of sudden failures in belt conveyors.

[0012] A method for early warning of sudden failures in a belt conveyor according to the present invention includes:

[0013] Steps for obtaining health indicators: After cleaning the sudden equipment failure data, perform time domain and frequency domain feature calculations, sort the features according to the order of their ability to represent failure, perform principal component analysis to reduce the dimension and fuse the preceding features, and obtain health indicators that represent the belt failure process during the sudden failure.

[0014] Fault prediction steps: Use a long short-term memory neural network to make short-term predictions of health indicators, predict the health status curve of the belt failure stage in the early stage of sudden failure, identify outliers in the health status curve, and locate the time of belt scratch failure based on the outliers.

[0015] Preferably, the sudden equipment failure data includes: missing values ​​of equipment detection data, sensor noise, and gross errors caused by external disturbances.

[0016] Preferably, the process of cleaning the equipment sudden failure data includes: acquiring the original equipment sudden failure data, eliminating irrelevant zeros in the original data to exclude missing values, removing gross errors in the data using the 3σ criterion, eliminating sensor observation noise using Kalman filtering, and finally obtaining the cleaned equipment sudden failure data.

[0017] Preferably, the time-domain and frequency-domain features include: mean, standard deviation, skewness, kurtosis, peak-to-peak value, root mean square, peak factor, waveform factor, impulse factor, margin factor, energy entropy, mean spectral kurtosis, standard deviation spectral kurtosis, skewness spectral kurtosis, and kurtosis kurtosis.

[0018] Preferably, the fault early warning method includes the following steps:

[0019] Step S1: Obtain the time domain and frequency domain characteristics of the belt speed data of the belt conveyor;

[0020] Step S2: Calculate the monotonicity and correlation of the eigenvalues ​​and use them as the basis for ranking importance. After combining the two, select the features with strong fault characterization capabilities and perform principal component analysis for dimensionality reduction and fusion.

[0021] Step S3: Construct a health index characterizing the trend of idiosyncratic failures using principal component analysis feature dimensionality reduction and fusion method;

[0022] Step S4: Use a long short-term memory neural network to make short-term predictions of health indicators;

[0023] Step S5: Use the Isolation Forest algorithm to identify outliers in the prediction data and locate the belt failure time based on the outliers.

[0024] Preferably, the monotonicity sorting method is as follows:

[0025]

[0026] In the formula, Monotonicity(x) i ) represents the monotonicity value. and It represents the number of positive and negative values ​​after differentiating the eigenvalues, where m is the number of eigenvalues ​​and n is the number of eigenvalue sampling points.

[0027] The relevance ranking method is as follows:

[0028]

[0029] In the formula, Correlation(x) i Corr(x) represents the correlation value. i x k ) represents the internal covariance of the same feature.

[0030] Preferably, the fault warning method further includes: dividing the data before failure and the data after failure according to a set ratio, selecting the data before failure as the training set for feature value acquisition and fusion, selecting the data after failure to verify the prediction results of health indicators, and adjusting the hyperparameters of the long short-term memory neural network, including the learning rate, the number of iterations, and the batch size, according to the verification results.

[0031] A sudden failure early warning system for a belt conveyor according to the present invention includes:

[0032] Health indicator acquisition module: After cleaning the equipment sudden failure data, time domain and frequency domain features are calculated, the features are sorted according to the strength of their ability to represent failure, and principal component analysis is performed on the preceding features to reduce dimensions and fuse them to obtain health indicators that represent the belt failure process during the sudden failure.

[0033] Fault prediction module: It uses a long short-term memory neural network to make short-term predictions of health indicators, predicts the health status curve of the belt failure stage in the early stage of sudden failure, identifies outliers in the health status curve, and locates the time of belt scratch failure based on the outliers.

[0034] Preferably, the sudden equipment failure data includes: missing values ​​of equipment detection data, sensor noise, and gross errors caused by external disturbances.

[0035] Preferably, the process of cleaning the equipment sudden failure data includes: acquiring the original equipment sudden failure data, eliminating irrelevant zeros in the original data to exclude missing values, removing gross errors in the data using the 3σ criterion, eliminating sensor observation noise using Kalman filtering, and finally obtaining the cleaned equipment sudden failure data.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] This invention proposes a fault early warning system for belt conveyors under complex geological conditions. By constructing a health index that reflects the changing trend of sudden faults, a short-term prediction algorithm is used to predict the waveform of the health index when failure occurs in the early stage of the fault. Finally, the fault is identified through an anomaly detection algorithm, thereby achieving the function of predicting and alarming sudden faults in the early stage, ensuring equipment safety, and improving the construction efficiency of tunnel boring machines. Attached Figure Description

[0038] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0039] Figure 1 This is a flowchart of the sudden fault early warning algorithm for the belt conveyor of the present invention;

[0040] Figure 2 This is a flowchart of the data cleaning process for sudden equipment failures in this invention.

