Belt conveyor roller fault monitoring system based on DAS technology

Abstract

The present invention discloses a belt conveyor roller fault monitoring system based on DAS technology, which relates to the technical field of industrial equipment status monitoring. The system obtains vibration signals from multiple monitoring points through distributed optical fiber sensing technology; pre-processes the vibration to obtain a target signal; performs two-dimensional characterization on the target signal to obtain a target image; and uses the target image as the input of a pre-trained fault detection model to obtain a fault detection result. The system uses optical fiber to achieve high-density vibration monitoring of all belt conveyor rollers, with a wide coverage range and high accuracy. Through signal pre-processing and image recognition, roller faults can be accurately identified, significantly reducing the misjudgment and missed judgment rate. The system has a high degree of automation, fast real-time alarm response, adaptability to complex environments, simple wiring, and low maintenance costs. It effectively solves the problems of insufficient coverage and low efficiency of traditional detection, and improves the level of intelligent operation and maintenance.

Description

Technical Field

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

[0002] Belt conveyors, as core material handling equipment, play a vital role in modern bulk material transportation industries, such as coal, electricity, ports, and mining. Rollers, a crucial component of belt conveyors, are numerous, widely distributed, and operate in complex and harsh environments. Long-term high-speed rotation makes them susceptible to wear, eccentricity, seizure, and bearing failure. Damage to these rollers not only affects the normal operation of the equipment but can also cause accidents such as belt tearing, material spillage, and even fires, resulting in serious economic losses and potential safety hazards.

[0003] Traditional roller fault detection relies primarily on manual inspections or point-based sensor deployments, resulting in low detection efficiency, long inspection cycles, and high labor intensity. Furthermore, due to the long distribution distances, large number of rollers, and complex operating environments, traditional fault detection suffers from limited coverage, frequent misjudgments and omissions, and low efficiency. This makes it difficult to meet the current industrial demand for high-reliability monitoring requiring "early detection, early warning, and early response." Furthermore, the complex wiring and maintenance difficulties of traditional electrical sensors limit their widespread application in large-scale conveyor lines.

[0004] Against the backdrop of the rapid development of industrial intelligence and informatization, distributed optical sensing (DAS) technology, an emerging advanced technology capable of long-distance, multi-point, real-time vibration monitoring, is gaining increasing attention. However, existing distributed optical fiber monitoring technologies are primarily used for pipeline leaks or perimeter security, and no method can be directly applied to 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 technology, and to propose a belt conveyor roller fault monitoring system based on DAS technology.

[0006] The present invention provides a belt conveyor roller fault monitoring system based on DAS technology, the system comprising:

[0007] The data acquisition module is used to obtain vibration signals of multiple monitoring points through distributed optical fiber sensing technology;

[0008] A preprocessing module is used to preprocess the first target signal to obtain a second target signal; the first target signal is a vibration signal of any monitoring point;

[0009] a signal image generating module, configured to perform two-dimensional characterization on the second target signal to obtain a target image;

[0010] a fault detection module, configured to use the target image as input to a pre-trained fault detection model to obtain a fault detection result;

[0011] The fault alarm module is used to determine the fault location according to the monitoring point corresponding to the first target signal and issue an alarm message if the fault detection result is that a fault exists.

[0012] Optionally, the preprocessing module includes:

[0013] A modal decomposition module, configured to decompose the first target signal into multiple modal components using a variational modal decomposition algorithm;

[0014] a component discrimination module, configured to determine, based on the kurtosis of the first target modal component and the distance between the first target modal component and the first target signal, whether the first target modal component is a valid component, a noisy component, or a noise component; the first target modal component is any one of a plurality of modal components;

[0015] a denoising module, configured to perform wavelet threshold denoising on the first target modal component to obtain a second target modal component if the first target modal component is a noisy component;

[0016] The signal reconstruction module is used to reconstruct the signal according to the effective component and the denoised second target modal component to obtain the second target signal.

[0017] Optionally, the modal decomposition module includes:

[0018] The parameter optimization module is used to optimize the parameters of the variational mode decomposition using the particle swarm optimization algorithm and the fitness function with the maximum kurtosis as the goal, and obtain the decomposition number K and the penalty factor α;

[0019] The variational decomposition module is used 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.

[0020] Optionally, the component discrimination module includes:

[0021] a kurtosis calculation module, configured to calculate the kurtosis KU of the first target modal component;

[0022] a distance calculation module, configured to calculate a Euclidean distance ED between the first target modal component and the first target signal;

[0023] An index calculation module is used to calculate the effective index value EV of the first target modal component according to the kurtosis and the Euclidean distance:

[0024] ;

[0025] Score calculation module, used to calculate the Z score of the effective index value of each modal component;

[0026] a noise determination module, configured to determine a modal component having a Z score less than a preset threshold as a noise component;

[0027] The clustering module is used to perform binary clustering on the remaining modal components, classifying the class with higher effective index value as effective component and the other class as noisy component.

