Belt conveyor carrier roller fault monitoring system based on DAS technology

Abstract

The invention discloses a belt conveyor carrier roller fault monitoring system based on a DAS technology, and relates to the technical field of industrial equipment state monitoring. According to the system, vibration signals of a plurality of monitoring points are obtained through a distributed optical fiber sensing technology; preprocessing the vibration to obtain a target signal; performing two-dimensional representation on the target signal to obtain a target image; and taking the target image as the input of a pre-trained fault detection model to obtain a fault detection result. According to the system, high-density vibration monitoring of the whole-line carrier rollers of the belt conveyor is achieved through optical fibers, the coverage range is wide, and precision is high. Through signal preprocessing and image recognition, carrier roller faults are accurately recognized, and the misjudgment and missed judgment rate is remarkably reduced. The system is high in automation degree, fast in real-time alarm response, suitable for complex environments, simple and convenient in wiring and low in maintenance cost, the problems of insufficient coverage and low efficiency of traditional detection are effectively solved, and the intelligent level of operation and maintenance is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment condition monitoring, and particularly relates to a belt conveyor idler fault monitoring system based on DAS technology. Background Art

[0002] In modern bulk material transportation industries such as coal, electricity, ports, and mines, belt conveyors, as core material conveying equipment, play a crucial role. Idlers are important components of belt conveyors. They are numerous, widely distributed, and work in complex and harsh environments. They are prone to failures such as wear, eccentricity, jamming, and bearing failure under long-term high-speed rotation. Once an idler is damaged, it will not only affect the normal operation of the equipment but also may cause safety accidents such as belt tearing, material spillage, and even fires, resulting in serious economic losses and safety hazards.

[0003] Traditional idler fault detection mainly relies on manual inspections or point sensor layouts, which have problems such as low detection efficiency, long cycle, and high labor intensity. Moreover, due to the long distribution distance, large quantity, and complex operating environment of idlers, traditional fault detection has problems such as limited coverage, many misjudgments and missed judgments, and low efficiency, making it difficult to meet the high-reliability monitoring requirements of "early detection, early warning, and early treatment" in the current industrial site. At the same time, the complex wiring and difficult maintenance of traditional electrical sensors also limit their popularization and application in large-scale conveying lines.

[0004] In the context of the rapid development of industrial intelligence and informatization, distributed optical fiber sensing (DAS) technology, as an emerging and advanced technology that can achieve long-distance and multi-point real-time vibration monitoring, has gradually attracted attention. However, existing distributed optical fiber monitoring technologies are mostly used for pipeline leakage or perimeter security, and there is no method that can be directly used for conveyor fault detection. Summary of the Invention

[0005] The purpose of the present invention is to solve the problems of limited coverage, many misjudgments and missed judgments, and low efficiency mentioned in the above background art, and to propose a belt conveyor idler fault monitoring system based on DAS technology.

[0006] The present invention provides a belt conveyor idler fault monitoring system based on DAS technology. The system includes: A data acquisition module for acquiring vibration signals of multiple monitoring points through distributed optical fiber sensing technology; A preprocessing module for preprocessing a first target signal to obtain a second target signal; the first target signal is the vibration signal of any one monitoring point; A signal image generation module for two-dimensionally characterizing the second target signal to obtain a target image; A fault detection module, configured to use the target image as the input of a pre-trained fault detection model to obtain a fault detection result; A fault warning module, configured to, if the fault detection result indicates a fault, determine the fault location according to the monitoring point corresponding to the first target signal and issue a warning message.

[0007] Optionally, the preprocessing module includes: A modal decomposition module, configured to decompose the first target signal into multiple modal components by using a variational modal decomposition algorithm; A component discrimination module, configured to determine whether the first target modal component is a valid component, a noise-containing component, or a noise component according to the kurtosis of the first target modal component and its distance from the first target signal; the first target modal component is any one of the multiple modal components; A denoising module, configured to, if the first target modal component is a noise-containing component, perform wavelet threshold denoising on the first target modal component to obtain a second target modal component; A signal reconstruction module, configured to perform signal reconstruction according to the valid component and the denoised second target modal component to obtain a second target signal.

[0008] Optionally, the modal decomposition module includes: A parameter optimization module, configured to use a particle swarm optimization algorithm and a fitness function with the maximum kurtosis as the target to optimize the parameters of the variational modal decomposition to obtain the decomposition number K and the penalty factor α; A variational decomposition module, configured to perform variational modal decomposition on the first target signal according to the decomposition number K and the penalty factor α to obtain multiple modal components.

