A belt conveyor state early warning method, device, equipment and storage medium

Abstract

This invention provides a method, device, equipment, and storage medium for early warning of belt conveyor status. The method includes: acquiring multi-source dynamic time-series signals of the belt conveyor and extracting multi-domain features from the multi-source dynamic time-series signals; performing feature screening on the multi-domain features based on a comprehensive screening index constructed from monotonicity, robustness, and trend indicators to obtain multi-domain target features; performing nonlinear dimensionality reduction on the multi-domain target features based on kernel principal component analysis to obtain multi-domain dimensionality-reduced features; inputting the multi-domain dimensionality-reduced features into a long short-term memory network model to obtain the predicted status of the belt conveyor, and issuing an early warning based on the predicted status. This invention uses a comprehensive screening index system composed of monotonicity, robustness, and trend indicators for feature screening, which can improve the accuracy and efficiency of status early warning.

Description

Technical Field

[0001] This invention relates to the field of intelligent fault diagnosis technology, specifically to a method, device, equipment, and storage medium for early warning of belt conveyor status. Background Technology

[0002] As a key continuous transportation device in mining production systems, the operating status of conveyor belts directly affects production efficiency and safety. In actual working conditions, belt conveyors operate under high loads, strong impacts, and dusty environments for extended periods, making them prone to typical faults such as belt misalignment, slippage, abnormal tension, and drum wear. Because the early characteristics of different faults are often weak and coupled with each other, coupled with the complex nonlinear relationships between multi-source monitoring signals, belt conveyor fault early warning faces technical challenges such as high feature dimension, low signal-to-noise ratio, and difficulty in capturing fault evolution trends.

[0003] Existing belt conveyor condition monitoring methods mainly fall into two categories: one is the empirical judgment method based on fixed thresholds, which relies on a single or few parameters and is difficult to cope with nonlinear coupling and strong noise interference, easily leading to false alarms or missed alarms; the other is the intelligent recognition method based on machine learning, among which Long Short-Term Memory (LSTM) networks have attracted attention due to their temporal modeling capabilities. However, directly inputting the original high-dimensional monitoring features into an LSTM has obvious drawbacks: on the one hand, the original features contain a large amount of redundancy and noise, which can easily lead to model overfitting and low training efficiency; on the other hand, different features contribute differently to fault evolution, lacking an objective screening mechanism. In addition, traditional linear dimensionality reduction methods (such as PCA) cannot effectively handle the strong nonlinear coupling relationships in belt conveyor monitoring data.

[0004] Therefore, there is an urgent need to provide a method, device, equipment, and storage medium for early warning of belt conveyor status, which can eliminate a large amount of redundancy and noise in the original features and, combined with nonlinear dimensionality reduction, improve the accuracy and representativeness of the features input to the LSTM model, thereby improving the accuracy and efficiency of early warning of belt conveyor status. Summary of the Invention

[0005] In view of this, it is necessary to provide a belt conveyor condition early warning method, device, equipment and storage medium to solve the technical problem in the prior art that the accuracy and efficiency of belt conveyor condition early warning are low because the original features are directly input into the model for prediction, and the large amount of redundancy and noise in the original features leads to a large amount of noise.

[0006] To address the aforementioned technical problems, in a first aspect, the present invention provides a belt conveyor status early warning method, comprising: Acquire multi-source dynamic time-series signals of the belt conveyor and extract multi-domain features of the multi-source dynamic time-series signals; Feature selection is performed on the multi-domain features based on a comprehensive screening index constructed from monotonicity, robustness, and trend indicators to obtain multi-domain target features. The monotonicity index is used to characterize the degree to which the feature exhibits a monotonically increasing or decreasing trend over time, reflecting the irreversibility of fault evolution. The robustness index is used to characterize the feature's ability to resist instantaneous shocks and random noise interference. The trend indicator is used to characterize the correlation between the feature and runtime, reflecting the feature's ability to follow long-term degradation processes. Based on kernel principal component analysis, nonlinear dimensionality reduction is performed on the multi-domain target features to obtain multi-domain dimensionality-reduced features. The multi-domain dimensionality reduction features are input into a long short-term memory network model to obtain the predicted state of the belt conveyor, and an early warning is issued based on the predicted state.

[0007] In one possible implementation, the feature filtering of the multi-domain features based on a comprehensive screening index constructed from monotonicity indicators, robustness indicators, and trend indicators to obtain multi-domain target features includes: Calculate the monotonicity index, robustness index, and trend index of each feature in the multi-domain features; The weighted sum of the monotonicity index value, the robustness index value, and the trend index value is used as the comprehensive index value; Features whose comprehensive index value is greater than the index threshold are used as the multi-domain target features.

