A method, system, device, and medium for fault recognition of a belt conveyor

Abstract

This application relates to a fault identification method, system, equipment, and medium for belt conveyors. The method includes: acquiring multi-channel vibration signals from the belt conveyor; mapping the multi-channel vibration signals into a two-dimensional spatiotemporal grayscale image; performing texture analysis on the two-dimensional spatiotemporal grayscale image to extract anisotropic texture features; analyzing the anisotropic texture features using a pre-trained classification model to obtain the fault type, the pre-trained classification model including a first classification model and a second classification model; determining the fault start position according to the fault type using a corresponding texture start position identification rule, and determining the fault severity level based on the anisotropic texture features; and outputting the fault identification result composed of the fault type, fault start position, and fault severity level to drive alarm prompts. This application improves the accuracy of fault identification for belt conveyors.

Description

Technical Field

[0001] This application relates to the technical field of intelligent monitoring and equipment fault diagnosis, and in particular to a fault identification method, system, equipment and medium for a belt conveyor. Background Technology

[0002] As core equipment in bulk material transportation industries such as coal, power, and ports, the operational stability of belt conveyors directly affects production efficiency and operational safety. Localized "point source failures" such as idler wear and bearing failure, and "propagating failures" that spread along the equipment's extension direction, such as belt misalignment, longitudinal tearing, and frame resonance, are two of the most typical types of failures. Among them, propagating failures have a wide impact and develop rapidly; if not identified accurately and promptly, they can easily lead to major safety accidents such as belt breakage, material spillage, and even fires, causing huge economic losses.

[0003] Existing fault monitoring solutions for belt conveyors based on DAS technology have solved the problem of insufficient coverage of traditional manual inspections or point sensors, but there are still significant technical bottlenecks that make it difficult to meet the needs of industrial sites for accurate fault classification and tracing.

[0004] The existing technical solutions mentioned above have the following drawbacks: most of the existing belt conveyor fault identification methods rely only on single-path feature extraction or fixed rule judgment, which cannot achieve high-precision type differentiation and starting position positioning, and lack a fusion identification mechanism that adapts to different fault characteristics, so there is room for improvement. Summary of the Invention

[0005] To improve the accuracy of fault identification for belt conveyors, this application provides a fault identification method, system, equipment, and medium for belt conveyors.

[0006] The above-mentioned objective of this application is achieved through the following technical solution:

[0007] A fault identification method for a belt conveyor, the fault identification method for a belt conveyor comprising:

[0008] Collect multi-channel vibration signals from a belt conveyor;

[0009] The multi-channel vibration signal is mapped into a two-dimensional spatiotemporal grayscale image;

[0010] Perform texture analysis on the two-dimensional spatiotemporal grayscale image to extract anisotropic texture features;

[0011] Based on the anisotropic texture features, a pre-trained classification model is used for analysis to obtain the fault type. The pre-trained classification model includes a first classification model and a second classification model.

[0012] According to the fault type, adopt the corresponding texture starting position recognition rule to determine the fault starting position, and determine the fault level based on the anisotropic texture features to obtain the fault severity level;

[0013] Form the fault recognition result by combining the fault type, the fault starting position and the fault severity level and output it for driving an alarm prompt.

[0014] By adopting the above technical solution, by collecting multi-channel vibration signals of the belt conveyor, it is possible to achieve full-coverage real-time perception of the operating state of the conveyor belt, thereby improving the comprehensiveness of fault recognition and the continuity of data acquisition; by mapping the multi-channel vibration signals into two-dimensional spatio-temporal grayscale images, the original time-series signals can be converted into information images with a spatial distribution structure, thereby providing a unified representation basis for subsequent image analysis and feature extraction; by performing texture analysis operations on the two-dimensional spatio-temporal grayscale images and extracting anisotropic texture features, it is possible to fully exploit the local structure and directional change information in the images, thereby improving the sensitivity and accuracy of fault type discrimination; by using the first classification model and the second classification model for analysis based on the anisotropic texture features to obtain the fault type, it is possible to integrate the advantages of artificial features and depth features, thereby significantly improving the fault recognition accuracy and adaptability under complex working conditions; by determining the fault starting position according to the fault type using the texture starting position recognition rule and determining the fault level based on the texture features, it is possible to achieve dual recognition of the fault position and severity, thereby providing a more targeted decision-making basis for equipment maintenance; by forming the fault recognition result by combining the fault type, the starting position and the severity level and outputting it, it is possible to form a structured output result for intelligent alarm and linkage response, thereby enhancing the intelligent early warning ability and operation and maintenance efficiency of the system.

[0015] In one example, the present application can be further configured as: the collecting of the multi-channel vibration signals of the belt conveyor specifically includes:

[0016] Collect the vibration information of the belt conveyor, perform a signal conversion operation on the vibration information, and output multi-channel vibration data in the triple format of optical fiber mileage, timestamp and vibration amplitude;

[0017] Perform filtering and noise reduction processing on the multi-channel vibration data, and the filtering and noise reduction processing includes power frequency notch filtering and moving average filtering to obtain the multi-channel vibration signals.

[0018] By adopting the above technical solution, and by collecting vibration information of the belt conveyor and outputting multi-channel vibration data in the format of fiber optic mileage, timestamp, and vibration amplitude triplet, the vibration intensity and spatial location at each moment can be accurately calibrated, thereby improving the traceability and accuracy of vibration data in the spatiotemporal dimension. By performing joint noise reduction processing of power frequency notch filtering and moving average filtering on the multi-channel vibration data, environmental interference and random noise can be effectively suppressed, thereby improving the signal quality and stability of downstream image mapping and feature extraction.

[0019] In one example, this application can be further configured such that mapping the multi-channel vibration signal into a two-dimensional spatiotemporal grayscale image specifically includes:

[0020] Set time window and spatial window parameters for image construction, and perform window segmentation processing on the multi-channel vibration signal based on the time window and spatial window parameters;

[0021] In each combination of spatial channel and time window, the vibration sampling sequence in the corresponding channel is extracted, and the vibration energy value of the vibration sampling sequence is calculated.

[0022] The vibration energy value is normalized and mapped to a gray value, and the gray value is filled into the corresponding image coordinate position to generate a two-dimensional spatiotemporal grayscale image;

[0023] The two-dimensional grayscale image is subjected to image enhancement processing, which includes histogram equalization.

