A fault positioning and identification method for underground coal mine equipment

Abstract

The application provides a fault positioning and identification method for underground coal mine equipment, which comprises the following steps: S1: distributed sensing optical fibers are laid in the underground coal mine to obtain operation data of the equipment to be measured, the real-time collected operation data is preprocessed, and characteristic parameters are extracted; S2: various faults of the equipment to be measured are simulated, the sensing optical cables laid are used to collect transportation data corresponding to various faults, a fault sample library is established, and a fault prediction model is constructed based on the fault sample library; S3: based on the characteristic parameters, the fault prediction model is used to identify the faults of the operation state of the equipment to be measured, and the spatial position of the fault point is positioned according to the spatial position information of the distributed sensing optical fibers; and S4: the residual life of the equipment to be measured at the fault point is detected, and targeted maintenance is performed according to the detection result. The application has comprehensive monitoring information, high fault identification accuracy, and can accurately position the fault source.

Description

Technical Field

[0001] This invention belongs to the field of safety monitoring technology, and relates to coal mine safety monitoring, and particularly to a method for fault location and identification of underground coal mine equipment. Background Technology

[0002] Fault diagnosis of underground equipment in coal mines is a crucial link in ensuring safe production. When abnormalities occur in underground equipment, if the source of the fault cannot be located with high precision, maintenance personnel will be unable to quickly locate the abnormal equipment in the narrow and dimly lit underground environment, which may even lead to irreparable catastrophic consequences for the mine.

[0003] Currently, monitoring the operational status of underground equipment mainly relies on manual inspections or the installation of various sensors on the equipment. However, due to the harsh environment of high humidity, high dust, and strong electromagnetic interference in coal mines, electrical sensors are susceptible to electromagnetic interference, leading to signal distortion. Prolonged exposure to humid environments also causes corrosion and aging. Furthermore, the inherent safety requirements are high, and power supply is limited, resulting in poor reliability of traditional monitoring solutions and frequent false alarms and missed alarms. Traditional methods are mostly "passive" troubleshooting after a fault occurs, making it difficult to achieve early warning and accurate location.

[0004] In recent years, distributed fiber optic sensing technology has been gradually introduced into the field of coal mine equipment condition monitoring due to its advantages such as resistance to electromagnetic interference, corrosion resistance, and passive distributed monitoring. However, existing fiber optic-based monitoring methods are mostly limited to the detection of single physical quantities, failing to comprehensively characterize the multidimensional features of equipment faults, resulting in low fault diagnosis accuracy and a high false alarm rate. Currently, the potential of fiber optics to simultaneously sense multiple physical fields has not been effectively utilized, leading to problems such as incomplete information, low identification accuracy, and difficulty in precise location during fault diagnosis.

[0005] Therefore, improving the accuracy of fault identification, accurately locating fault points, and reducing safety risks in coal mine production have become the focus of existing safety monitoring operations. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the present invention aims to provide a method for fault location and identification of underground coal mine equipment, solving the problems of incomplete monitoring information, low fault identification accuracy, and inability to accurately locate fault sources caused by relying on a single type of fiber optic sensor data in existing technologies.

[0007] To achieve this objective, the present invention adopts the following technical solution: This invention provides a method for fault location and identification of underground equipment in coal mines, the method comprising: S1: Deploy distributed sensing optical fibers underground in the coal mine to acquire the operating data of the device under test, preprocess the real-time acquired operating data, and extract feature parameters. S2: Simulate various faults of the equipment under test, collect transportation data corresponding to various faults using the deployed sensing optical cables, establish a fault sample library, and build a fault prediction model based on the fault sample library. S3: Based on feature parameters, use the fault prediction model to identify faults in the operating status of the equipment under test, and locate the fault point spatially based on the spatial location information of the distributed sensing fiber. S4: Then perform a remaining life test on the device under test at the fault point, and carry out targeted maintenance based on the test results.

[0008] As a preferred technical solution of the present invention, in step S1, the operating data includes at least temperature data, vibration data and pressure data.

[0009] As a preferred embodiment of the present invention, in step S1, the preprocessing includes: Singular value decomposition and reconstructed signal processing are performed on the running data to obtain the noise-reduced running data; Then use linear interpolation or spline interpolation methods to fill in the missing values ​​in various types of data; Statistical methods are used to identify and remove outlier data that exceeds a reasonable range; Then, the min-max normalization method is used to unify the data format and eliminate the differences between different units.