[0008] In one possible implementation, the comprehensive screening index is:

[0009]

[0010]

[0011]

[0012]

[0013] In the formula, For comprehensive screening indicators; It is a monotonicity indicator; For robustness indicators; It is a trend indicator; , , Weights for monotonicity indicators, robustness indicators, and trend indicators; The number of samples in the feature sequence F; Take the derivative of the characteristic sequence F with respect to time; for The number of; for The number of; Let i be the i-th feature at time t; The residual of the average trend of the i-th feature at time t and the i-th feature at time t after exponentially weighted moving average smoothing is given. The average trend of the i-th feature at time t after exponentially weighted moving average smoothing; It is the difference between the rank of the i-th feature and the time series and feature sequence F after sorting them in order.

[0014] In one possible implementation, the nonlinear dimensionality reduction of the multi-domain target features based on kernel principal component analysis to obtain multi-domain dimensionality-reduced features includes: The multi-domain target features are implicitly mapped to a high-dimensional feature space using a preset kernel function; Principal component analysis is performed in the high-dimensional feature space, and the principal components are extracted as the multi-domain dimensionality reduction features.

[0015] In one possible implementation, the implicit mapping of the multi-domain target features to a high-dimensional feature space using a preset kernel function includes: The kernel function is used to calculate the kernel function value between any two features in the multi-domain target features, and a kernel matrix is ​​constructed based on the kernel function value. The kernel matrix is ​​used to describe the inner product relationship between sample points in the high-dimensional feature space. The kernel matrix is ​​centered to obtain a centered kernel matrix.

[0016] In one possible implementation, performing principal component analysis in the high-dimensional feature space and extracting principal components as the multi-domain dimensionality reduction features includes: The eigenvalues ​​and corresponding eigenvectors are obtained by performing eigenvalue decomposition on the centered kernel matrix. Calculate the contribution rate of each feature value, sort the feature values ​​in descending order of contribution rate, and select the feature vector corresponding to the feature value whose cumulative contribution rate is greater than the contribution rate threshold as the target feature vector; The multi-domain target features are projected onto the direction of the target feature vector to obtain the multi-domain dimensionality reduction features.

[0017] In one possible implementation, the multi-domain features include time-domain features, frequency-domain features, and time-frequency-domain features. The time-domain features include dimensional time-domain features and dimensionless time-domain features. The dimensional time-domain features include root mean square value, mean, variance, absolute mean, root mean amplitude, peak value, and peak-to-peak value. The dimensionless time-domain features include skewness index, kurtosis index, peak value index, margin index, impulse index, and waveform index. The frequency-domain features include spectral amplitude mean, spectral amplitude variance, spectral frequency centroid, second-order centroid, root mean square frequency, and spectral amplitude skewness. The time-frequency-domain features include the energy percentage of each node in the last layer after performing wavelet packet decomposition of the multi-source dynamic time-series signal at a specified number of layers.

[0018] Secondly, the present invention also provides a belt conveyor status early warning device, comprising: The feature extraction unit is used to acquire the multi-source dynamic time-series signal of the belt conveyor and extract the multi-domain features of the multi-source dynamic time-series signal; The feature filtering unit is used to filter the multi-domain features based on a comprehensive filtering index constructed from monotonicity, robustness, and trend indicators to obtain multi-domain target features. The monotonicity index is used to characterize the degree to which the feature exhibits a monotonically increasing or decreasing trend over time, reflecting the irreversibility of fault evolution. The robustness index is used to characterize the feature's ability to resist instantaneous shocks and random noise interference. The trend indicator is used to characterize the correlation between the feature and runtime, reflecting the feature's ability to follow long-term degradation processes. A nonlinear dimensionality reduction unit is used to perform nonlinear dimensionality reduction on the multi-domain target features based on kernel principal component analysis to obtain multi-domain dimensionality-reduced features. The state warning unit is used to input the multi-domain dimensionality reduction features into the long short-term memory network model to obtain the predicted state of the belt conveyor, and to issue a warning based on the predicted state.

[0019] Thirdly, the present invention also provides a belt conveyor status early warning device, including a data acquisition module, a memory, and a processor, wherein, The data acquisition module is used to acquire multi-source dynamic timing signals of the belt conveyor; The memory is used to store programs; The processor, coupled to the memory and the data acquisition module, is used to execute the program stored in the memory to implement the steps in the belt conveyor status early warning method described in any of the above possible implementations.

[0020] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps in the belt conveyor status early warning method described in any of the above possible implementations.

[0021] The beneficial effects of this invention are as follows: The belt conveyor condition early warning method provided by this invention constructs a comprehensive screening index system composed of monotonicity indicators, robustness indicators, and trend indicators. Monotonicity ensures that the feature changes monotonically as the fault worsens, avoiding misleading reverse fluctuations; robustness suppresses false jumps caused by impact noise; and trend ensures that the feature is synchronized with the long-term degradation process. The multi-domain target features selected through the comprehensive screening of these three factors inevitably possess stable, monotonic, and time-dependent characteristics, thereby eliminating inferior interference at the data source, retaining true fault information, significantly enhancing the sensitivity to early, minor faults, and thus providing a reliable data foundation for subsequent accurate belt conveyor condition early warning.