[0009] Optionally, the component discrimination module includes: A kurtosis calculation module, configured to calculate the kurtosis KU of the first target modal component; A distance calculation module, configured to calculate the Euclidean distance ED between the first target modal component and the first target signal; An index calculation module, configured to calculate the effective index value EV of the first target modal component according to the kurtosis and the Euclidean distance: ; A score calculation module, configured to calculate the Z-score of the effective index value of each modal component; A noise determination module, configured to determine the modal components with Z-scores less than a preset threshold as noise components; A clustering division module, configured to perform binary classification clustering on the remaining modal components, classify the category with a higher effective index value as valid components, and classify the other category as noise-containing components.

[0010] Optionally, the denoising module includes: A wavelet decomposition module for performing multi-level wavelet decomposition on the first target modal component to obtain a set of wavelet coefficients, including low-frequency approximation coefficients and multiple high-frequency detail coefficients; A threshold processing module for performing threshold processing on multiple high-frequency detail coefficients using a preset threshold function to obtain multiple target detail coefficients; the threshold function is: ; Wherein, is the target detail coefficient after threshold processing; is the high-frequency detail coefficient at the j-th scale, is its corresponding wavelet threshold; sgn() is the sign function; exp is the exponential function with the natural constant e as the base; A component reconstruction module for performing inverse wavelet transform according to the low-frequency approximation coefficients and multiple target detail coefficients to obtain a denoised second target modal component.

[0011] Optionally, the signal image generation module includes: A two-dimensional characterization module for converting the second target signal into a two-dimensional coefficient matrix using continuous wavelet transform; A color map mapping module for performing pseudo-color mapping on the two-dimensional coefficient matrix using the Jet color map to generate a target image.

[0012] Optionally, the fault detection model includes a feature extraction network and a classifier; the feature extraction network includes a first convolutional block, a second convolutional block, a third convolutional block, a fourth convolutional block, and a fifth convolutional block connected in sequence; the classifier includes a flattening layer, a first feature compressor, a second feature compressor, and an output layer connected in sequence; wherein, the flattening layer serves as an interface between the feature extraction network and the classifier, converting the multi-dimensional features output by the fifth convolutional block into a one-dimensional vector as the input of the first feature compressor; specifically: Each convolutional block includes multiple convolutional layers and a pooling layer; the size of the convolutional kernel of the convolutional layer is 3×3, the stride is 1, and the activation function is ReLU; the size of the pooling layer kernel is 2×2, and the stride is 2; the number of convolutional layers in the first convolutional block, the second convolutional block, the third convolutional block, the fourth convolutional block, and the fifth convolutional block are 2, 2, 3, 3, and 3 respectively; Each feature compressor consists of a fully connected layer, ReLU activation, and Dropout regularization; during the training process, each feature compressor will randomly lose 50% of the neurons to prevent overfitting; The output layer consists of a fully connected layer and Sigmoid activation.

[0013] Advantages of the present invention: The present invention proposes a roller fault monitoring system for belt conveyors based on DAS technology. The system acquires vibration signals at multiple monitoring points through distributed optical fiber sensing technology, preprocesses the first target signal to obtain the second target signal, where the first target signal is the vibration signal of any one monitoring point, performs two-dimensional characterization on the second target signal to obtain the target image, uses the target image as the input of a pre-trained fault detection model to obtain the fault detection result, and if the fault detection result indicates a fault, determines the fault location according to the monitoring point corresponding to the first target signal and issues an alarm message.

[0014] Based on DAS technology, the system uses optical fibers to achieve high-density vibration monitoring of all rollers of the belt conveyor, with a wide coverage range and high accuracy. Through signal preprocessing and image recognition, it can accurately identify roller faults and significantly reduce the misjudgment and missed judgment rates. The system has a high degree of automation, a fast real-time alarm response, can adapt to complex environments, has simple wiring and low maintenance costs, effectively solves the problems of insufficient traditional detection coverage and low efficiency, and improves the intelligent level of operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 FIG. 10 is an architecture diagram of a roller fault monitoring system for belt conveyors based on DAS technology provided by an embodiment of the present invention; Figure 2 FIG. 13 is a schematic diagram of optical cable installation provided by an embodiment of the present invention; Figure 3 FIG. 16 is a network architecture diagram of a fault detection model provided by an embodiment of the present invention; Among them, 1 - optical fiber, 2 - spiral tensioning device, 3 - troughing idler, 4 - belt, 5 - return idler, 6 - intermediate frame, and 7 - electric roller. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0017] An embodiment of the present invention provides a roller fault monitoring system for belt conveyors based on DAS technology. Refer to Figure 1 , Figure 1 FIG. 10 is an architecture diagram of a roller fault monitoring system for belt conveyors based on DAS technology provided by an embodiment of the present invention. The system includes: A data acquisition module, configured to acquire vibration signals at multiple monitoring points through distributed optical fiber sensing technology.