[0010] As a preferred embodiment of the present invention, step S1, the extraction and processing of the feature parameters includes the following steps: Edge computing nodes are used to extract time-domain, frequency-domain, and time-frequency-domain features from vibration data to form vibration feature vectors; Based on the least squares sliding window method, feature extraction is performed on temperature data and pressure data respectively to obtain feature vectors of temperature change rate and pressure change rate. A temperature distribution feature vector is formed by extracting spatial temperature difference features from multiple measurement points; Deep feature fusion of multiple feature vectors is performed to obtain multi-source fused data.

[0011] As a preferred technical solution of the present invention, a convolutional neural network is used to enhance multi-parameter fused data.

[0012] As a preferred technical solution of the present invention, in step S2, the fault sample library includes abnormal conditions, usage time and maintenance frequency of conveyors, motors and fans; The abnormal conditions include at least one of the following: idler roller jamming, idler roller damage, idler roller shaft breakage, bracket loosening, bracket deformation, bearing failure, rotor imbalance, rotor misalignment, blade damage, idler roller jamming, bearing overheating, and motor overload.

[0013] As a preferred technical solution of the present invention, in step S3, the fault prediction model is trained based on the fault sample data corresponding to the fault type in the fault sample library, using the Bayesian learning algorithm as the training framework, and the fault prediction model is iteratively trained using superimposed weight analysis to obtain an optimized fault prediction model.

[0014] As a preferred technical solution of the present invention, in step S3, the spatial coordinate information of the abnormal sensing fiber corresponding to the fault identification result is obtained, the spatial position of the fault point along the fiber deployment path is determined according to the spatial coordinate information, and mapped to the three-dimensional spatial coordinates of the underground equipment in the coal mine, and the spatial positioning result of the fault point is output.

[0015] As a preferred embodiment of the present invention, the fault location and identification method further includes: The fault identification result is associated and bound with the fault point spatial location result to generate fault alarm information containing fault type label and three-dimensional spatial coordinates of fault point; The fault alarm information is sent to the ground monitoring center and displayed in a visual manner on the distribution map of underground equipment in the coal mine.

[0016] As a preferred technical solution of the present invention, in step S4, the remaining lifetime detection includes: performing time-series regression prediction on the feature parameters at the fault point based on hybrid deep learning, and outputting the remaining lifetime prediction value.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves distributed, passive, multi-physics synchronous monitoring of the entire length of equipment by deploying sensing optical fibers along the equipment line. This reduces the risk of electromagnetic interference and corrosion that traditional sensors are susceptible to underground, significantly improving the reliability and security of the collected data. At the same time, through the comprehensive characteristic indicators of multi-parameter fusion, the fault identification capability is significantly improved, effectively reducing the risk of false alarms and missed alarms. It also enables precise spatial positioning of fault points. By performing remaining life detection on the faulty equipment, staff can make targeted maintenance decisions, thereby shortening the fault investigation time and improving the safety of coal mine production. Detailed Implementation

[0018] It should be understood that in the methods described in the embodiments of this invention, the order in which the steps are written does not imply a strict execution order. The actual execution order of each step should be determined according to its function and possible internal logic. Unless otherwise specified, all steps of this invention can be executed in the order they are written, or in any order without technical conflict. For example, if the method includes steps (a) and (b), it means that the method may include steps (a) and (b) executed sequentially, or it may include steps (b) and (a) executed sequentially. If the method also includes step (c), then step (c) can be added to the method in any order without conflict, including but not limited to the execution order of steps (a), (b), and (c), steps (a), (c), and (b), steps (c), (a), and (b), etc.

[0019] The technical solution of the present invention will be further illustrated below through specific embodiments.

[0020] This invention provides a method for fault location and identification of underground equipment in coal mines, specifically including the following: S1: Deploy distributed sensing optical fibers underground in the coal mine to acquire the operating data of the device under test, preprocess the real-time acquired operating data, and extract feature parameters.

[0021] The distributed sensing optical fibers are deployed along the underground line, covering key monitoring areas to form a sensing network. For example, the key areas include, but are not limited to, the equipment's windings, bearings, rotors, housings, supports, idlers, and blades.

[0022] The operational data includes at least temperature, vibration, and pressure data. Motor overheating is a common cause of many electrical and mechanical failures; monitoring temperature data helps prevent equipment burnout. Analysis of the vibration signals provides data support for the early detection of damage to core transmission components such as bearings and rotors. The pressure data reflects the equipment's load status and fluid sealing condition. This invention achieves continuous, distributed, and high-precision synchronous monitoring of multiple physical quantities through distributed sensing optical fibers.