[0024] By adopting the above technical solutions, and by setting the time and spatial window parameters required for image construction and performing window segmentation processing, long-term continuous signals can be reasonably divided into analysis units with consistent structures, thereby improving the standardization and alignment of image generation. By extracting the vibration sampling sequence under each channel and time window combination and calculating its vibration energy value, the original time-domain signal features can be compressed into statistical values ​​expressing fault activity, thereby simplifying the data structure and retaining key energy information. By normalizing and mapping to grayscale values ​​and filling them into image coordinate positions, a two-dimensional spatiotemporal grayscale image is generated, which can form a clear vibration evolution image structure, thereby providing a unified data input for image-based modeling. By performing histogram equalization processing on the grayscale image, image contrast can be enhanced and texture boundary clarity can be improved, thereby providing a higher quality input foundation for subsequent feature extraction and model recognition.

[0025] In one example, this application can be further configured as follows: performing texture analysis on the two-dimensional spatiotemporal grayscale image to extract anisotropic texture features specifically includes:

[0026] A gray-level co-occurrence matrix is ​​constructed based on the two-dimensional spatiotemporal grayscale image, and the construction direction of the gray-level co-occurrence matrix includes the temporal dimension and the spatial dimension.

[0027] Texture parameters are extracted in each direction, including the image's contrast, correlation, energy, and uniformity. The texture parameters extracted in multiple directions are combined into a feature vector to form the anisotropic texture feature.

[0028] By adopting the above technical solution, a gray-level co-occurrence matrix is ​​constructed based on a two-dimensional spatiotemporal gray-level image, and two directions, time and space, are set. This can comprehensively reflect the gray-level co-change characteristics of the image in multiple directions, thereby enhancing the directional sensitivity of texture representation. By extracting texture parameters such as contrast, correlation, energy and uniformity of the image in each direction and combining them into a feature vector, the structural nature and complexity of the image texture can be comprehensively characterized, thereby improving the classification model's ability to distinguish different fault modes.

[0029] In one example, this application can be further configured as follows: based on the anisotropic texture features, the analysis is performed using a pre-trained classification model to obtain the fault type, specifically including:

[0030] The anisotropic texture features are input into a first classification model for pattern recognition analysis, and a first fault type prediction result is output. The first classification model is at least one of a support vector machine classifier or a random forest classifier.

[0031] The two-dimensional spatiotemporal grayscale image is input into the second classification model to extract deep feature information representing the spatial propagation structure of the image, and based on the deep feature information, the second fault type prediction result is output. The second classification model is a pre-trained convolutional neural network model.

[0032] A fusion analysis operation is performed on the first fault type prediction result and the second fault type prediction result, and the final fault type is determined based on the fusion result.

[0033] By adopting the above technical solutions, and by inputting anisotropic texture features into the first classification model for recognition, it is possible to achieve efficient recognition of typical fault modes based on manual statistical features, thereby improving the robustness under known modes. By inputting two-dimensional spatiotemporal grayscale images into the second classification model and extracting deep feature information for recognition, it is possible to discover abstract patterns in complex spatial structures, thereby improving the model's sensitivity to unknown or minor fault modes. By fusing and analyzing the fault type prediction results of the two paths, it is possible to take into account the judgment results of explicit features and deep patterns, thereby improving the accuracy and stability of the final fault recognition and enhancing the system's ability to adapt to different working conditions.

[0034] In one example, this application can be further configured as follows: the fusion analysis operation performed on the first fault type prediction result and the second fault type prediction result, and the final fault type determined based on the fusion result, specifically includes:

[0035] Based on a preset weighted fusion strategy, the prediction results of the first fault type and the prediction results of the second fault type are weighted and calculated to obtain a weighted result.

[0036] The final fault type is determined based on the weighted result.

[0037] By adopting the above technical solution, and by using a preset weighted fusion strategy to perform weighted calculations on the output results of the first classification model and the second classification model, the predictive advantages of the two models can be combined and the importance of different models can be differentiated and adjusted, thereby improving the rationality and generalization ability of the fusion judgment. By determining the final fault type based on the weighted result, the stable output of the decision result can be achieved, thereby effectively reducing the risk of misjudgment by a single model and improving the reliability of fault identification in real-world environments.

[0038] In one example, this application can be further configured as follows: determining the fault start position according to the fault type using the corresponding texture start position recognition rule specifically includes:

[0039] Based on the fault type, the corresponding fault propagation direction is determined, and the texture analysis direction corresponding to the fault propagation direction is determined in the two-dimensional spatiotemporal grayscale image;

[0040] Traverse the image texture structure along the texture analysis direction and detect changes in texture parameters to identify the starting region where the change in texture parameters exceeds a preset threshold;

[0041] The fiber optic mileage location corresponding to the starting region is marked as the fault initiation location.

[0042] By adopting the above technical solution, the fault propagation direction is determined based on the fault type, and the corresponding texture analysis direction is matched. This allows the characteristic patterns of different types of faults to guide the starting position analysis direction, thereby improving the accuracy of fault location. By traversing the image texture structure along the analysis direction and detecting the variation amplitude of texture parameters, regions exceeding a preset threshold are identified as starting regions. This enables the automatic extraction of the fault location based on image abrupt change points, thereby improving the objectivity and real-time performance of starting point identification. By mapping the image position corresponding to the identified starting region to the fiber optic mileage position, a precise mapping between the image domain and the physical domain can be achieved, thereby realizing high-precision spatial positioning and providing clear location support for maintenance tasks.

[0043] The second objective of this invention is achieved through the following technical solution:

[0044] A fault identification system for a belt conveyor, the fault identification system for a belt conveyor comprising:

[0045] Vibration signal acquisition module, used to acquire multi-channel vibration signals of belt conveyors;

[0046] A grayscale image construction module is used to map the multi-channel vibration signal into a two-dimensional spatiotemporal grayscale image;

[0047] The texture analysis module is used to perform texture analysis operations on the two-dimensional spatiotemporal grayscale image and extract anisotropic texture features;

[0048] The fault identification module is used to analyze the anisotropic texture features using a pre-trained classification model to obtain the fault type. The pre-trained classification model includes a first classification model and a second classification model.

[0049] The starting position identification and level determination module is used to determine the fault starting position according to the fault type and the corresponding texture starting position identification rule, and to determine the fault level based on the anisotropic texture features to obtain the fault severity level.

[0050] The alarm output module is used to output a fault identification result composed of the fault type, the fault start location, and the fault severity level, which is used to drive alarm prompts.