[0028] Optionally, the denoising module includes:

[0029] a wavelet decomposition module, configured to perform wavelet multi-layer 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;

[0030] The threshold processing module is used to perform threshold processing on multiple high-frequency detail coefficients using a preset threshold function to obtain multiple target detail coefficients; the threshold function is:

[0031] ;

[0032] in, is the target detail coefficient after threshold processing; is the high-frequency detail coefficient of the j-th scale, is the corresponding wavelet threshold; sgn() is the sign function; exp is the exponential function with the natural constant e as the base;

[0033] The component reconstruction module is used to perform inverse wavelet transform according to the low-frequency approximation coefficient and multiple target detail coefficients to obtain a denoised second target modal component.

[0034] Optionally, the signal image generation module includes:

[0035] a two-dimensional characterization module, configured to convert the second target signal into a two-dimensional coefficient matrix using a continuous wavelet transform;

[0036] The color map mapping module is used to use the Jet color map to perform pseudo color mapping on the two-dimensional coefficient matrix to generate a target image.

[0037] Optionally, the fault detection model includes a feature extraction network and a classifier; the feature extraction network includes a first convolution block, a second convolution block, a third convolution block, a fourth convolution block, and a fifth convolution 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 multidimensional features output by the fifth convolution block into a one-dimensional vector as input to the first feature compressor; specifically:

[0038] Each convolution block consists of multiple convolutional layers and a pooling layer; the convolution kernel size of the convolution layer is 3×3, the stride is 1, and the activation function is ReLU; the pooling layer kernel size is 2×2, and the stride is 2; the number of convolutional layers in the first convolution block, the second convolution block, the third convolution block, the fourth convolution block, and the fifth convolution block are 2, 2, 3, 3, and 3 respectively;

[0039] Each feature compressor consists of a fully connected layer, ReLU activation, and Dropout regularization. During training, each feature compressor randomly loses 50% of its neurons to prevent fitting from solidification.

[0040] The output layer consists of a fully connected layer and a Sigmoid activation.

[0041] Beneficial effects of the present invention:

[0042] The present invention proposes a belt conveyor roller fault monitoring system based on DAS technology. The system obtains vibration signals of multiple monitoring points through distributed optical fiber sensing technology; preprocesses the first target signal to obtain a second target signal; the first target signal is the vibration signal of any monitoring point; the second target signal is two-dimensionally represented to obtain a target image; the target image is used as the input of a pre-trained fault detection model to obtain a fault detection result; if the fault detection result is that a fault exists, the fault location is determined according to the monitoring point corresponding to the first target signal, and an alarm message is issued.

[0043] Based on DAS technology, this system utilizes optical fiber to achieve high-density vibration monitoring of rollers along the entire belt conveyor line, offering wide coverage and high accuracy. Through signal preprocessing and image recognition, it accurately identifies roller faults, significantly reducing false positives and missed detections. The system boasts a high degree of automation, rapid real-time alarm response, adaptability to complex environments, simplified wiring, and low maintenance costs. It effectively addresses the inadequate coverage and low efficiency of traditional detection, enhancing the intelligence of operations and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 An architecture diagram of a belt conveyor roller fault monitoring system based on DAS technology is provided for an embodiment of the present invention;

[0045] Figure 2 A schematic diagram of optical cable installation is provided for an embodiment of the present invention;

[0046] Figure 3 A network architecture diagram of a fault detection model is provided for an embodiment of the present invention;

[0047] Among them, 1-optical fiber, 2-spiral tensioning device, 3-grooved roller, 4-belt, 5-lower roller, 6-middle frame and 7-electric roller. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0049] The embodiment of the present invention provides a belt conveyor roller fault monitoring system based on DAS technology. Figure 1 , Figure 1 This is an architecture diagram of a belt conveyor roller fault monitoring system based on DAS technology provided by an embodiment of the present invention. The system includes:

[0050] The data acquisition module is used to obtain vibration signals from multiple monitoring points through distributed optical fiber sensing technology.

[0051] The preprocessing module is used to preprocess the first target signal to obtain the second target signal; the first target signal is the vibration signal of any monitoring point.

[0052] The signal image generation module is used to perform two-dimensional characterization on the second target signal to obtain a target image.