[0023] In some implementations, the preprocessing includes: S01: Perform singular value decomposition and reconstruct signal processing on the running data to obtain the noise-reduced running data; S02: Then use linear interpolation or spline interpolation methods to fill in the missing values ​​in various types of data; S03: Identify and remove abnormal data that exceeds a reasonable range based on statistical methods; S04: Then, the minimum-maximum normalization method is used to unify the data format and eliminate the differences between different units.

[0024] In S01, singular value decomposition is used to construct the Hankel matrix from the collected multi-dimensional time-series signals such as vibration data, temperature data, and pressure data, and eigenvalue decomposition is performed. This can effectively suppress downhole environmental noise and electromagnetic interference while preserving the original fault characteristic frequencies to the maximum extent. This provides a high signal-to-noise ratio data foundation for subsequent multi-parameter feature extraction and fusion, and significantly reduces the misjudgment of fault identification due to environmental interference.

[0025] The linear interpolation method assumes a linear relationship between two adjacent known data points, determines the equation of a straight line using these two points, and then calculates the estimated value for any missing point between them. The spline interpolation method divides all known data points into several intervals, fits each interval with a low-order polynomial (usually a cubic polynomial), and mandates that the polynomials of adjacent intervals have continuous first and second derivatives at the connection points, thus ensuring the smoothness of the overall interpolation curve. Those skilled in the art can select either method to fill in missing values ​​in the collected data based on the actual data characteristics and processing requirements, ensuring the integrity and continuity of the data sequence.

[0026] This invention employs a robust interquartile range method to set time-series data of single physical quantities such as temperature, vibration, and pressure, adaptively adapting to threshold boundaries, resisting extreme value interference, and automatically eliminating noise samples caused by occasional sensor drift or instantaneous impact.

[0027] In some implementations, the feature parameter extraction process includes the following steps: S11 uses edge computing nodes to extract time-domain, frequency-domain, and time-frequency-domain features from vibration data, forming a vibration feature vector; S12 uses the least squares sliding window method to extract features from temperature and pressure data respectively, and obtains feature vectors of temperature change rate and pressure change rate. S13 extracts spatial temperature difference features from multiple measurement points to form a temperature distribution feature vector; S14 performs deep feature fusion on multiple feature vectors to obtain multi-source fused data.

[0028] The time-domain features described in S11 of this invention include peak value, root mean square value, absolute mean, waveform factor, kurtosis, peak factor, etc.; the frequency-domain features are obtained by performing a fast Fourier transform on the data to convert it into a frequency-domain signal, including centroid frequency, mean square frequency, and spectral amplitude, etc.; the time-frequency domain features are obtained by using wavelet packet decomposition based on multi-resolution analysis to calculate the energy proportion of wavelet nodes in each frequency band, thereby enhancing the ability to identify weak pulse signals such as equipment damage.

[0029] In step S12 of this invention, sampling sequences for temperature and pressure are set separately. Covering at least two complete frequency cycles of the equipment, and based on the statistical stability reflecting the operating state, a window length and sliding step size are set. Starting from the beginning of the sequence, the window slides with the set sliding step size to obtain multiple analysis windows, each containing multiple data points. Then, least-squares linear fitting is used to fit a first-order linear function to the data in each window, obtaining the slope and intercept parameters for that window. Finally, the fitting results for each window are used to construct a feature vector, which is then arranged in chronological order to form a feature sequence. All calculations involved above employ computational functions or formulas known in the art.

[0030] The deep feature fusion in S14 of this invention is based on an attention-based feature aggregation architecture. Initial concatenation is performed at the input layer, followed by deep semantic feature extraction through multiple layers of networks and then fusion again. Finally, the fusion result is optimized and adjusted using an attention mechanism. The specific network structures, inputs and outputs, parameter values, operating procedures, and calculation processes mentioned above are all known in the art, and this invention does not impose specific limitations on them.

[0031] The vibration feature vector is used to identify idler roller jamming, bearing wear, rotor imbalance, and gear failure; the temperature rise rate of the temperature data is used to identify idler roller jamming, bearing overheating, and motor overload failure; the pressure change rate feature vector is used to identify frame deformation, support loosening, and belt damage failure.

[0032] In some embodiments, the present invention also utilizes convolutional neural networks to enhance multi-parameter fused data. Specifically, the execution steps include constructing a pre-enhancement dataset, noise injection enhancement, graph transformation enhancement, and CNN training and incremental output. The specific network structures, inputs and outputs, parameter values, operating procedures, and calculation processes mentioned above are all known in the art, and the present invention does not specifically limit them.