[0041] Figure 3 This is a schematic diagram illustrating the construction of health indicators using principal component analysis in this invention.

[0042] Figure 4 This is a schematic diagram illustrating the prediction of health indicators using a long short-term memory neural network, as described in this invention.

[0043] Figure 5 This is a schematic diagram illustrating the detection of abnormal predicted values ​​using the isolated forest algorithm of this invention. Detailed Implementation

[0044] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0045] This invention discloses a method for early warning of sudden failures in belt conveyors, referring to... Figure 1 ,include:

[0046] Steps for obtaining health indicators: After cleaning the equipment sudden failure data, perform time-domain and frequency-domain feature calculations. Sort the features according to their ability to characterize the failure, and perform principal component analysis (PCA) to reduce and fuse the preceding features to obtain health indicators characterizing the belt failure process during the sudden failure. For example, sort the features according to monotonicity and correlation, and then perform PCA to reduce and fuse the features with the highest overall ranking.

[0047] Fault prediction steps: Use a long short-term memory neural network to make short-term predictions of health indicators, predict the health status curve of the belt failure stage in the early stage of sudden failure, use the isolated forest algorithm to identify outliers in the health status curve, and locate the time of belt scratch failure based on outliers.

[0048] Based on the above process, the system can predict and identify belt failure in the early stages of a fault by using the constructed emergency fault health indicators, and inform maintenance personnel to replace the damaged parts before the equipment stops abnormally.

[0049] The monitoring data of belt conveyors contains missing values, sensor noise, and gross errors caused by complex external disturbances. Figure 2 This demonstrates the data cleaning steps before analysis, including... Figure 2 (a) is a section of belt speed monitoring data that caused belt scratches due to a large rock suddenly jamming the belt conveyor. This data is used to train the algorithm to realize the function of early fault alarm in the belt scratches caused by the rock. Figure 2 (b) shows the data after removing missing values ​​by eliminating irrelevant zeros; Figure 2 (c) shows the data after removing gross errors using the 3σ criterion; Figure 2 (d) shows the data after sensor observation noise has been eliminated by Kalman filtering.

[0050] Figure 3 This demonstrates the process of constructing health indicators based on principal component analysis (PCA), including the following steps:

[0051] Step S1: Obtain the time-domain and frequency-domain characteristics of the belt conveyor speed data. These characteristics include, but are not limited to: mean, standard deviation, skewness, kurtosis, peak-to-peak value, root mean square, peak factor, waveform factor, impulse factor, margin factor, energy entropy, mean spectral kurtosis, standard deviation spectral kurtosis, skewness spectral kurtosis, and kurtosis spectral kurtosis. The calculation formulas are shown in Table 1, x i Let N represent the i-th input data, N be the number of data points contained in a single feature sampling point, and E be the number of data points in a single feature sampling point. i Let t represent the energy value of the i-th data, and let f represent the time and frequency, respectively. STFT(t, f) represents the short-time Fourier transform of the data.

[0052] Table 1 Formulas for calculating time-domain and frequency-domain eigenvalues

[0053]

[0054]

[0055] Step S2: Calculate the monotonicity and correlation of the feature values ​​and use them as the basis for ranking importance. After combining the two, select features with strong fault characterization capabilities for principal component analysis, dimensionality reduction, and fusion. Divide the data before and after failure according to a set ratio. Select the data before failure as the training set for feature value acquisition and fusion, and select the data after failure to verify the health indicator prediction results. Adjust the hyperparameters of the long short-term memory neural network, including the learning rate, number of iterations, and batch size, based on the verification results.

[0056] The monotonic sorting method is as follows:

[0057]

[0058] In the formula, Monotonicity(x) i ) represents the monotonicity value. and It represents the number of positive and negative values ​​after differentiating the eigenvalues, where m is the number of eigenvalues ​​and n is the number of eigenvalue sampling points.

[0059] The relevance ranking method is as follows:

[0060]

[0061] In the formula, Correlation(x) i Corr(x) represents the correlation value. i x k) represents the internal covariance of the same feature.

[0062] Step S3: Construct a health index characterizing the trend of idiosyncratic failures using principal component analysis feature dimensionality reduction and fusion method. That is, use orthogonal transformation to map high-dimensional data to low-dimensional space to obtain a set of orthogonal, low-dimensional representations, so as to reduce the amount of data while retaining sufficient key features.