[0053] The fault detection module is used to take the target image as the input of the pre-trained fault detection model to obtain the fault detection result.

[0054] The fault alarm module is used to determine the fault location according to the monitoring point corresponding to the first target signal and issue an alarm message if the fault detection result shows that a fault exists.

[0055] A belt conveyor roller fault monitoring system based on DAS technology, provided by an embodiment of the present invention, utilizes optical fiber to achieve high-density vibration monitoring of rollers along the entire belt conveyor line, offering wide coverage and high accuracy. Through signal preprocessing and image recognition, roller faults can be accurately identified, significantly reducing the rate of false positives and missed detections. The system boasts a high degree of automation, rapid real-time alarm response, adaptability to complex environments, simplified wiring, and low maintenance costs. This effectively addresses the challenges of insufficient coverage and low efficiency of traditional detection systems, enhancing the intelligence of operations and maintenance.

[0056] In one embodiment, the hardware consists of a sensor cable, a DAS host, a server, and a monitoring platform, wherein:

[0057] See also Figure 2 , Figure 2 This diagram illustrates an optical cable installation according to an embodiment of the present invention, including an optical fiber 1, a spiral tensioning device 2, trough rollers 3, a belt 4, a lower roller 5, an intermediate frame 6, and a motorized drum 7. The sensor cable utilizes a single-mode armored optical cable (5 mm diameter, tensile strength ≥ 800 N, bending radius ≥ 70 mm) and is laid along the entire length of the belt conveyor to the intermediate frame channel steel. A "binding + bonding" composite fastening method ensures a tight fit between the cable and the channel steel (binding spacing ≤ 1 meter, with weather-resistant silicone resin adhesive as the bonding material).

[0058] The DAS host is deployed in the power room and is equipped with a laser wavelength of 1550nm, a pulse width of 10ns, and a dynamic range of ≥90dB. The host is connected to the sensor cable via a single-mode fiber optic patch cord (FC / APC interface), collecting and analyzing backscattered signals in real time. The sampling rate is set to 3kHz to capture high-frequency vibration characteristics.

[0059] The server receives the vibration signal collected by the host and implements the data analysis and processing process of the above monitoring system.

[0060] Monitoring platform, which displays fault location and alarm information in real time.

[0061] In one embodiment, the pre-processing module includes:

[0062] The modal decomposition module is used to decompose the first target signal into multiple modal components using a variational modal decomposition algorithm.

[0063] A component discrimination module is used to determine whether the first target modal component is a valid component, a noisy component, or a noise component based on 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.

[0064] The denoising module is configured to perform wavelet threshold denoising on the first target modal component if the first target modal component is a noisy component, so as to obtain a second target modal component.

[0065] The signal reconstruction module is used to reconstruct the signal according to the effective component and the denoised second target modal component to obtain the second target signal.

[0066] This embodiment decomposes the vibration signal into multiple modal components through variational modal decomposition, combines kurtosis and distance to determine the component type, uses the wavelet threshold method to reduce the noise of the noisy components, and finally reconstructs the signal, effectively retaining the fault characteristics and suppressing noise interference. This solution improves signal purity and feature extraction accuracy, providing a more reliable data foundation for subsequent image generation and fault detection models.

[0067] In one implementation, the modal decomposition module includes:

[0068] The parameter optimization module is used to optimize the parameters of variational mode decomposition using the particle swarm optimization algorithm and the fitness function with the maximum kurtosis as the goal, and obtain the decomposition number K and the penalty factor α.

[0069] The variational decomposition module is used 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.

[0070] Compared with fixed parameters, the optimized decomposition better fits the signal characteristics, helps reduce modal aliasing and information loss, and significantly enhances the subsequent denoising and fault identification effects.

[0071] In one implementation, the component discrimination module includes:

[0072] The kurtosis calculation module is used to calculate the kurtosis KU of the first target modal component.

[0073] The distance calculation module is used to calculate the Euclidean distance ED between the first target modal component and the first target signal.

[0074] The index calculation module is used to calculate the effective index value EV of the first target modal component based on the kurtosis and Euclidean distance: ;

[0075] The score calculation module is used to calculate 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 values ​​respectively; EV i is the effective index value of the i-th modal component; Z i is the Z-score of the ith modal component.

[0076] The noise determination module is configured to determine a modal component with a Z score less than a preset threshold as a noise component. Specifically, the preset threshold may be set to -1.5.

[0077] The clustering module is used to perform binary clustering on the remaining modal components, classifying the class with higher effective index value as effective component and the other class as noisy component.