[0022] Furthermore, this invention employs kernel principal component analysis (KPCA) to perform nonlinear dimensionality reduction on multi-domain target features. KPCA can fully uncover the low-dimensional essential features hidden in strong nonlinear relationships, while removing relevant noise and redundant information. This avoids the information loss and feature distortion of traditional linear dimensionality reduction methods (such as PCA) under nonlinear conditions. While ensuring the representativeness of the multi-domain dimensionality reduction features, redundant features are proposed, thereby improving the efficiency of the condition warning while ensuring the accuracy of the belt conveyor condition warning.

[0023] Furthermore, this invention uses a long short-term memory network model for state prediction and its gating mechanism to capture the long-term evolution trend of belt conveyor misalignment or wear, overcoming the limitation of static networks in modeling time dependence and further improving the accuracy of state and early warning. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A schematic flowchart of an embodiment of the belt conveyor status early warning method provided by the present invention; Figure 2 For the present invention Figure 1 A schematic diagram of an embodiment of S102; Figure 3 For the present invention Figure 1 A schematic diagram of an embodiment of S103; Figure 4 For the present invention Figure 3 A schematic diagram of an embodiment of S302; Figure 5A schematic diagram of an embodiment of the belt conveyor status early warning device provided by the present invention; Figure 6 This is a schematic diagram of an embodiment of the belt conveyor status early warning device provided by the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0027] It should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this invention illustrate operations implemented according to some embodiments of the invention. It should be understood that the operations in the flowcharts may be implemented out of order, and steps without logical contextual relationships may be reversed or performed simultaneously. Furthermore, those skilled in the art, guided by the content of this invention, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0028] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0029] This invention provides a method, device, equipment, and storage medium for early warning of belt conveyor status, which will be described below.

[0030] Figure 1 This is a schematic flowchart of an embodiment of the belt conveyor status early warning method provided by the present invention, as shown below. Figure 1 As shown, the belt conveyor status early warning methods include: S101. Acquire the multi-source dynamic timing signal of the belt conveyor and extract the multi-domain features of the multi-source dynamic timing signal.

[0031] Among them, the multi-source dynamic timing signals include, but are not limited to: belt speed, motor current, multi-point tension, offset and idler vibration signals.

[0032] In a specific embodiment of the present invention, the multi-domain features include time-domain features, frequency-domain features, and time-frequency-domain features.

[0033] Specifically, time-domain characteristics include dimensional time-domain characteristics and dimensionless time-domain characteristics. Dimensional time-domain characteristics refer to characteristics that have clear physical meaning but are easily affected by complex operating conditions such as heavy ore load impact. Dimensionless time-domain characteristics, also known as dimensionless time-domain characteristics, refer to characteristics that are sensitive to early minor belt misalignment or local wear faults and are not easily affected by external factors.

[0034] Among them, dimensional time-domain characteristics include the root mean square (RMS), mean, variance, absolute mean, root square amplitude, peak value, and peak-to-peak value of each time-series signal. The RMS reflects the average energy or effective value of the signal and is sensitive to fluctuations and vibrations. The mean represents the DC component or static offset trend of the signal. The variance measures the dispersion of the signal around the mean, reflecting the amplitude of fluctuations. The absolute mean is the arithmetic mean of the absolute amplitude values; it is simple to calculate and has better noise resistance than the RMS. The root square amplitude is calculated by taking the square root of the amplitude, averaging, and then squaring; it is more sensitive to changes in weak signals. The peak value refers to the maximum instantaneous amplitude of the signal, used to capture transient impacts. The peak-to-peak value is the difference between the maximum and minimum values, reflecting the total fluctuation range of the signal. Because these characteristics have definite physical units and dimensions, they can intuitively describe the energy and amplitude levels of the signal, but they are susceptible to interference from heavy loads and other operating conditions.

[0035] Dimensionless time-domain characteristics include skewness, kurtosis, peak value, margin, impulse, and waveform indices. Skewness reflects the degree of asymmetry in the signal amplitude distribution and is used to determine the presence of unidirectional offset or friction. Kurtosis measures the sharpness of the signal distribution and is highly sensitive to impulses, often used for early fault detection. Peak value is the ratio of peak value to RMS value, characterizing the instantaneous impulse characteristics of the signal. Margin is the ratio of peak value to RMS amplitude and has a good response to wear-related faults. Impulse is the ratio of peak value to absolute mean, highlighting periodic impulse components. Waveform indices are the ratio of RMS value to absolute mean, reflecting the overall waveform changes of the signal and are often used for stability evaluation. Due to their dimensionless nature, these indices can effectively suppress operating condition fluctuations and highlight fault characteristics.