[0063] Step S4: Use a Long Short-Term Memory (LSTM) neural network to make short-term predictions of health indicators. (Refer to...) Figure 4 As shown in the diagram, the health index values ​​from the time the large rock block jams to the time the conveyor belt fails can effectively predict the health status at the time of belt failure. The LSTM structure consists of cell states, a forget gate, an input gate, and an output gate. The forget gate is used to filter out invalid information, the input gate is used to supplement valid information, and the output gate is used to update the cell state. The calculation formulas for different gates at each time step are as follows:

[0064] f t =σ(W f ·[h t-1 x t ]+b f )

[0065] i t =σ(W i ·[h t-1 x t ]+b i )

[0066]

[0067]

[0068] o t =σ(W o [h t-1 x t ]+b o )

[0069] h t =o t *tanh(C t )

[0070] In the formula f t i t o t , C t W represents the forget gate, input gate, output gate, input cell state, and output cell state, respectively. f W i W C W o bf b i b C b o These represent the corresponding weight coefficient matrix and bias term, respectively, and σ and tanh represent the sigmoid and hyperbolic tangent activation functions, respectively.

[0071] Step S5: Use the Isolation Forest algorithm (iForest) to identify outliers in the prediction data, such as... Figure 5 As shown, the failure time of the belt is located based on outliers. iForest is an unsupervised fast anomaly detection method suitable for continuous data, with linear time complexity and high accuracy. It adopts a strategy of randomly recursively splitting the dataset until the sample points are isolated, and regards sparsely distributed points that are far away from the high-density group as outliers.

[0072] This invention also discloses a sudden fault early warning system for belt conveyors, comprising:

[0073] Health indicator acquisition module: After cleaning the equipment sudden failure data, perform time domain and frequency domain feature calculations, and perform principal component analysis to reduce the dimension and fuse some features with strong fault characterization capabilities to obtain health indicators characterizing the belt failure process during the sudden failure.

[0074] Fault prediction module: It uses a long short-term memory neural network to make short-term predictions of health indicators, predicts the health status curve of the belt failure stage in the early stage of sudden failure, identifies outliers in the health status curve, and locates the time of belt scratch failure based on the outliers.

[0075] In a preferred embodiment, the sudden equipment failure data includes: missing values ​​of equipment detection data, sensor noise, and gross errors caused by external disturbances.

[0076] In a preferred embodiment, the process of cleaning the equipment sudden failure data includes: acquiring the original equipment sudden failure data, eliminating irrelevant zeros in the original data to exclude missing values, removing gross errors in the data using the 3σ criterion, eliminating sensor observation noise using Kalman filtering, and finally obtaining the cleaned equipment sudden failure data.

[0077] In a preferred embodiment, the time-domain and frequency-domain features include: mean, standard deviation, skewness, kurtosis, peak-to-peak value, root mean square, peak factor, waveform factor, impulse factor, margin factor, energy entropy, mean spectral kurtosis, standard deviation spectral kurtosis, skewness spectral kurtosis, and kurtosis kurtosis.

[0078] In a preferred embodiment, the belt conveyor emergency failure early warning system includes the following modules:

[0079] Module M1: Acquires the time-domain and frequency-domain characteristics of belt speed data for belt conveyors;

[0080] Module M2: Calculates the monotonicity and correlation of eigenvalues ​​and uses them as the basis for ranking importance. After combining the two, features with strong fault characterization capabilities are selected for principal component analysis, dimensionality reduction and fusion.

[0081] Module M3: Constructs a health index characterizing the trend of idiosyncratic failures using principal component analysis feature dimensionality reduction and fusion methods;

[0082] Module M4: Uses a long short-term memory neural network to make short-term predictions of health indicators;

[0083] Module M5: Uses the isolated forest algorithm to identify outliers in the predicted data and locates the belt failure time based on the outliers.

[0084] In a preferred embodiment, the monotonicity sorting method is as follows:

[0085]

[0086] In the formula, Monotonicity(x) i ) represents the monotonicity value. and It represents the number of positive and negative values ​​after differentiating the eigenvalues, where m is the number of eigenvalues ​​and n is the number of eigenvalue sampling points.

[0087] The relevance ranking method is as follows:

[0088]

[0089] In the formula, Correlation(x) i Corr(x) represents the correlation value. i x k ) represents the internal covariance of the same feature.

[0090] In a preferred embodiment, data before failure and data after failure are divided according to a set ratio. Data before failure is selected as the training set for feature value acquisition and fusion, and data after failure is selected to verify the prediction results of health indicators.