[0033] S2: Simulate various faults of the device under test, collect transportation data corresponding to various faults using deployed optical sensing cables, establish a fault sample library, and build a fault prediction model based on the fault sample library.

[0034] The fault sample database includes abnormal conditions, usage duration, and maintenance frequency of conveyors, motors, and fans.

[0035] The abnormal conditions include at least one of the following: idler roller jamming, idler roller damage, idler roller shaft breakage, bracket loosening, bracket deformation, bearing failure, rotor imbalance, rotor misalignment, blade damage, idler roller jamming, bearing overheating, and motor overload.

[0036] S3: Based on feature parameters, the fault prediction model is used to identify faults in the operating status of the equipment under test, and the fault point is spatially located based on the spatial location information of the distributed sensing fiber.

[0037] In some embodiments, the present invention trains the fault prediction model based on fault sample data corresponding to fault types in the fault sample library using a Bayesian learning algorithm as the training framework, and iteratively trains the fault prediction model using superimposed weight analysis to obtain an optimized fault prediction model.

[0038] Specifically, the training and optimization of the fault prediction model are performed according to the following steps: S31: Extract sample data corresponding to each fault type from the fault sample library, and divide it into training set and validation set according to the proportion. Each sample contains multi-dimensional feature vectors such as temperature, vibration and pressure obtained by the aforementioned feature extraction steps, as well as the corresponding fault type label. S32: Construct a three-layer Bayesian fully connected network, with the input layer dimension equal to the feature vector dimension, the hidden layer containing 64 neurons, and the output layer dimension equal to the number of fault types. S33: Input the training set into the Bayesian network described above, and train it using the Adam optimizer with ELBO as the objective function; S34: Superimposed weight analysis and iterative correction. After each weight correction iteration, the corrected weights are used as the initial mean parameters of the variational distribution. Variational inference training is re-executed, and the contribution of each weight is re-evaluated based on the updated posterior distribution. S35: The Bayesian network at the end of the iteration is the optimized fault prediction model.

[0039] In some implementations, the spatial coordinate information of the abnormal sensing fiber corresponding to the fault identification result is obtained, the spatial location of the fault point along the fiber deployment path is determined based on the spatial coordinate information, and mapped to the three-dimensional spatial coordinates of the underground coal mine equipment, and the spatial positioning result of the fault point is output.

[0040] In some embodiments, the fault location and identification method further includes: The fault identification result is associated and bound with the fault point spatial location result to generate fault alarm information containing fault type label and three-dimensional spatial coordinates of fault point; The fault alarm information is sent to the ground monitoring center and displayed in a visual manner on the distribution map of underground equipment in the coal mine.

[0041] S4: Then perform a remaining life test on the device under test at the fault point, and carry out targeted maintenance based on the test results.

[0042] This invention analyzes the collected information to determine if the equipment has malfunctioned. If so, it identifies the fault point of the component, monitors the lifespan of the malfunctioning equipment, and determines whether the malfunctioning equipment needs to be replaced. Finally, a solution is provided manually through online monitoring equipment.

[0043] In some implementations, the remaining lifetime detection includes: performing temporal regression prediction on the feature parameters at the fault point based on hybrid deep learning, and outputting a predicted remaining lifetime value.

[0044] Specifically, this invention employs a hybrid deep learning model cascaded with CNN and LSTM to perform temporal regression prediction of feature parameters at the fault point. The specific steps include: S41: Spatial feature extraction is performed on the input multi-channel feature tensor using a CNN sub-network. The input data is a historical feature sequence with the current time as the endpoint and a backtracking time of a set duration, such as 30 minutes or 60 minutes. This sequence consists of three channels of features: temperature, vibration, and pressure. The features are aligned according to the sampling time to form a multi-dimensional input tensor. S42: Receive the feature sequence output by the CNN through the LSTM sub-network and use its gating mechanism to capture the long-term dependencies of the feature sequence in the time dimension; S43: Edge nodes slide to extract the feature sequences of the most recent 30 or 60 minutes with a period of 5 or 10 minutes and input them into the CNN-LSTM hybrid model to output the remaining life prediction in real time.

[0045] The CNN sub-network structure includes three convolutional layers and three pooling layers, enabling the extraction of spatial local features and compression of sequence length. The LSTM sub-network structure includes a two-layer LSTM layer, a fully connected layer, and an output layer. The operation of these network structures is known in the art, and this invention does not specifically limit their implementation.