[0036] Specifically, the frequency domain features include the mean spectral amplitude, variance spectral amplitude, centroid of spectral frequency, second-order centroid, root-mean-square frequency, and skewness of spectral amplitude. These features can be calculated based on the spectrum after a Fast Fourier Transform. The mean spectral amplitude reflects the overall energy level of the spectrum; the variance spectral amplitude measures the dispersion of each frequency component's amplitude relative to the mean, reflecting the spectrum's volatility; the centroid of spectral frequency (i.e., the first-order spectral moment) characterizes the concentrated position of spectral energy on the frequency axis, used to determine the shift of the main vibration frequency; the second-order centroid (i.e., the second-order spectral moment) reflects the degree of dispersion of spectral energy and is sensitive to changes in frequency distribution; the root-mean-square frequency is the weighted root-mean-square of the frequency squares, highlighting the influence of high-frequency components; and the skewness of spectral amplitude describes the asymmetry of the spectral amplitude distribution, used to detect the presence of asymmetric modulation or fault sidebands. These features comprehensively characterize the frequency distribution pattern of the conveyor belt vibration signal, providing excellent identification capabilities for frequency structure changes caused by faults such as belt misalignment and wear.

[0037] It should be understood that, in addition to the frequency domain features mentioned above, other frequency domain features can be introduced to construct 12 frequency domain features.

[0038] Specifically, the time-frequency domain features include the energy percentage of each node in the last layer after performing wavelet packet decomposition of a specified number of layers on a multi-source dynamic time-series signal.

[0039] The energy of each node after wavelet packet decomposition is expressed as:

[0040] In the formula, Let be the node coefficients of the i-th layer and the j-th node after wavelet decomposition. Let be the energy of the coefficient of the i-th layer and the j-th node. The energy of the node coefficients will change after the belt conveyor runs off-track or deteriorates mechanically. Therefore, the ratio of the energy of the node coefficients of a certain layer after wavelet packet decomposition to the sum of the energies of all nodes in that layer can be used as the time-frequency domain feature of that node, and can be expressed as:

[0041] In the formula, E i,j This represents the energy percentage of the i-th layer and the j-th node.

[0042] In a specific embodiment of the present invention, the time-series signal is decomposed into three layers of wavelet packets and the energy proportion of the last layer node is extracted, resulting in a total of eight features as time-frequency domain features.

[0043] It should be noted that wavelet packet decomposition has a strong ability to decompose time and frequency locally, and is suitable for processing complex non-stationary signals caused by the impact of falling ore.

[0044] In summary, the embodiments of the present invention extracted a total of 33 features, including 13 time-domain features, 12 frequency-domain features, and 8 time-frequency-domain features.

[0045] It should also be noted that, in order to eliminate the influence of the dimensions among the features and improve the convergence speed of the subsequent prediction model, the time domain features, frequency domain features, and time-frequency domain features are all subjected to max-min normalization.

[0046] S102. Feature screening is performed on multi-domain features based on a comprehensive screening index constructed from monotonicity index, robustness index, and trend index to obtain multi-domain target features. The monotonicity index is used to characterize the degree to which a feature exhibits a monotonically increasing or decreasing trend with time series, reflecting the irreversibility of fault evolution. The robustness index is used to characterize the ability of a feature to resist instantaneous shocks and random noise interference. The trend index is used to characterize the correlation between a feature and running time, reflecting the feature's ability to follow long-term degradation processes.

[0047] S103. Nonlinear dimensionality reduction of multi-domain target features is performed based on kernel principal component analysis to obtain multi-domain dimensionality-reduced features. S104. Input the multi-domain dimensionality reduction features into the Long Short-Term Memory (LSTM) network model to obtain the predicted state of the belt conveyor, and issue an early warning based on the predicted state.

[0048] It should be understood that an alert is issued when the predicted state is faulty, but no alert is issued when the predicted state is healthy. A more refined approach could be to issue alerts based on the severity of the fault.

[0049] Specifically, LSTM effectively captures long-term temporal dependencies in the multidimensional operational data of belt conveyors through its unique internal gating mechanism, successfully overcoming the gradient vanishing or gradient exploding problems that traditional recurrent neural networks (RNNs) are prone to when processing long sequences. Inside the LSTM unit, the state update process is driven by a series of core formulas. The first is the Forget Gate, which determines the degree to which historical operational information (such as historical deviation trends) is retained from the previous time step, thus effectively filtering out strong noise interference such as instantaneous ore falling. Its calculation formula is: 。

[0050] Secondly, the input gate controls the extent to which newly acquired nonlinear principal component information, after KPCA dimensionality reduction, is written into the memory cell. Its formula is: .

[0051] At the same time, the network generates candidate cell states to characterize the latest operational features of the current device, calculated using the following formula: .

[0052] Subsequently, cell state updates are performed. This step dynamically fuses historical effective feature information with the latest candidate state to form the belt health cell state throughout the entire time series. The update formula is as follows: .

[0053] Next is the Output Gate, which controls which key fault evolution features in the currently updated cell state need to be activated and output. The formula is: .

[0054] Finally, the hidden state output is obtained, which is the health and deviation feature representation at the current moment, and the formula is: .