[0051] By adopting the above technical solutions, by collecting multi-channel vibration signals of the belt conveyor, it is possible to achieve full-coverage real-time perception of the operating state of the conveyor belt, thereby improving the comprehensiveness of fault identification and the continuity of data acquisition; by mapping the multi-channel vibration signals into two-dimensional spatio-temporal grayscale images, the original time-series signals can be converted into information images with spatial distribution structures, thereby providing a unified representation basis for subsequent image analysis and feature extraction; by performing texture analysis operations on the two-dimensional spatio-temporal grayscale images and extracting anisotropic texture features, the local structures and directional change information in the images can be fully exploited, thereby improving the sensitivity and accuracy of fault type discrimination; by analyzing based on the anisotropic texture features using the first classification model and the second classification model to obtain the fault type, the advantages of artificial features and deep features can be integrated, thereby significantly improving the fault identification accuracy and adaptability under complex working conditions; by determining the fault starting position according to the fault type using the texture starting position recognition rule and judging the fault level based on the texture features, the dual recognition of the fault position and severity can be achieved, thereby providing a more targeted decision-making basis for equipment maintenance; by forming the fault identification results by combining the fault type, starting position and severity level for output, a structured output result that can be used for intelligent alarm and linkage response can be formed, thereby enhancing the intelligent early warning ability and operation and maintenance efficiency of the system.

[0052] The above object three of the present application is achieved by the following technical solutions:

[0053] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above method for fault identification of a belt conveyor are implemented.

[0054] The above object four of the present application is achieved by the following technical solutions:

[0055] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the steps of the above method for fault identification of a belt conveyor are implemented.

[0056] In summary, the present application includes the following beneficial technical effects:

[0057] 1. By acquiring multi-channel vibration signals from belt conveyors, real-time full-coverage perception of the conveyor belt's operating status can be achieved, thereby improving the comprehensiveness of fault identification and the continuity of data acquisition. By mapping multi-channel vibration signals into two-dimensional spatiotemporal grayscale images, the original time-series signals can be converted into information images with spatial distribution structures, thus providing a unified representation basis for subsequent image analysis and feature extraction. By performing texture analysis on the two-dimensional spatiotemporal grayscale images and extracting anisotropic texture features, the local structure and directional change information in the images can be fully explored, thereby improving the sensitivity and accuracy of fault type discrimination.

[0058] 2. By analyzing anisotropic texture features using a first and second classification model, fault types are obtained. This approach combines the advantages of artificial and deep features, significantly improving the accuracy and adaptability of fault identification under complex operating conditions. Furthermore, by determining the fault starting position based on the texture starting position recognition rule according to the fault type and judging the fault level based on texture features, dual identification of fault location and severity can be achieved, providing more targeted decision-making basis for equipment maintenance.

[0059] 3. By determining the corresponding fault propagation direction based on the fault type and matching it with the corresponding texture analysis direction, the feature patterns of different types of faults can be combined to guide the starting position analysis direction, thereby improving the accuracy of fault location; by traversing the image texture structure along the analysis direction and detecting the change amplitude of texture parameters, regions exceeding a preset threshold are identified as starting regions, enabling automatic extraction of the fault location based on image abrupt change points, thereby improving the objectivity and real-time performance of starting point identification; by mapping the image position corresponding to the identified starting region to the fiber optic mileage position, a precise mapping between the image domain and the physical domain can be achieved, thereby realizing high-precision spatial positioning and providing clear location support for maintenance tasks. Attached Figure Description

[0060] Figure 1 This is a flowchart of a fault identification method for a belt conveyor according to one embodiment of this application;

[0061] Figure 2 This is a schematic block diagram of a fault identification system for a belt conveyor according to one embodiment of this application;

[0062] Figure 3 This is a schematic diagram of a device according to one embodiment of this application. Detailed Implementation

[0063] The present application will be further described in detail below with reference to the accompanying drawings.

[0064] In one embodiment, such as Figure 1As shown, this application discloses a fault identification method for a belt conveyor, which specifically includes the following steps:

[0065] S10: Collects multi-channel vibration signals from the belt conveyor.

[0066] Specifically, a distributed fiber optic acoustic sensing system is used to lay single-mode armored sensing optical cables along the entire belt conveyor. The backscattered Rayleigh signals on each monitoring channel are collected in real time by the DAS host. During the acquisition process, the optical signals are converted into electrical signals and output to the data processing end in the format of "fiber mileage-timestamp-vibration amplitude". At the same time, a 50Hz notch filter is performed to suppress power frequency interference, and a moving average filter is used to reduce random noise, thereby extracting high-fidelity, multi-channel, and multi-scale vibration response data for subsequent modeling and analysis.

[0067] S20: Maps multi-channel vibration signals into a two-dimensional spatiotemporal grayscale image.

[0068] Specifically, time window T and spatial window L parameters are set for image conversion. The time window is used to cover at least one complete idler roller cycle, and the spatial window is used to cover multiple adjacent monitoring channels to ensure texture continuity. The vibration sequence of the corresponding channel is extracted within each window, and the vibration energy of each spatiotemporal sampling point is calculated based on the root mean square value. Then, the vibration energy is normalized and mapped to a gray value of [0, 255], and the gray value is filled into the corresponding image coordinate position to generate a two-dimensional spatiotemporal grayscale image with the horizontal axis as the fiber optic mileage and the vertical axis as the time axis. At the same time, histogram equalization processing is performed on the image to enhance the image contrast of weak fault areas, thereby improving the accuracy of subsequent texture extraction.

[0069] S30: Perform texture analysis on a two-dimensional spatiotemporal grayscale image to extract anisotropic texture features.

[0070] Specifically, in each grayscale image, grayscale co-occurrence matrices are constructed based on the temporal and spatial dimensions, respectively. When constructing the matrices, the pixel spacing is set to 1. Four key texture parameters that reflect the image structure pattern are extracted, including image contrast, correlation, energy, and uniformity. Each parameter is used to measure the local sharpness, temporal / spatial propagation consistency, local energy concentration, and overall texture smoothness of the image. Finally, the multiple parameters extracted from the two directions are combined into a feature vector to form an anisotropic texture feature that characterizes the vibration propagation pattern, providing numerical support for subsequent fault identification.

[0071] S40: Based on anisotropic texture features, a pre-trained classification model is used for analysis to obtain the fault type. The pre-trained classification model includes a first classification model and a second classification model.

[0072] Specifically, the texture features are input into a first classification model, such as a support vector machine or a random forest model, for pattern recognition, and are combined with typical fault texture samples labeled in the training data for classification. At the same time, the original two-dimensional spatiotemporal grayscale image is input into a second classification model, namely a convolutional neural network model, to automatically extract multi-layer spatial propagation structure features. Then, the classification results of the two models are fused and analyzed to determine the type of point source faults, instantaneous diffusion faults, and propagation faults. The fusion process uses a weighted voting mechanism to improve robustness in complex fault scenarios, thereby obtaining accurate fault type output.