[0018] A preprocessing module for preprocessing a first target signal to obtain a second target signal; the first target signal is a vibration signal of any monitoring point.

[0019] A signal image generation module for two-dimensionally characterizing the second target signal to obtain a target image.

[0020] A fault detection module for taking the target image as the input of a pre-trained fault detection model to obtain a fault detection result.

[0021] A fault warning module for, if the fault detection result indicates a fault, determining the fault location according to the monitoring point corresponding to the first target signal and sending out a warning message.

[0022] A belt conveyor idler fault monitoring system based on the DAS technology provided by an embodiment of the present invention uses an optical fiber to achieve high-density vibration monitoring of the entire line of idlers of the belt conveyor, with a wide coverage range and high accuracy. Through signal preprocessing and image recognition, idler faults can be accurately identified, significantly reducing the false judgment and missed judgment rates. The system has a high degree of automation, a fast real-time warning response, is suitable for complex environments, has simple wiring and low maintenance costs, effectively solves the problems of insufficient coverage and low efficiency of traditional detection, and improves the intelligent level of operation and maintenance.

[0023] In one embodiment, the hardware part consists of a sensing optical cable, a DAS host, a server, and a monitoring platform, where: See Figure 2 , Figure 2 is a schematic diagram of the optical cable installation provided by an embodiment of the present invention, which includes an optical fiber 1, a spiral tensioning device 2, a troughing idler 3, a belt 4, a lower idler 5, an intermediate frame 6, and an electric roller 7. The sensing optical cable uses a single-mode armored optical cable (with a diameter of 5 mm, a tensile strength of ≥800 N, and a bending radius of ≥70 mm), and is laid along the entire length of the belt conveyor on the channel steel of the intermediate frame. The "tying + bonding" composite fixing method is used to ensure that the optical cable is closely attached to the channel steel (the tying spacing ≤1 m, and the bonding material is weather-resistant silicone resin glue).

[0024] The DAS host is deployed in the power room, configured with a laser wavelength of 1550 nm, a pulse width of 10 ns, and a dynamic range of ≥90 dB. The host is connected to the sensing optical cable through a single-mode fiber optic jumper (FC / APC interface), and the backscattered signal is collected and analyzed in real time. The sampling rate is set to 3 kHz to capture high-frequency vibration characteristics.

[0025] The server receives the vibration signals collected by the host and implements the data analysis and processing process of the above monitoring system.

[0026] The monitoring platform displays the fault location and warning information in real time.

[0027] In one embodiment, the preprocessing module includes: A modal decomposition module, configured to decompose a first target signal into a plurality of modal components by using a variational mode decomposition algorithm.

[0028] A component discrimination module, configured to determine whether a first target modal component is a valid component, a noise-containing component, or a noise component according to the kurtosis of the first target modal component and its distance from the first target signal; the first target modal component is any one of the plurality of modal components.

[0029] A denoising module, configured to perform wavelet threshold denoising on the first target modal component if the first target modal component is a noise-containing component, to obtain a second target modal component.

[0030] A signal reconstruction module, configured to perform signal reconstruction according to the valid components and the denoised second target modal component, to obtain a second target signal.

[0031] In this embodiment, the vibration signal is decomposed into a plurality of modal components by variational mode decomposition, and the component type is judged by combining kurtosis and distance. The wavelet threshold method is used to denoise the noise-containing components, and finally the signal is reconstructed, effectively retaining the fault characteristics and suppressing the noise interference. This solution improves the signal purity and the accuracy of feature extraction, and provides a more reliable data basis for subsequent image generation and fault detection models.

[0032] In one implementation, the modal decomposition module includes: A parameter optimization module, configured to optimize the parameters of the variational mode decomposition by using a particle swarm optimization algorithm and a fitness function with the maximum kurtosis as the target, to obtain the decomposition number K and the penalty factor α.

[0033] A variational decomposition module, configured to perform variational mode decomposition on the first target signal according to the decomposition number K and the penalty factor α, to obtain a plurality of modal components.