[0055] Of all the calculation formulas mentioned above, This represents the multi-domain dimensionality reduction features of the input at the current moment. This indicates the hidden state at the previous moment. This represents the cell state at the previous moment. For activation function, The hyperbolic tangent activation function is used. This represents element-wise multiplication (Hadamard product). , , , These represent the weight matrices for the corresponding gating structure and state update, respectively. , , , These represent the corresponding bias vectors, and the final output feature state. This is the predicted state, which serves as the core basis for subsequent early warning decisions.

[0056] It should be understood that the belt conveyor status early warning method in this embodiment of the invention can be implemented in any system based on belt conveyor operation status monitoring, such as belt conveyor intelligent monitoring terminals, fault early warning servers, or mining integrated management and control platforms. Specifically, the belt conveyor status early warning method is pre-stored in the memory of the aforementioned device in the form of a computer program. When the device is powered on or the system triggers a monitoring task, the program is called and executed by the processor to achieve dynamic monitoring and early warning of belt conveyor misalignment, wear, and other faults.

[0057] Compared with existing technologies, the belt conveyor status early warning method provided in this invention constructs a comprehensive screening index system composed of monotonicity, robustness, and trend indicators. Monotonicity ensures that the feature changes monotonically as the fault worsens, avoiding misleading reverse fluctuations; robustness suppresses false jumps caused by impact noise; and trend ensures that the feature is synchronized with the long-term degradation process. The multi-domain target features selected through this comprehensive screening inevitably possess stable, monotonic, and time-dependent characteristics, thereby eliminating inferior interference at the data source, retaining true fault information, significantly enhancing the sensitivity to early, minor faults, and thus providing a reliable data foundation for subsequent accurate belt conveyor status early warning.

[0058] Furthermore, this embodiment of the invention employs kernel principal component analysis (KPCA) to perform nonlinear dimensionality reduction on multi-domain target features. KPCA can fully uncover the low-dimensional essential features hidden in strong nonlinear relationships, while removing relevant noise and redundant information. This avoids information loss and feature distortion under nonlinear conditions, as is the case with traditional linear dimensionality reduction methods (such as PCA). While ensuring the representativeness of the multi-domain dimensionality reduction features, redundant features are proposed, thereby improving the efficiency of the condition warning while ensuring the accuracy of the belt conveyor condition warning.

[0059] Furthermore, this embodiment of the invention uses a long short-term memory network model for state prediction, and utilizes its gating mechanism to capture the long-term evolution trend of belt conveyor misalignment or wear, overcoming the defect of static networks in being unable to model time dependence, and further improving the accuracy of state and early warning.

[0060] In some embodiments of the present invention, such as Figure 2 As shown, step S102 includes: S201. Calculate the monotonicity index, robustness index, and trend index of each feature in the multi-domain features.

[0061] Specifically, the formulas for calculating the monotonicity index, robustness index, and trend index are as follows:

[0062]

[0063]

[0064]

[0065] In the formula, It is a monotonicity indicator; For robustness indicators; It is a trend indicator; The number of samples in the feature sequence F; Take the derivative of the characteristic sequence F with respect to time; for The number of; for The number of; Let i be the i-th feature at time t; The residual of the average trend of the i-th feature at time t and the i-th feature at time t after exponentially weighted moving average smoothing is given. The average trend of the i-th feature at time t after exponentially weighted moving average smoothing; It is the difference between the rank of the i-th feature and the time series and feature sequence F after sorting them in order.

[0066] The difference in rank between the time series and the feature sequence F after sorting is defined as follows: First, time points are assigned ranks in natural order (rank 1 for the first time point, rank 2 for the second, and so on). Then, the values ​​in the feature sequence F are sorted from smallest to largest, and the rank of each feature value at its original position is obtained after sorting (if the values ​​are equal, the average rank is taken). For the same original time point i, the difference between the time rank and the feature rank at that point is the rank difference. This difference reflects the degree of deviation between the order of feature values ​​and the chronological order at that time point: the smaller the difference, the stronger the monotonic trend of the feature over time; the larger the difference, the weaker the correlation between the feature and time.

[0067] S202. The weighted sum of the monotonicity index value, the robustness index value, and the trend index value is used as the comprehensive index value.

[0068] Specifically, the formula for calculating the comprehensive index value is as follows:

[0069] In the formula, For comprehensive screening indicators; , , The weights are assigned to monotonicity indicators, robustness indicators, and trend indicators.

[0070] S203. Features whose comprehensive index value is greater than the index threshold are used as multi-domain target features.

[0071] It should be understood that the threshold values ​​for indicators can be set or adjusted according to the actual application scenario, and no specific restrictions are imposed here.

[0072] In some embodiments of the present invention, such as Figure 3 As shown, step S103 includes: S301. Implicitly map multi-domain target features to a high-dimensional feature space using a preset kernel function.

[0073] Kernel functions include, but are not limited to, Gaussian kernel functions and polynomial kernel functions.

[0074] S302. Perform principal component analysis in the high-dimensional feature space and extract the principal components as multi-domain dimensionality reduction features.