[0073] Furthermore, the construction and training of this first classification model includes collecting data on five typical operating conditions: normal operation (2000 samples), point source faults (roller wear / bearing failure, 3000 samples), instantaneous diffusion faults (material impact, 1500 samples), propagation faults (belt misalignment / longitudinal tear / frame resonance, 3000 samples), and compound faults (roller failure + local misalignment, 500 samples). The training set, validation set, and test set are divided into three sets in a 7:2:1 ratio, and spatiotemporal grayscale images are generated and labeled (normal = 0, point source fault = 1, instantaneous diffusion = 2, propagation = 3, compound fault = 4). A 4D feature vector is input into a Support Vector Machine (SVM) using a Radial Basis Function (RBF) kernel with a regularization parameter C=10 and a kernel parameter γ=0.1. The model architecture of this convolutional neural network is as follows: It uses MobileNetv2, with an input of a 224×224×1 grayscale image. The network structure is as follows: Input layer: 224×224×1 grayscale image; Depthwise separable convolutional blocks (3 layers): convolutional kernel size 3×3, stride 1, ReLU6 activation function, each block followed by BatchNormal... m layers; pooling layers (2 layers): 2×2 max pooling, stride 2, compressing feature dimensions; global average pooling layer: converts multi-dimensional features into 128-dimensional vectors; feature output layer: outputs 128-dimensional texture feature vectors; training parameters: batch size = 32, training epochs = 30, optimizer = Adam (initial learning rate = 0.0001, decays by 10% every 10 epochs), loss function = cross-entropy loss, overfitting is prevented by Dropout (probability 0.5), and finally the convolutional neural network model is obtained.

[0074] S50: Based on the fault type, the corresponding texture start position recognition rule is used to determine the fault start position, and the fault severity level is determined based on the anisotropic texture features.

[0075] Specifically, based on the identified fault type, the image texture morphology corresponding to its propagation structure is determined. If it is a vertical stripe, the center coordinates of the highlighted area are taken as the location of the point source fault. If it is a V-shaped or X-shaped texture, the diffusion intersection point is taken as the starting position of the instantaneous diffusion fault. If it is a horizontal or diffuse texture, the starting edge coordinates of the energy distribution are taken as the starting point of the propagation fault. At the same time, the coverage length is calculated based on the span of the texture features on the image space axis, and the fault is divided into three levels: Level 1 (less than 10 meters), Level 2 (10 to 50 meters), or Level 3 (greater than 50 meters) according to the set length classification threshold. Finally, a judgment result with clear location and accurate classification is formed.

[0076] S60: Outputs the fault identification result, which is composed of fault type, fault start location and fault severity level, to drive alarm prompts.

[0077] Specifically, the fault type classification labels, starting position coordinate information and fault level values ​​obtained from the aforementioned identification are integrated, and a fault identification report is generated in a structured format. The results are transmitted to the monitoring platform at the output end and trigger the corresponding visual and audible alarm mechanisms. Level 1 faults are marked in blue to indicate routine maintenance, Level 2 faults are marked in yellow and the on-duty personnel are notified to intervene and investigate, and Level 3 faults trigger a red warning and simultaneously execute a shutdown command and SMS notification mechanism, so as to achieve the automated monitoring goal of early warning, fast response and graded handling.

[0078] In one embodiment, step S10, namely acquiring multi-channel vibration signals of the belt conveyor, specifically includes:

[0079] S11: Collect vibration information from the belt conveyor, perform signal conversion on the vibration information, and output multi-channel vibration data in a triplet format of fiber optic mileage, timestamp, and vibration amplitude.

[0080] Specifically, during the acquisition process, based on the fiber backscattering principle of the DAS device, the sensing optical cable laid along the belt conveyor converts the received vibration response into a reflected optical interference signal in the form of an acoustic signal. Through the photoelectric conversion mechanism, an electrical signal data stream with intensity modulation characteristics is generated, and the data on each channel is labeled in real time with its fiber optic mileage position and sampling timestamp. The fiber optic mileage is used to accurately calibrate the spatial position corresponding to the current data, the timestamp calibrates the dynamic response at the corresponding moment, and the vibration amplitude is used to quantify the vibration energy of that point in the current sampling period. Finally, a multi-channel vibration dataset oriented towards the space-time-intensity dimension is constructed with a triplet structure of "fiber optic mileage, timestamp, and vibration amplitude". This dataset can be continuously written to a buffer queue and used for subsequent image mapping processing. For example, in a certain actual working condition, a structured vibration data segment containing 1024 channels, a sampling frequency of 1kHz, and a duration of 10 seconds can be formed to support the accurate characterization of typical cascading fault chains.

[0081] S12: Perform filtering and noise reduction processing on the multi-channel vibration data. The filtering and noise reduction processing includes power frequency notch filtering and moving average filtering to obtain multi-channel vibration signals.

[0082] Specifically, the original multi-channel vibration data contains a large number of background interference signals unrelated to conveyor belt vibration, among which 50Hz power frequency noise is the most common. Therefore, the energy of the target frequency band is suppressed by setting the cutoff frequency and bandwidth parameters of the notch filter to eliminate the periodic pulse signal caused by power frequency electromagnetic interference. At the same time, in order to avoid the amplified effect of single-point random interference or instantaneous abnormal sampling values ​​on subsequent analysis, a moving average filtering algorithm is used to locally smooth the vibration amplitude of each channel in the time dimension. Specifically, a weighted average is performed with 3 to 5 sampling points as a sliding window. In practical engineering applications, the window length and filtering parameters can be dynamically adjusted according to the sensor layout density and sampling frequency. This combined filtering process can significantly improve the stability and continuity of the vibration signal, providing a clear and drift-free grayscale base input for image mapping.

[0083] In one embodiment, step S20, which maps the multi-channel vibration signal into a two-dimensional spatiotemporal grayscale image, specifically includes:

[0084] S21: Set the time window and spatial window parameters for image construction, and perform window segmentation processing on the multi-channel vibration signal based on the time window and spatial window parameters.

[0085] Specifically, the time window length T and spatial window length L are set as segmentation parameters for constructing a two-dimensional spatiotemporal image. The time window length is usually set according to one cycle or half cycle of the belt conveyor operation to capture stable dynamic response characteristics. The spatial window length is divided into several continuous channel combinations according to the number of fiber optic channels to ensure the texture continuity in the spatial dimension. For example, T=1s and L=32 channels can be set as a basic window unit to segment and extract multi-channel vibration signals in segments for each time period and channel, generating multiple local data segments with consistent size and distribution characteristics to support the image mapping process. The data in each window segment will be regarded as an element input in an image block.