[0034] Compared with the fixed parameters, the optimized decomposition is more suitable for the signal characteristics, helps to reduce modal aliasing and information loss, and significantly enhances the subsequent denoising and fault identification effects.

[0035] In one implementation, the component discrimination module includes: A kurtosis calculation module, configured to calculate the kurtosis KU of the first target modal component.

[0036] A distance calculation module, configured to calculate the Euclidean distance ED between the first target modal component and the first target signal.

[0037] An index calculation module, configured to calculate the effective index value EV of the first target modal component according to the kurtosis and the Euclidean distance: ; A fraction calculation module for calculating the Z-score of the effective index value of each modal component: ; u and σ are the mean and standard deviation of the effective index value respectively; EV i is the effective index value of the i-th modal component; Z i is the Z-score of the i-th modal component.

[0038] A noise determination module for determining the modal components with Z-scores less than a preset threshold as noise components. Specifically, the preset threshold can be set to -1.5.

[0039] A clustering and partitioning module for performing binary clustering on the remaining modal components, classifying the category with higher effective index values as effective components and the other category as noisy components.

[0040] This implementation method constructs the effective index value by fusing kurtosis and Euclidean distance, and introduces the Z-score and clustering algorithm to perform multi-level discrimination on modal components, which can more accurately distinguish effective components, noisy components and noise components. This method comprehensively considers the sharpness of modal components and the similarity with the original signal, and the discrimination process is more objective and stable, avoiding human intervention. Compared with the traditional threshold method, it can significantly improve the accuracy of modal screening, enhance the effect of retaining fault features, and lay a more reliable data foundation for subsequent noise reduction and reconstruction.

[0041] In one implementation, the denoising module includes: A wavelet decomposition module for performing multi-level wavelet decomposition on the first target modal component to obtain a set of wavelet coefficients, including low-frequency approximation coefficients and multiple high-frequency detail coefficients. Specifically, the db4 wavelet basis can be used to perform discrete wavelet transform on the modal component.

[0042] A threshold processing module for performing threshold processing on multiple high-frequency detail coefficients using a preset threshold function to obtain multiple target detail coefficients; the threshold function is: ; where is the target detail coefficient after threshold processing; is the high-frequency detail coefficient at the j-th scale, is its corresponding wavelet threshold; sgn() is the sign function; exp is the exponential function with the natural constant e as the base.

[0043] A component reconstruction module for performing inverse wavelet transform according to the low-frequency approximation coefficients and multiple target detail coefficients to obtain the denoised second target modal component.

[0044] This method extracts the multi-scale features of the modal components through multi-level wavelet decomposition, and uses a smooth and continuous improved threshold function to perform non-linear processing on the high-frequency detail coefficients, effectively suppressing noise while retaining the fault signal. Compared with the traditional hard / soft threshold method, this threshold function achieves a better balance between edge preservation and continuity, reducing the risk of signal distortion. Finally, the signal is restored through wavelet reconstruction, which helps to obtain a second target modal component with a higher signal-to-noise ratio and more integrity, providing a cleaner data basis for subsequent fault identification.

[0045] In one embodiment, the signal image generation module includes: A two-dimensional characterization module for converting the second target signal into a two-dimensional coefficient matrix using continuous wavelet transform.

[0046] A color map mapping module for performing pseudo-color mapping on the two-dimensional coefficient matrix using the Jet color map to generate a target image.

[0047] This embodiment can fully retain the time-domain and frequency-domain characteristics of the signal and realize the two-dimensional visual expression of fault characteristics. Compared with the one-dimensional waveform input, the two-dimensional image is more easily recognized by the deep learning model, improving the accuracy of fault detection. The Jet color map enhances the visual contrast of high-frequency abnormal features, helping to highlight weak fault information.

[0048] In one implementation, the complex Morlet wavelet function can be used for continuous wavelet transform, with the sampling rate set to 3200Hz, the scale range from 1 to 128, and the step size of 1.