[0075] This invention implicitly maps multi-domain target features to a high-dimensional feature space using kernel functions, enabling data that was originally strongly nonlinearly coupled in the low-dimensional space to become linearly separable, thus overcoming the limitations of traditional linear dimensionality reduction methods. Based on this, principal component analysis is performed, which can remove redundancy and noise while retaining the core information of fault evolution, obtaining low-dimensional, decorrelation-resistant, and interference-resistant multi-domain dimensionality reduction features, providing high-quality input for subsequent time series prediction models.

[0076] In a specific embodiment of the present invention, step S301 includes: The kernel function is used to calculate the kernel function value between any two features in the multi-domain target features, and a kernel matrix is ​​constructed based on the kernel function value. The kernel matrix is ​​used to describe the inner product relationship between sample points in the high-dimensional feature space. The kernel matrix is ​​then centered to obtain a centered kernel matrix.

[0077] Specifically, if the multi-domain target features obtained after screening are represented as ,in For multi-domain target features, the feature parameter is a parameter of length . m The vector, n The dimension of the multi-domain target features is the number of features contained within them. A nonlinear function is used to map the multi-domain target features onto a high-dimensional space F to obtain... And assuming the mapping has been decentralized, then The covariance matrix can be expressed as: (1) In the formula: Since it is a coefficient, it can be ignored.

[0078] covariance matrix The eigenvalue decomposition is expressed as: (2) In the formula: Covariance matrix eigenvectors, These are the corresponding eigenvalues.

[0079] According to equations (1) and (2), the eigenvalues ​​can be expressed as: (3) because As a scalar, there exists a set of coefficients. = , can make eigenvalues Represented as: (4) Substituting equation (4) into equation (2), multiply both sides by . ,make Then equation (2) can be further expressed as: (5) In the formula: It is an n×n symmetric positive semi-definite matrix; for eigenvectors.

[0080] set up eigenvalues Dimensionality reduction is achieved using the first p eigenvalues, normalizing the features in space F, i.e. According to equations (4) and (5), we can obtain: (6) The principal components in a high-dimensional space F are... The projection onto the eigenvectors of the covariance matrix can be obtained from the matrix and its eigenvectors: (7) In the formula: Representing vectors The j-th value.

[0081] matrix Obtained from the kernel function, it can be expressed as: (8) The above calculations are performed under the premise of decentralization. After decentralization, the matrix can be represented as: (9) In the formula: .

[0082] The centered kernel matrix can be obtained based on the above process.

[0083] In specific embodiments of the present invention, such as Figure 4 As shown, step S302 includes: S401. Perform eigenvalue decomposition on the centered kernel matrix to obtain eigenvalues ​​and corresponding eigenvectors.

[0084] Specifically, based on formula (5), eigenvalue decomposition of the centered kernel matrix can be achieved.

[0085] S402. Calculate the contribution rate of each feature value, sort the feature values ​​in descending order of contribution rate, and select the feature vector corresponding to the feature value whose cumulative contribution rate is greater than the contribution rate threshold as the target feature vector. S403. Project the multi-domain target features onto the direction of the target feature vector to obtain multi-domain dimensionality reduction features.

[0086] Specifically, based on formula (7), the projection of the multi-domain target features onto the direction of the target feature vector can be obtained, which is the multi-domain dimensionality reduction feature.

[0087] The contribution rate of the k-th eigenvalue is calculated using the following formula: (10) In the formula, This is the k-th eigenvalue; Let be the i-th eigenvalue; n is the total number of eigenvalues.

[0088] Specifically, the contribution rate expresses the ratio of the effective information contained in the feature value to the information contained in the high-dimensional space.

[0089] It should be understood that the contribution rate threshold can be set or adjusted according to the actual application scenario, and no specific limit is made here.

[0090] In summary, the belt conveyor status early warning method proposed in this invention has the following advantages: First, by using a multi-domain feature extraction and quantitative feature comprehensive screening index system, it can effectively eliminate inferior feature interference caused by heavy load impact at the data source, significantly improving its anti-noise and anti-interference capabilities. Second, by employing the KPCA nonlinear dimensionality reduction method, it can map the highly nonlinearly coupled multivariate belt signals to a high-dimensional space for thorough redundancy removal, greatly preserving the true fault evolution structure of the original data and avoiding feature distortion problems. Third, by utilizing the powerful time series processing capabilities of Long Short-Term Memory (LSTM) networks, it can not only identify the current belt misalignment status but also accurately predict future belt deterioration trends based on hidden historical evolution patterns, gaining valuable time margin for the intervention of automatic flexible belt correction equipment and ensuring the safe and efficient operation of the coal mine transportation system.