[0091] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0092] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for early warning of sudden failures in a belt conveyor, characterized in that, include: Steps for obtaining health indicators: After cleaning the sudden equipment failure data, perform time domain and frequency domain feature calculations, sort the features according to the order of their ability to represent failure, perform principal component analysis to reduce the dimension and fuse the preceding features, and obtain health indicators that represent the belt failure process during the sudden failure. Fault prediction steps: Use a long short-term memory neural network to make short-term predictions of health indicators, predict the health status curve of the belt failure stage in the early stage of sudden failure, identify outliers in the health status curve, and locate the time of belt scratch failure based on the outliers. The steps for obtaining health indicators also include: Step S1: Obtain the time domain and frequency domain characteristics of the belt speed data of the belt conveyor; Step S2: Calculate the monotonicity and correlation of the eigenvalues ​​and use them as the basis for ranking importance. After combining the two, select the features with strong fault characterization capabilities and perform principal component analysis for dimensionality reduction and fusion.

2. The method for early warning of sudden failures in a belt conveyor according to claim 1, characterized in that: The equipment failure data includes: missing values ​​in equipment detection data, sensor noise, and gross errors caused by external disturbances.

3. The method for early warning of sudden failures in a belt conveyor according to claim 2, characterized in that: The process of cleaning the equipment sudden failure data includes: acquiring the original equipment sudden failure data, eliminating irrelevant zeros in the original data to exclude missing values, removing gross errors in the data using the 3σ criterion, eliminating sensor observation noise using Kalman filtering, and finally obtaining the cleaned equipment sudden failure data.

4. The method for early warning of sudden failures in a belt conveyor according to claim 1, characterized in that: The time-domain and frequency-domain features include: mean, standard deviation, skewness, kurtosis, peak-to-peak value, root mean square, peak factor, waveform factor, impulse factor, margin factor, energy entropy, mean spectral kurtosis, standard deviation spectral kurtosis, skewness spectral kurtosis, and kurtosis kurtosis.

5. The method for early warning of sudden failures in a belt conveyor according to claim 1, characterized in that, Includes the following steps: Step S3: Construct a health index characterizing the trend of idiosyncratic failures using principal component analysis feature dimensionality reduction and fusion method; Step S4: Use a long short-term memory neural network to make short-term predictions of health indicators; Step S5: Use the Isolation Forest algorithm to identify outliers in the prediction data and locate the belt failure time based on the outliers.

6. The method for early warning of sudden failures in a belt conveyor according to claim 5, characterized in that: The monotonic sorting method is as follows: In the formula, Monotonicity(x) i ) represents the monotonicity value. and It represents the number of positive and negative values ​​after differentiating the eigenvalues, where m is the number of eigenvalues ​​and n is the number of eigenvalue sampling points. The relevance ranking method is as follows: In the formula, Correlation(x) i Corr(x) represents the correlation value. i ,x k ) represents the internal covariance of the same feature.

7. The method for early warning of sudden failures in a belt conveyor according to claim 5, characterized in that: Also includes: Data is divided into pre-failure and post-failure data according to a set ratio. Pre-failure data is selected as the training set for feature acquisition and fusion, while post-failure data is selected to verify the health indicator prediction results. Based on the verification results, the hyperparameters of the long short-term memory neural network, including the learning rate, number of iterations, and batch size, are adjusted.

8. A sudden fault early warning system for a belt conveyor, characterized in that, include: Health indicator acquisition module: After cleaning the equipment sudden failure data, time domain and frequency domain features are calculated, the features are sorted according to the strength of their ability to represent failure, and principal component analysis is performed on the preceding features to reduce dimensions and fuse them to obtain health indicators that represent the belt failure process during the sudden failure. Fault prediction module: Utilizes a long short-term memory neural network to make short-term predictions of health indicators, predicts the health status curve of the belt failure stage in the early stage of a sudden failure, identifies outliers in the health status curve, and locates the time of belt scratch failure based on the outliers. The health indicator acquisition module also includes: Module M1: Acquires the time-domain and frequency-domain characteristics of belt speed data for belt conveyors; Module M2: Calculates the monotonicity and correlation of eigenvalues ​​and uses them as the basis for ranking importance. After combining the two, features with strong fault characterization capabilities are selected for principal component analysis dimensionality reduction and fusion.

9. The belt conveyor sudden fault early warning system according to claim 8, characterized in that: The equipment failure data includes: missing values ​​in equipment detection data, sensor noise, and gross errors caused by external disturbances.

10. The belt conveyor sudden fault early warning system according to claim 9, characterized in that: The process of cleaning the equipment sudden failure data includes: acquiring the original equipment sudden failure data, eliminating irrelevant zeros in the original data to exclude missing values, removing gross errors in the data using the 3σ criterion, eliminating sensor observation noise using Kalman filtering, and finally obtaining the cleaned equipment sudden failure data.

Images