[0046] This invention achieves distributed, passive, multi-physics synchronous monitoring of the entire length of equipment by deploying sensing optical fibers along the equipment line. This reduces the risk of electromagnetic interference and corrosion that traditional sensors are susceptible to underground, significantly improving the reliability and security of the collected data. At the same time, through the comprehensive characteristic indicators of multi-parameter fusion, the fault identification capability is significantly improved, effectively reducing the risk of false alarms and missed alarms. It also enables precise spatial positioning of fault points. By performing remaining life detection on the faulty equipment, staff can make targeted maintenance decisions, thereby shortening the fault investigation time and improving the safety of coal mine production.

[0047] The applicant declares that the above description is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Those skilled in the art should understand that any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention fall within the protection and disclosure scope of the present invention.

Claims

1. A method for fault location and identification of underground equipment in coal mines, characterized in that, The fault location and identification method includes: S1: Deploy distributed sensing optical fibers underground in the coal mine to acquire the operating data of the device under test, preprocess the real-time acquired operating data, and extract feature parameters. S2: Simulate various faults of the equipment under test, collect transportation data corresponding to various faults using the deployed sensing optical cables, establish a fault sample library, and build a fault prediction model based on the fault sample library. S3: Based on feature parameters, use the fault prediction model to identify faults in the operating status of the equipment under test, and locate the fault point spatially based on the spatial location information of the distributed sensing fiber. S4: Then perform a remaining life test on the device under test at the fault point, and carry out targeted maintenance based on the test results.

2. The fault location and identification method according to claim 1, characterized in that, In step S1, the operating data includes at least temperature data, vibration data, and pressure data.

3. The fault location and identification method according to claim 1 or 2, characterized in that, In step S1, the preprocessing includes: Singular value decomposition and reconstructed signal processing are performed on the running data to obtain the noise-reduced running data; Then use linear interpolation or spline interpolation methods to fill in the missing values ​​in various types of data; Statistical methods are used to identify and remove outlier data that exceeds a reasonable range; Then, the min-max normalization method is used to unify the data format and eliminate the differences between different units.

4. The fault location and identification method according to claim 3, characterized in that, In step S1, the feature parameter extraction process includes the following steps: Edge computing nodes are used to extract time-domain, frequency-domain, and time-frequency-domain features from vibration data to form vibration feature vectors; Based on the least squares sliding window method, feature extraction is performed on temperature data and pressure data respectively to obtain feature vectors of temperature change rate and pressure change rate. A temperature distribution feature vector is formed by extracting spatial temperature difference features from multiple measurement points; Deep feature fusion of multiple feature vectors is performed to obtain multi-source fused data.

5. The fault location and identification method according to claim 4, characterized in that, We use convolutional neural networks to enhance multi-parameter fused data.

6. The fault location and identification method according to claim 1, characterized in that, In step S2, the fault sample database includes abnormal conditions, usage duration, and maintenance frequency of conveyors, motors, and fans; The abnormal conditions include at least one of the following: idler roller jamming, idler roller damage, idler roller shaft breakage, bracket loosening, bracket deformation, bearing failure, rotor imbalance, rotor misalignment, blade damage, idler roller jamming, bearing overheating, and motor overload.

7. The fault location and identification method according to claim 1, characterized in that, In step S3, based on the fault sample data corresponding to the fault type in the fault sample library, the fault prediction model is trained using the Bayesian learning algorithm as the training framework, and the fault prediction model is iteratively trained using superimposed weight analysis to obtain an optimized fault prediction model.

8. The fault location and identification method according to claim 1, characterized in that, In step S3, the spatial coordinate information of the abnormal sensing fiber corresponding to the fault identification result is obtained. Based on the spatial coordinate information, the spatial location of the fault point along the fiber deployment path is determined and mapped to the three-dimensional spatial coordinates of the underground equipment in the coal mine. The spatial positioning result of the fault point is then output.

9. The fault location and identification method according to claim 1 or 8, characterized in that, The fault location and identification method further includes: The fault identification result is associated and bound with the fault point spatial location result to generate fault alarm information containing fault type label and three-dimensional spatial coordinates of fault point; The fault alarm information is sent to the ground monitoring center and displayed in a visual manner on the distribution map of underground equipment in the coal mine.

10. The fault location and identification method according to claim 1, characterized in that, In step S4, the remaining lifetime detection includes: performing temporal regression prediction on the feature parameters at the fault point based on hybrid deep learning, and outputting the remaining lifetime prediction value.