[0091] On the other hand, embodiments of the present invention also provide a belt conveyor status early warning device, such as... Figure 5 As shown, the belt conveyor status early warning device 500 includes: The feature extraction unit 501 is used to acquire the multi-source dynamic time-series signal of the belt conveyor and extract the multi-domain features of the multi-source dynamic time-series signal. The feature selection unit 502 is used to select multi-domain features based on a comprehensive selection index constructed from monotonicity index, robustness index, and trend index to obtain multi-domain target features. The monotonicity index is used to characterize the degree to which a feature exhibits a monotonically increasing or decreasing trend with time series, reflecting the irreversibility of fault evolution. The robustness index is used to characterize the feature's ability to resist instantaneous shocks and random noise interference. The trend index is used to characterize the correlation between the feature and the running time, reflecting the feature's ability to follow the long-term degradation process. Nonlinear dimensionality reduction unit 503 is used to perform nonlinear dimensionality reduction on multi-domain target features based on kernel principal component analysis to obtain multi-domain dimensionality reduction features. The state warning unit 504 is used to input multi-domain dimensionality reduction features into the long short-term memory network model to obtain the predicted state of the belt conveyor and to issue warnings based on the predicted state.

[0092] The belt conveyor status early warning device 500 provided in the above embodiments can realize the technical solutions described in the above belt conveyor status early warning method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above belt conveyor status early warning method embodiments, and will not be repeated here.

[0093] like Figure 6 As shown, the present invention also provides a belt conveyor status early warning device 600. The belt conveyor status early warning device 600 includes a data acquisition module 601, a processor 602, a memory 603, and a display 604. Figure 6 Only some components of the belt conveyor condition warning device 600 are shown. However, it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0094] The data acquisition module 601 is used to acquire multi-source dynamic timing signals of the belt conveyor.

[0095] In some embodiments, processor 602 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in memory 603 or process data, such as the belt conveyor status early warning method of the present invention.

[0096] In some embodiments, the memory 603 may be an internal storage unit of the conveyor belt status warning device 600, such as a hard disk or memory of the conveyor belt status warning device 600. In other embodiments, the memory 603 may also be an external storage device of the conveyor belt status warning device 600, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the conveyor belt status warning device 600.

[0097] Furthermore, the memory 603 may include both internal storage units of the belt conveyor status early warning device 600 and external storage devices. The memory 603 is used to store the application software and various data of the belt conveyor status early warning device 600.

[0098] In some embodiments, display 604 may be an LED display, a liquid crystal display, a touch-screen liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 604 is used to display information from the conveyor belt status warning device 600 and to display a visual user interface. Components 601-604 of the conveyor belt status warning device 600 communicate with each other via a device bus.

[0099] In some embodiments of the present invention, when the processor 602 executes the belt conveyor status early warning program in the memory 603, the following steps can be implemented: Acquire multi-source dynamic time-series signals of the belt conveyor and extract multi-domain features of the multi-source dynamic time-series signals; Feature selection is performed on multi-domain features based on a comprehensive screening index constructed from monotonicity, robustness, and trend indicators to obtain multi-domain target features. The monotonicity index is used to characterize the degree to which a feature exhibits a monotonically increasing or decreasing trend over time, reflecting the irreversibility of fault evolution. The robustness index is used to characterize the ability of a feature to resist instantaneous shocks and random noise interference. The trend indicator is used to characterize the correlation between a feature and runtime, reflecting the feature's ability to follow long-term degradation processes. Multi-domain target features are nonlinearly reduced in dimensionality based on kernel principal component analysis to obtain multi-domain dimensionality-reduced features. Multi-domain dimensionality reduction features are input into a long short-term memory network model to obtain the predicted state of the belt conveyor, and early warning is issued based on the predicted state.

[0100] It should be understood that when the processor 602 executes the belt conveyor status warning program in the memory 603, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.

[0101] Accordingly, this application also provides a computer-readable storage medium for storing a computer-readable program or instruction. When the program or instruction is executed by a processor, it can implement the method steps or functions provided in the above-described method embodiments.

[0102] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0103] The present invention provides a detailed description of a belt conveyor status early warning method, device, equipment, and storage medium. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A belt conveyor state early warning method, characterized in that, include: Acquire multi-source dynamic time-series signals of the belt conveyor and extract multi-domain features of the multi-source dynamic time-series signals; The multi-domain features are filtered based on a comprehensive screening index constructed from monotonicity, robustness, and trend indicators to obtain multi-domain target features. The monotonicity index is used to characterize the degree to which a feature exhibits a monotonically increasing or decreasing trend over time, reflecting the irreversibility of fault evolution; the robustness index is used to characterize the feature's ability to resist instantaneous shocks and random noise interference; the trend index is used to characterize the correlation between the feature and runtime, reflecting the feature's ability to follow long-term degradation processes. Based on kernel principal component analysis, nonlinear dimensionality reduction is performed on the multi-domain target features to obtain multi-domain dimensionality-reduced features. The multi-domain dimensionality reduction features are input into a long short-term memory network model to obtain the predicted state of the belt conveyor, and an early warning is issued based on the predicted state.