[0086] S22: Under each combination of spatial channel and time window, extract the vibration sampling sequence in the corresponding channel and calculate the vibration energy value of the vibration sampling sequence.

[0087] Specifically, in each window combination, the complete vibration amplitude sequence of the corresponding channel is obtained, and the root mean square calculation is performed to characterize the average energy level of the channel within the current time window. For the multi-channel data within each spatiotemporal window, the spatiotemporal points are calculated. Vibrational energy: quantified using the root mean square (RMS) value, the formula is as follows: ,in The number of sampling points for a single spatiotemporal point. The sampling amplitude; grayscale mapping: the RMS value is normalized to the [0, 255] grayscale range through min-max normalization, i.e. Generate spatiotemporal grayscale images —Horizontal axis For fiber optic mileage, the vertical axis is... For time, pixel brightness is positively correlated with vibration energy. For example, if a section of idler roller is damaged during the operation of a conveyor belt, it will cause high-amplitude short-cycle vibration. The RMS energy value of this area will be significantly higher than that of other locations, thus providing a strong feature response for subsequent image mapping.

[0088] S23: The vibration energy value is normalized and mapped to a gray value, and the gray value is filled into the corresponding image coordinate position to generate a two-dimensional spatiotemporal gray image.

[0089] Specifically, the vibration energy values ​​under all window combinations are uniformly subjected to maximum and minimum normalization processing, which linearly maps them to the integer grayscale range of [0, 255]. The normalized grayscale values ​​are then filled into a two-dimensional matrix according to row and column coordinates, where the row coordinates correspond to the time window index and the column coordinates correspond to the fiber optic channel number. Finally, a grayscale image matrix is ​​obtained, which intuitively reflects the distribution of vibration energy in the spatial and temporal dimensions. High-energy areas are represented as bright areas on the image, and low-energy areas are represented as dark areas. The generated grayscale image is uniformly scaled to 224×224 pixels. For example, if the fault's influence range expands to multiple channels over time, the image will show horizontal or diagonal bright stripes, exhibiting typical propagation characteristics.

[0090] S24: Perform image enhancement processing on a two-dimensional grayscale image, including histogram equalization.

[0091] Specifically, to improve the contrast and texture recognition of detailed areas in the image, a global histogram equalization process is performed on the initially generated two-dimensional grayscale image. The specific steps include statistically analyzing the distribution histogram of all grayscale values ​​in the original image, calculating its cumulative distribution function, and transforming the grayscale value of each pixel in the original image according to the mapping rules. This makes the grayscale distribution in the original image more balanced, stretches the contrast to a larger range, and enhances the clarity of the boundary between low-energy and high-energy areas. This process is particularly crucial for the identification of weak signal areas. For example, in some conveyor sections, there may be slight belt loosening or initial roller abnormalities. These conditions have low energy levels but can be clearly presented in the image after enhancement, thereby improving the ability to identify faults early.

[0092] In one embodiment, step S30, which involves performing texture analysis on the two-dimensional spatiotemporal grayscale image to extract anisotropic texture features, specifically includes:

[0093] S31: Construct a gray-level co-occurrence matrix based on a two-dimensional spatiotemporal gray-level image. The construction direction of the gray-level co-occurrence matrix includes the temporal dimension and the spatial dimension.

[0094] Specifically, after the grayscale image is constructed, the image texture structure is extracted based on the grayscale co-occurrence relationship between pixels. The co-occurrence matrix generation operation is performed on the image according to the set directional dimension. In the time dimension, the grayscale co-occurrence frequency of adjacent pixel pairs along the vertical axis is constructed, reflecting the local persistence characteristics of vibration energy in the time dimension. In the spatial dimension, the grayscale co-occurrence frequency of adjacent pixel pairs along the horizontal axis is constructed, reflecting the consistency and diffusion behavior of vibration response in the conveyor belt channel. To improve stability, the pixel spacing is set to 1 and the diagonal neighborhood is ignored during the construction process. The combination of all grayscale value pairs will form a two-dimensional grayscale co-occurrence matrix, where the matrix element (i,j) represents the number or frequency of co-occurrence of pixels with grayscale values ​​i and j in the target direction. For example, in the region where the fall of the conveyor roller causes a concentrated energy response in a short time, its co-occurrence matrix will show a high-frequency diagonal offset structure in the spatial dimension, which helps to reflect its heterogeneous extension characteristics in subsequent extraction.

[0095] S32: Extract texture parameters in each direction. Texture parameters include image contrast, correlation, energy and uniformity. Combine the texture parameters extracted in multiple directions into a feature vector to form anisotropic texture features.

[0096] Specifically, based on the constructed temporal and spatial gray-level co-occurrence matrices, multiple statistical indicators are calculated for each matrix. Contrast measures the clarity and intensity of local changes in image texture; correlation reflects the degree of gray-level correlation between pixels in the image; energy represents the sum of squares of the elements in the co-occurrence matrix and is used to measure the overall regularity and repeatability of the texture; uniformity measures the consistency of the distribution of element values ​​in the matrix and is used to determine the smoothness and complexity of the image region. After extracting the above four parameters separately in each direction, the extracted values ​​in the temporal and spatial dimensions are concatenated to form an eight-dimensional vector, which serves as the anisotropic texture feature description vector for the image patch. In practical applications, if the high-frequency energy region in the image expands along the spatial direction but remains stable along the temporal direction, its energy and contrast are significantly higher in the spatial direction than in the temporal direction, thus reflecting that it is a segment-by-segment expansion type fault structure, providing a direction-sensitive input basis for subsequent model recognition.

[0097] Furthermore, four key texture parameters are calculated, as follows:

[0098] Contrast: This reflects texture clarity; faulty textures have higher contrast than noise.

[0099] Correlation: This reflects the spatiotemporal continuity of texture, and the spatial correlation of propagating faults is higher;

[0100] energy: This reflects the uniformity of the texture, and the energy of point source faults is concentrated, resulting in higher energy values;

[0101] Uniformity: This reflects the uniformity of texture distribution, and the uniformity of propagating faults is higher;

[0102] Feature vector construction: Combine the four parameters in the order of "contrast-correlation-energy-uniformity" to form a 4-dimensional feature vector.