[0049] In one embodiment, refer to Figure 3 , Figure 3 which is the network architecture diagram of a fault detection model provided by an embodiment of the present invention. The fault detection model includes a feature extraction network and a classifier; the feature extraction network includes a first convolutional block, a second convolutional block, a third convolutional block, a fourth convolutional block, and a fifth convolutional block connected in sequence; the classifier includes a flattening layer, a first feature compressor, a second feature compressor, and an output layer connected in sequence; among them, the flattening layer serves as the interface between the feature extraction network and the classifier, converting the multi-dimensional features output by the fifth convolutional block into a one-dimensional vector as the input of the first feature compressor; specifically: Each convolutional block includes multiple convolutional layers and a pooling layer; the size of the convolutional kernel of the convolutional layer is 3×3, the step size is 1, and the activation function is ReLU; the size of the pooling kernel of the pooling layer is 2×2, and the step size is 2; the number of convolutional layers in the first convolutional block, the second convolutional block, the third convolutional block, the fourth convolutional block, and the fifth convolutional block are 2, 2, 3, 3, and 3 respectively; Each feature compressor consists of a fully connected layer, ReLU activation, and Dropout regularization; during the training process, each feature compressor will randomly lose 50% of the neurons to prevent overfitting; The output layer consists of a fully connected layer and a Sigmoid activation.

[0050] The fault detection model gradually extracts features through five convolutional blocks. Each convolutional block contains multiple convolutional layers and pooling layers, which can gradually extract multi-scale and multi-level features from low-level to high-level, helping to capture details and global information in the fault signal and improving the detection accuracy. Two cascaded feature compressors can gradually compress and refine features. At the same time, Dropout randomly drops 50% of the neurons, effectively preventing the model from overfitting during training and enhancing the generalization ability. The overall structure of the model is neither too complex (to avoid wasting computing resources) nor can it fully extract and utilize fault-related features, which is conducive to achieving efficient and accurate fault detection.

[0051] The training of this model is mainly divided into the following steps: Step 1, data collection and generation of the dataset Extract the vibration signals of the files containing vibration signals generated by the DAS device. After preprocessing, shuffle the order through a random seed to ensure reproducibility. Generate the training set, validation set, and test set for the fault signals and normal signals at the same time according to the ratio of 7:2:1, generate the dataset, and save it as a joblib file.

[0052] Step 2: Wavelet transform feature extraction Since this model is an image recognition model, a dataset of time-frequency images needs to be generated according to the generated joblib file. Use continuous wavelet transform to convert the one-dimensional vibration signal into a two-dimensional time-frequency diagram. For each dataset, that is, the samples of the training set, validation set, and test set generated before, generate an image with a size of 224*224 pixels for both the fault signals and normal signals, adapt to the input size of the model network, and generate them in batches in the training set, validation set, and test set directories in the way of {fault / normal}_{sequence number}.png (such as normal_1.png), and convert all images into RGB mode uniformly.

[0053] Then convert the picture dataset into a PyTorch dataset, convert each picture into a PyTorch tensor (shape: [3,224,224]), and normalize each channel of the RGB image. Generate labels according to the file names, where normal represents '0' and fault represents '1', and save the tensorized dataset (including the training set, validation set, and test set) again.

[0054] Step 3: Construction of the fault detection model: Define the network architecture of the fault detection model, see Figure 3 , Figure 3It is a network architecture diagram of a fault detection model provided by an embodiment of the present invention. After instantiating the model and moving the model to the GPU, it is initialized.

[0055] Step 4: Model training: When training the data, the samples are divided into batches, with 32 samples in each batch. When training the training set, the order is shuffled, while the validation set is not shuffled. The processed data is put into the established binary classification convolutional neural network model. The optimizer used for training is the Adam optimizer. The loss function uses the binary cross-entropy loss for classification tasks combined with the loss function with Sigmoid activation. Set 30 training epochs, set the learning rate to 0.00001. During training, record the average loss and accuracy of each epoch, monitor the performance of the model on the validation set, and save the optimal parameter model obtained in each epoch in the pre-specified path. Finally, obtain the parameter model with the lowest validation loss.

[0056] It should be noted that in this article, terms such as "including", "comprising" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article or device.

[0057] The above has described the embodiments of the present invention in detail, but the content described is only the preferred embodiments of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of the application of the present invention should still fall within the scope covered by the patent of the present invention.

Claims

1. Belt conveyor idler fault monitoring system based on DAS technology, characterized in that, The system includes: A data acquisition module, configured to acquire vibration signals of multiple monitoring points through distributed optical fiber sensing technology; A preprocessing module, configured to preprocess a first target signal to obtain a second target signal; the first target signal is the vibration signal of any one of the monitoring points; A signal image generation module, configured to perform two-dimensional characterization on the second target signal to obtain a target image; A fault detection module, configured to use the target image as the input of a pre-trained fault detection model to obtain a fault detection result; A fault warning module, configured to, if the fault detection result indicates a fault, determine the fault location according to the monitoring point corresponding to the first target signal and send out a warning message.