2. The belt conveyor state early warning method according to claim 1, characterized in that, The multi-domain feature selection is performed based on a comprehensive screening index constructed from monotonicity, robustness, and trend indicators to obtain multi-domain target features, including: Calculate the monotonicity index, robustness index, and trend index of each feature in the multi-domain features; The weighted sum of the monotonicity index value, the robustness index value, and the trend index value is used as the comprehensive index value; Features whose comprehensive index value is greater than the index threshold are used as the multi-domain target features.

3. The belt conveyor state early warning method according to claim 1 or 2, characterized in that, The comprehensive screening criteria are as follows: In the formula, For comprehensive screening indicators; It is a monotonicity indicator; For robustness indicators; It is a trend indicator; , , Weights for monotonicity indicators, robustness indicators, and trend indicators; The number of samples in the feature sequence F; Take the derivative of the characteristic sequence F with respect to time; for The number of; for The number of; Let i be the i-th feature at time t; The residual of the average trend of the i-th feature at time t and the i-th feature at time t after exponentially weighted moving average smoothing is given. The average trend of the i-th feature at time t after exponentially weighted moving average smoothing; It is the difference between the rank of the i-th feature and the time series and feature sequence F after sorting them in order.

4. The belt conveyor condition early warning method according to claim 1, characterized in that, The nonlinear dimensionality reduction of the multi-domain target features based on kernel principal component analysis to obtain multi-domain dimensionality-reduced features includes: The multi-domain target features are implicitly mapped to a high-dimensional feature space using a preset kernel function; Principal component analysis is performed in the high-dimensional feature space, and the principal components are extracted as the multi-domain dimensionality reduction features.

5. The belt conveyor condition early warning method according to claim 4, characterized in that, The implicit mapping of the multi-domain target features to a high-dimensional feature space using a preset kernel function includes: The kernel function is used to calculate the kernel function value between any two features in the multi-domain target features, and a kernel matrix is ​​constructed based on the kernel function value. The kernel matrix is ​​used to describe the inner product relationship between sample points in the high-dimensional feature space. The kernel matrix is ​​centered to obtain a centered kernel matrix.

6. The belt conveyor condition early warning method according to claim 5, characterized in that, The step of performing principal component analysis in the high-dimensional feature space and extracting principal components as the multi-domain dimensionality reduction features includes: The eigenvalues ​​and corresponding eigenvectors are obtained by performing eigenvalue decomposition on the centered kernel matrix. Calculate the contribution rate of each feature value, sort the feature values ​​in descending order of contribution rate, and select the feature vector corresponding to the feature value whose cumulative contribution rate is greater than the contribution rate threshold as the target feature vector; The multi-domain target features are projected onto the direction of the target feature vector to obtain the multi-domain dimensionality reduction features.

7. The belt conveyor condition early warning method according to claim 1, characterized by, The multi-domain features include time-domain features, frequency-domain features, and time-frequency-domain features. The time-domain features include dimensional time-domain features and dimensionless time-domain features. The dimensional time-domain features include root mean square value, mean, variance, absolute mean, root mean amplitude, peak value, and peak-to-peak value. The dimensionless time-domain features include skewness index, kurtosis index, peak value index, margin index, impulse index, and waveform index. The frequency-domain features include spectral amplitude mean, spectral amplitude variance, spectral frequency centroid, second-order centroid, root mean square frequency, and spectral amplitude skewness. The time-frequency-domain features include the energy percentage of each node in the last layer after performing wavelet packet decomposition of the multi-source dynamic time-series signal at a specified number of layers.

8. A belt conveyor state early warning device, characterized in that, include: The feature extraction unit is used to acquire the multi-source dynamic time-series signal of the belt conveyor and extract the multi-domain features of the multi-source dynamic time-series signal; The feature filtering unit is used to filter the multi-domain features based on a comprehensive filtering index constructed from monotonicity index, robustness index, and trend index to obtain multi-domain target features. The monotonicity index is used to characterize the degree to which a feature exhibits a monotonically increasing or decreasing trend over time, reflecting the irreversibility of fault evolution; the robustness index is used to characterize the feature's ability to resist instantaneous shocks and random noise interference; the trend index is used to characterize the correlation between the feature and runtime, reflecting the feature's ability to follow long-term degradation processes. A nonlinear dimensionality reduction unit is used to perform nonlinear dimensionality reduction on the multi-domain target features based on kernel principal component analysis to obtain multi-domain dimensionality-reduced features. The state warning unit is used to input the multi-domain dimensionality reduction features into the long short-term memory network model to obtain the predicted state of the belt conveyor, and to issue a warning based on the predicted state.

9. A belt conveyor state early warning device characterized by, It includes a data acquisition module, a memory, and a processor, among which, The data acquisition module is used to acquire multi-source dynamic timing signals of the belt conveyor; The memory is used to store programs; The processor, coupled to the memory and the data acquisition module, is used to execute the program stored in the memory to implement the steps in the belt conveyor status early warning method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the belt conveyor status early warning method according to any one of claims 1 to 7.

Images