[0103] In one embodiment, in step S40, based on the anisotropic texture features, a pre-trained classification model is used for analysis to obtain the fault type, specifically including:

[0104] S41: Input the anisotropic texture features into the first classification model for pattern recognition analysis, and output the first fault type prediction result. The first classification model is at least one of support vector machine classifier or random forest classifier.

[0105] Specifically, anisotropic texture features extracted from two-dimensional spatiotemporal grayscale images are used as feature vectors and input into the first classification model for predictive analysis. This model is trained offline, and a training set is constructed by combining various typical fault scenarios of belt conveyors. During the training phase, the optimal support vector machine or random forest algorithm is selected as the main model through cross-validation, and the input format that matches the texture feature dimension with the model parameters is set to ensure classification performance. During the runtime phase, the input anisotropic texture features are used to perform feature matching and classification judgment with the decision boundary or tree structure stored in the model, and the corresponding first fault type label is output. For example, for an image whose spatial texture is characterized by horizontal stretching and high energy concentration, its texture features have been labeled as "belt misalignment fault" in the training set. The first classification model can quickly complete the matching based on the feature value range of its energy and contrast in the spatial dimension, thereby improving the ability to quickly identify known patterns and the robustness of the model.

[0106] S42: Input the two-dimensional spatiotemporal grayscale image into the second classification model, extract deep feature information representing the spatial propagation structure of the image, and output the second fault type prediction result based on the deep feature information. The second classification model is a pre-trained convolutional neural network model.

[0107] Specifically, the original two-dimensional spatiotemporal grayscale image without manual extraction of texture parameters is directly input into a pre-trained convolutional neural network model. This model contains multiple convolutional layers, pooling layers, and fully connected layers to automatically extract deep spatial features reflecting the fault propagation structure in the image. The first few convolutional kernels focus on extracting local high-frequency energy change regions, the middle convolutional structure is used to capture the directionality and morphological boundary information of fault diffusion, and the subsequent fully connected layers map the extracted features into low-dimensional embedding vectors for classification. During the training phase, image data augmentation techniques are used to improve the model's adaptability to different texture deformations and lighting conditions. During the inference phase, the Softmax output layer generates probability prediction results for various fault labels, and finally outputs the second fault type label. For example, when there is an obvious V-shaped or X-shaped energy diffusion structure in the image, the network model can focus on the intersection area through convolutional features and determine it as a "transient diffusion type fault", thereby improving the model's ability to recognize complex texture structures.

[0108] S43: Perform a fusion analysis on the prediction results of the first fault type and the prediction results of the second fault type, and determine the final fault type based on the fusion results.

[0109] Specifically, a weighted fusion analysis is performed based on the fault type prediction results output by the two classification models. First, initial weight parameters are configured for the first and second models respectively, and the ratio is set according to their accuracy and stability in the training set test. During the fusion process, the prediction results of the two models are weighted or probabilistically voted according to the set weights. If the output results of the two models are consistent, the fault type is directly confirmed. If the results are inconsistent, the judgment of the model with higher weight is given priority or a confidence threshold strategy is combined for decision-making. For example, if the first model predicts "point anomaly" and the second model predicts "diffusion fault", but the confidence of the second model is significantly higher than the set threshold, the final fault type is determined to be "diffusion fault". This fusion strategy can take into account both the stability based on statistical features and the perception ability of deep models in complex scenarios, thereby improving the final recognition accuracy and the overall robustness of the system.

[0110] In one embodiment, step S43, namely, performing a fusion analysis operation on the first fault type prediction result and the second fault type prediction result, and determining the final fault type based on the fusion result, specifically includes:

[0111] S431: Based on a preset weighted fusion strategy, the prediction results of the first fault type and the prediction results of the second fault type are weighted and calculated to obtain a weighted result.

[0112] Specifically, the fusion weight parameters corresponding to the first classification model and the second classification model are set as α and β, respectively, where α+β=1. During the offline evaluation phase of the model, the weight allocation is set according to its accuracy, confidence discrimination and anti-perturbation performance on the training set and validation set. For example, if the first model performs well under simple working conditions while the second model is more sensitive to complex textures, a weighting ratio of α=0.4 and β=0.6 can be set. During the running phase, when the first model outputs a probability of p1 for fault type A and the second model outputs a probability of p2 for fault type B, α×p1 and β×p2 are used as fusion score reference values, respectively. A weighted merging operation is performed on all prediction types to generate a set of fusion score vectors. Each dimension in this vector represents the fusion score of the current input corresponding to a certain fault type. Finally, it constitutes the intermediate result of the fusion analysis. For example, when both models output a high probability prediction value for the "belt edge anomaly" fault, this fault category obtains the maximum score in the fusion vector, reflecting the consistency and confidence superposition of the results of the two paths.

[0113] S432: Determine the final fault type based on the weighted result.

[0114] Specifically, the scores corresponding to each fault type in the fusion scoring vector are sorted, and the fault type with the highest score is selected as the final identification result. If the highest score does not reach the set fusion confidence threshold, a secondary decision-making mechanism can be triggered for anomaly analysis or manual intervention. In practical applications, the confidence threshold can be set to 0.7. When the highest item in the weighted score exceeds this threshold, the type is directly determined as the target fault type. Otherwise, the candidate set weighted re-identification process is entered to improve decision robustness. This strategy can effectively avoid misjudgment caused by model divergence or image interference. For example, when the first model identifies the fault as "diffusion-type fault" and the second model identifies it as "roller anomaly", and the fusion-type diffusion score is 0.82, which is significantly higher than the other types and exceeds the threshold, the final output fault type is determined to be "diffusion-type fault". This decision-making method takes into account both accuracy and fault tolerance, ensuring the reliability and credibility of the final output result.

[0115] In one embodiment, step S50, namely, determining the fault start position according to the fault type and using the corresponding texture start position recognition rule, specifically includes:

[0116] S51: Determine the corresponding fault propagation direction based on the fault type, and determine the texture analysis direction corresponding to the fault propagation direction in the two-dimensional spatiotemporal grayscale image.

[0117] Specifically, the corresponding fault propagation direction is determined based on the fault type. Priority is given to identifying the texture distribution features common to this type in spatiotemporal grayscale images. For example, propagation faults are usually accompanied by continuous horizontal or sloping strip textures, while point source faults are manifested as vertical or isolated texture blocks. Based on this, the analysis direction consistent with the texture distribution direction is selected as the texture analysis direction. For example, when the fault type is determined to be propagating along the belt running direction, the time dimension is selected as the main axis direction, and when the fault is determined to be a structural transmission type, the spatial dimension direction is selected. This ensures that subsequent texture traversal operations can capture feature changes along the actual propagation path, thereby improving the accuracy of the starting position identification.