2. The belt conveyor idler fault monitoring system based on the DAS technology according to claim 1, wherein The preprocessing module includes: A modal decomposition module, configured to decompose the first target signal into multiple modal components by using a variational modal decomposition algorithm; A component discrimination module, configured to determine whether the first target modal component is a valid component, a noise-containing component or a noise component according to the kurtosis of the first target modal component and its distance from the first target signal; the first target modal component is any one of the multiple modal components; A denoising module, configured to, if the first target modal component is a noise-containing component, perform wavelet threshold denoising on the first target modal component to obtain a second target modal component; A signal reconstruction module, configured to perform signal reconstruction according to the valid components and the denoised second target modal components to obtain a second target signal.

3. The belt conveyor idler fault monitoring system based on DAS technology according to claim 2, characterized in that, The modal decomposition module includes: A parameter optimization module, configured to use a particle swarm optimization algorithm and a fitness function with the maximum kurtosis as the target to optimize the parameters of the variational modal decomposition to obtain the decomposition number K and the penalty factor α; A variational decomposition module, configured to perform variational modal decomposition on the first target signal according to the decomposition number K and the penalty factor α to obtain multiple modal components.

4. The idler fault monitoring system of the belt conveyor based on the DAS technology according to claim 2, wherein, The component discrimination module includes: A kurtosis calculation module, configured to calculate the kurtosis KU of the first target modal component; A distance calculation module, configured to calculate the Euclidean distance ED between the first target modal component and the first target signal; An index calculation module, configured to calculate the effective index value EV of the first target modal component according to the kurtosis and the Euclidean distance; ； A score calculation module, configured to calculate the Z-score of the effective index values of each modal component; A noise determination module, configured to determine the modal components with Z-scores less than a preset threshold as noise components; A clustering division module, configured to perform binary classification clustering on the remaining modal components, classify the category with a higher effective index value as valid components, and classify the other category as noise-containing components.

5. The belt conveyor idler fault monitoring system based on DAS technology according to claim 2, characterized in that, The denoising module includes: A wavelet decomposition module, configured to perform multi-level wavelet decomposition on the first target modal component to obtain a set of wavelet coefficients, including low-frequency approximation coefficients and multiple high-frequency detail coefficients; A threshold processing module, configured to perform threshold processing on the multiple high-frequency detail coefficients by using a preset threshold function to obtain multiple target detail coefficients; the threshold function is: ； Among them, is the target detail coefficient after threshold processing; is the high-frequency detail coefficient at the j-th layer scale, is its corresponding wavelet threshold; sgn() is the sign function; exp is the exponential function with the natural constant e as the base; A component reconstruction module, configured to perform inverse wavelet transform according to the low-frequency approximation coefficients and the multiple target detail coefficients to obtain the denoised second target modal component.

6. The belt conveyor idler fault monitoring system based on the DAS technology according to claim 1, characterized in that The signal image generation module includes: A two-dimensional characterization module, which is used to convert the second target signal into a two-dimensional coefficient matrix by using continuous wavelet transform; A color map mapping module, which is used to perform pseudo-color mapping on the two-dimensional coefficient matrix by using the Jet color map to generate a target image.

7. The belt conveyor idler fault monitoring system based on DAS technology according to claim 1, characterized in that, The fault detection model includes a feature extraction network and a classifier; the feature extraction network includes a first convolutional block, a second convolutional block, a third convolutional block, a fourth convolutional block, and a fifth convolutional block connected in sequence; the classifier includes a flattening layer, a first feature compressor, a second feature compressor, and an output layer connected in sequence; wherein, the flattening layer serves as an interface between the feature extraction network and the classifier, and converts the multi-dimensional features output by the fifth convolutional block into a one-dimensional vector as the input of the first feature compressor; specifically: Each convolutional block includes a plurality of convolutional layers and a pooling layer; the size of the convolutional kernel of the convolutional layer is 3×3, the stride is 1, and the activation function is ReLU; the size of the pooling layer kernel is 2×2, and the stride is 2; the number of convolutional layers in the first convolutional block, the second convolutional block, the third convolutional block, the fourth convolutional block, and the fifth convolutional block are 2, 2, 3, 3, and 3 respectively; Each feature compressor consists of a fully connected layer, ReLU activation, and Dropout regularization; during the training process, each feature compressor will randomly lose 50% of the neurons to prevent overfitting; The output layer consists of a fully connected layer and Sigmoid activation.

Images