[0118] S52: Traverse the image texture structure along the texture analysis direction and detect changes in texture parameters, identifying the starting region where the change in texture parameters exceeds a preset threshold.

[0119] Specifically, image traversal operations are performed pixel-by-pixel or in a sliding window manner along a determined texture analysis direction. The changing trends of texture parameters such as contrast, energy, and uniformity in continuous image regions are monitored in real time. When the change in any texture parameter exceeds a preset threshold, such as a sudden increase in local contrast or a sudden drop in uniformity, the region can be identified as a potential fault initiation region. To reduce the risk of misjudgment, differential comparison can also be performed by combining the texture evolution trends of the preceding and following frames of the image to confirm that the change is caused by a fault rather than background disturbance.

[0120] S53: Mark the fiber optic mileage location corresponding to the starting area as the fault start location.

[0121] Specifically, after obtaining the image coordinates of the starting region in a two-dimensional spatiotemporal grayscale image, the pixel index of the region in the fiber direction is extracted according to the construction method of the image with the horizontal axis as fiber optic mileage and the vertical axis as time. The fiber optic mileage position of the region in the actual conveyor structure is then calculated by combining the mapping relationship. For example, based on the channel spacing of 5 meters per channel and the setting of a 100-meter space window, the 10th column pixel in the image corresponds to the 50-meter position. Finally, the fiber optic mileage position is output as the starting position of the fault for subsequent maintenance personnel to locate the specific area on site.

[0122] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0123] In one embodiment, a fault identification system for a belt conveyor is provided, which corresponds one-to-one with the fault identification method for a belt conveyor described in the above embodiments. For example... Figure 2As shown, the fault identification system for a belt conveyor includes a vibration signal acquisition module, a grayscale image construction module, a texture analysis module, a fault identification module, a starting position identification and level determination module, and an alarm output module. Detailed descriptions of each functional module are as follows:

[0124] Vibration signal acquisition module, used to acquire multi-channel vibration signals of belt conveyors;

[0125] A grayscale image construction module is used to map the multi-channel vibration signal into a two-dimensional spatiotemporal grayscale image;

[0126] The texture analysis module is used to perform texture analysis operations on the two-dimensional spatiotemporal grayscale image and extract anisotropic texture features;

[0127] The fault identification module is used to analyze the anisotropic texture features using a pre-trained classification model to obtain the fault type. The pre-trained classification model includes a first classification model and a second classification model.

[0128] The starting position identification and level determination module is used to determine the fault starting position according to the fault type and the corresponding texture starting position identification rule, and to determine the fault level based on the anisotropic texture features to obtain the fault severity level.

[0129] The alarm output module is used to output a fault identification result composed of the fault type, the fault start location, and the fault severity level, which is used to drive alarm prompts.

[0130] Optionally, the vibration signal acquisition module includes:

[0131] The vibration information acquisition submodule is used to acquire the vibration information of the belt conveyor, perform signal conversion operation on the vibration information, and output multi-channel vibration data in a triplet format of fiber optic mileage, timestamp and vibration amplitude.

[0132] The signal noise reduction submodule is used to perform filtering and noise reduction processing on the multi-channel vibration data. The filtering and noise reduction processing includes power frequency notch filtering and moving average filtering to obtain the multi-channel vibration signal.

[0133] Optionally, the grayscale image building module includes:

[0134] The window parameter setting submodule is used to set the time window and spatial window parameters for image construction, and to perform window segmentation processing on the multi-channel vibration signal based on the time window and spatial window parameters;

[0135] The energy calculation submodule is used to extract the vibration sampling sequence in the corresponding channel under each combination of spatial channel and time window, and calculate the vibration energy value of the vibration sampling sequence.

[0136] The image mapping submodule is used to normalize and map the vibration energy value into a gray value, and fill the gray value into the corresponding image coordinate position to generate a two-dimensional spatiotemporal grayscale image;

[0137] The image enhancement submodule is used to perform image enhancement processing on the two-dimensional grayscale image, the image enhancement processing including histogram equalization.

[0138] Optionally, the texture analysis module includes:

[0139] The co-occurrence matrix construction submodule is used to construct a gray-level co-occurrence matrix based on the two-dimensional spatiotemporal gray-level image. The construction direction of the gray-level co-occurrence matrix includes the temporal dimension and the spatial dimension.

[0140] The texture parameter extraction submodule is used to extract texture parameters in each direction. The texture parameters include the contrast, correlation, energy and uniformity of the image. The texture parameters extracted in multiple directions are combined into a feature vector to form the anisotropic texture feature.

[0141] Optionally, the fault identification module includes:

[0142] The traditional model classification submodule is used to input the anisotropic texture features into the first classification model for pattern recognition analysis and output the first fault type prediction result. The first classification model is at least one of support vector machine classifier or random forest classifier.

[0143] The deep feature recognition submodule is used to input the two-dimensional spatiotemporal grayscale image into the second classification model, extract deep feature information representing the spatial propagation structure of the image, and output the second fault type prediction result based on the deep feature information. The second classification model is a pre-trained convolutional neural network model.

[0144] The fusion analysis submodule is used to perform fusion analysis on the prediction results of the first fault type and the prediction results of the second fault type, and determine the final fault type based on the fusion results.

[0145] Optionally, the fusion analysis submodule includes:

[0146] The weighted calculation unit is used to perform weighted calculation on the prediction results of the first fault type and the prediction results of the second fault type based on a preset weighted fusion strategy to obtain a weighted result;

[0147] The fault type output unit is used to determine the final fault type based on the weighted result.

[0148] Optionally, the starting position recognition and level determination module includes:

[0149] The propagation direction determination submodule is used to determine the corresponding fault propagation direction based on the fault type, and to determine the texture analysis direction corresponding to the fault propagation direction in the two-dimensional spatiotemporal grayscale image;

[0150] The texture mutation recognition submodule is used to traverse the image texture structure along the texture analysis direction and detect changes in texture parameters, and identify the starting region where the change in texture parameters exceeds a preset threshold.

[0151] The location mapping submodule is used to mark the fiber optic mileage location corresponding to the starting area as the fault start location.

[0152] Specific limitations regarding the fault identification system for a belt conveyor can be found in the above description of the fault identification method for a belt conveyor, and will not be repeated here. Each module in the aforementioned fault identification system for a belt conveyor can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0153] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a fault identification method for a belt conveyor.

[0154] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps:

[0155] Collect multi-channel vibration signals from a belt conveyor;

[0156] Multi-channel vibration signals are mapped into two-dimensional spatiotemporal grayscale images;

[0157] Texture analysis is performed on a two-dimensional spatiotemporal grayscale image to extract anisotropic texture features;

[0158] Based on anisotropic texture features, a pre-trained classification model is used for analysis to obtain the fault type. The pre-trained classification model includes a first classification model and a second classification model.

[0159] Based on the fault type, the corresponding texture start position recognition rule is used to determine the fault start position, and the fault severity level is determined based on the anisotropic texture features.

[0160] The fault identification result, composed of fault type, fault starting location, and fault severity level, is output to drive alarm prompts.

[0161] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0162] Collect multi-channel vibration signals from a belt conveyor;

[0163] Multi-channel vibration signals are mapped into two-dimensional spatiotemporal grayscale images;

[0164] Texture analysis is performed on a two-dimensional spatiotemporal grayscale image to extract anisotropic texture features;

[0165] Based on anisotropic texture features, a pre-trained classification model is used for analysis to obtain the fault type. The pre-trained classification model includes a first classification model and a second classification model.

[0166] Based on the fault type, the corresponding texture start position recognition rule is used to determine the fault start position, and the fault severity level is determined based on the anisotropic texture features.

[0167] The fault identification result, composed of fault type, fault starting location, and fault severity level, is output to drive alarm prompts.

[0168] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0169] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0170] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A fault identification method for a belt conveyor, characterized in that, The fault identification method for a belt conveyor includes: Collect multi-channel vibration signals from a belt conveyor; The multi-channel vibration signal is mapped into a two-dimensional spatiotemporal grayscale image; Perform texture analysis on the two-dimensional spatiotemporal grayscale image to extract anisotropic texture features; Based on the anisotropic texture features, a pre-trained classification model is used for analysis to obtain the fault type. The pre-trained classification model includes a first classification model and a second classification model. Based on the fault type, the corresponding texture start position recognition rule is used to determine the fault start position, and the fault severity level is determined based on the anisotropic texture features. The fault identification result, composed of the fault type, the fault start location, and the fault severity level, is output to drive alarm prompts.

2. The fault identification method for a belt conveyor according to claim 1, characterized in that, The acquisition of multi-channel vibration signals from the belt conveyor specifically includes: The vibration information of the belt conveyor is collected, the vibration information is converted into a signal, and multi-channel vibration data is output in a triplet format of fiber optic mileage, timestamp and vibration amplitude. The multi-channel vibration data is subjected to filtering and noise reduction processing, which includes power frequency notch filtering and moving average filtering, to obtain the multi-channel vibration signal.

3. The fault identification method for a belt conveyor according to claim 1, characterized in that, The process of mapping the multi-channel vibration signal into a two-dimensional spatiotemporal grayscale image specifically includes: Set time window and spatial window parameters for image construction, and perform window segmentation processing on the multi-channel vibration signal based on the time window and spatial window parameters; In each combination of spatial channel and time window, the vibration sampling sequence in the corresponding channel is extracted, and the vibration energy value of the vibration sampling sequence is calculated. The vibration energy value is normalized and mapped to a gray value, and the gray value is filled into the corresponding image coordinate position to generate a two-dimensional spatiotemporal grayscale image; The two-dimensional grayscale image is subjected to image enhancement processing, which includes histogram equalization.

4. The fault identification method for a belt conveyor according to claim 1, characterized in that, The step of performing texture analysis on the two-dimensional spatiotemporal grayscale image to extract anisotropic texture features specifically includes: A gray-level co-occurrence matrix is ​​constructed based on the two-dimensional spatiotemporal grayscale image, and the construction direction of the gray-level co-occurrence matrix includes the temporal dimension and the spatial dimension. Texture parameters are extracted in each direction, including the image's contrast, correlation, energy, and uniformity. The texture parameters extracted in multiple directions are combined into a feature vector to form the anisotropic texture feature.

5. The fault identification method for a belt conveyor according to claim 1, characterized in that, The fault types are obtained by analyzing the anisotropic texture features using a pre-trained classification model, specifically including: The anisotropic texture features are input into a first classification model for pattern recognition analysis, and a first fault type prediction result is output. The first classification model is at least one of a support vector machine classifier or a random forest classifier. The two-dimensional spatiotemporal grayscale image is input into the second classification model to extract deep feature information representing the spatial propagation structure of the image, and based on the deep feature information, the second fault type prediction result is output. The second classification model is a pre-trained convolutional neural network model. A fusion analysis operation is performed on the first fault type prediction result and the second fault type prediction result, and the final fault type is determined based on the fusion result.

6. The fault identification method for a belt conveyor according to claim 5, characterized in that, The step of performing a fusion analysis on the prediction results of the first fault type and the prediction results of the second fault type, and determining the final fault type based on the fusion result, specifically includes: Based on a preset weighted fusion strategy, the prediction results of the first fault type and the prediction results of the second fault type are weighted and calculated to obtain a weighted result. The final fault type is determined based on the weighted result.

7. The fault identification method for a belt conveyor according to claim 1, characterized in that, The step of determining the fault start position based on the fault type and the corresponding texture start position recognition rule specifically includes: Based on the fault type, the corresponding fault propagation direction is determined, and the texture analysis direction corresponding to the fault propagation direction is determined in the two-dimensional spatiotemporal grayscale image; Traverse the image texture structure along the texture analysis direction and detect changes in texture parameters to identify the starting region where the change in texture parameters exceeds a preset threshold; The fiber optic mileage location corresponding to the starting region is marked as the fault initiation location.

8. A fault identification system for a belt conveyor, characterized in that, The fault identification system for the belt conveyor includes: Vibration signal acquisition module, used to acquire multi-channel vibration signals of belt conveyors; A grayscale image construction module is used to map the multi-channel vibration signal into a two-dimensional spatiotemporal grayscale image; The texture analysis module is used to perform texture analysis operations on the two-dimensional spatiotemporal grayscale image and extract anisotropic texture features; The fault identification module is used to analyze the anisotropic texture features using a pre-trained classification model to obtain the fault type. The pre-trained classification model includes a first classification model and a second classification model. The starting position identification and level determination module is used to determine the fault starting position according to the fault type and the corresponding texture starting position identification rule, and to determine the fault level based on the anisotropic texture features to obtain the fault severity level. The alarm output module is used to output a fault identification result composed of the fault type, the fault start location, and the fault severity level, which is used to drive alarm prompts.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the fault identification method for a belt conveyor as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the fault identification method for a belt conveyor as described in any one of claims 1 to 7.

Images