[0078] This implementation constructs an effective index value by integrating kurtosis and Euclidean distance, and introduces Z-scores and clustering algorithms for multi-layered discrimination of modal components, enabling more accurate distinction between effective, noisy, and noise components. This method comprehensively considers the sharpness of modal components and their similarity to the original signal, making the discrimination process more objective and stable, and avoiding human intervention. Compared with traditional threshold methods, this method significantly improves modal screening accuracy and enhances fault feature retention, laying a more reliable data foundation for subsequent noise reduction and reconstruction.

[0079] In one implementation, the denoising module includes:

[0080] The wavelet decomposition module is used to perform wavelet multi-layer decomposition on the first target modal component to obtain a set of wavelet coefficients, including low-frequency approximate coefficients and multiple high-frequency detail coefficients. Specifically, the modal component can be discretely transformed using the db4 wavelet basis.

[0081] The threshold processing module is used to perform threshold processing on multiple high-frequency detail coefficients using a preset threshold function to obtain multiple target detail coefficients; the threshold function is:

[0082] ;

[0083] in, is the target detail coefficient after threshold processing; is the high-frequency detail coefficient of the j-th scale, is the corresponding wavelet threshold; sgn() is the sign function; exp is the exponential function with the natural constant e as the base.

[0084] The component reconstruction module is used to perform inverse wavelet transform based on the low-frequency approximation coefficient and multiple target detail coefficients to obtain a denoised second target modal component.

[0085] This approach extracts multiscale features of modal components through multi-layer wavelet decomposition and employs a smooth, continuous, and improved threshold function to perform nonlinear processing on high-frequency detail coefficients, effectively suppressing noise while preserving the fault signal. Compared to traditional hard / soft thresholding methods, this threshold function achieves a better balance between edge preservation and continuity, reducing the risk of signal distortion. Ultimately, wavelet reconstruction and restoration of the signal help obtain a higher signal-to-noise ratio and a more complete second target modal component, providing a cleaner data foundation for subsequent fault identification.

[0086] In one embodiment, the signal image generation module includes:

[0087] The two-dimensional characterization module is used to convert the second target signal into a two-dimensional coefficient matrix by using continuous wavelet transform.

[0088] The color map mapping module is used to use the Jet color map to perform pseudo color mapping on the two-dimensional coefficient matrix to generate a target image.

[0089] This embodiment fully preserves the signal's time and frequency domain characteristics, enabling two-dimensional visualization of fault features. Compared to one-dimensional waveform input, two-dimensional images are more easily recognized by deep learning models, improving fault detection accuracy. Jet color maps enhance the visual contrast of high-frequency anomalies, helping to highlight subtle fault information.

[0090] In one implementation, a complex Morlet wavelet function may be used for continuous wavelet transform, with a sampling rate of 3200 Hz, a scale range of 1 to 128, and a step size of 1.

[0091] In one embodiment, see Figure 3 , Figure 3 This is a 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 convolution block, a second convolution block, a third convolution block, a fourth convolution block, and a fifth convolution 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; the flattening layer serves as an interface between the feature extraction network and the classifier, converting the multidimensional features output by the fifth convolution block into a one-dimensional vector, which serves as the input to the first feature compressor. Specifically:

[0092] Each convolution block consists of multiple convolutional layers and a pooling layer; the convolution kernel size of the convolution layer is 3×3, the stride is 1, and the activation function is ReLU; the pooling layer kernel size is 2×2, and the stride is 2; the number of convolutional layers in the first convolution block, the second convolution block, the third convolution block, the fourth convolution block, and the fifth convolution block are 2, 2, 3, 3, and 3 respectively;

[0093] Each feature compressor consists of a fully connected layer, ReLU activation, and Dropout regularization. During training, each feature compressor randomly loses 50% of its neurons to prevent fitting from solidification.

[0094] The output layer consists of a fully connected layer and a Sigmoid activation.

[0095] This fault detection model progressively extracts features through five convolutional blocks. Each convolutional block contains multiple convolutional and pooling layers. This allows for the progressive extraction of multi-scale, multi-layer features from low-level to high-level, helping to capture both detailed and global information in fault signals and improve detection accuracy. Two cascaded feature compressors progressively compress and refine features, while dropout, which randomly drops 50% of neurons, effectively prevents overfitting during training and improves generalization. The overall model structure is both moderately complex (to avoid wasting computational resources) and fully extracts and utilizes fault-related features, facilitating efficient and accurate fault detection.

[0096] The training of this model is mainly divided into the following steps:

[0097] Step 1: Data collection and dataset generation

[0098] Extract vibration signals from files containing vibration signals generated by DAS equipment. After preprocessing, scramble the order using a random seed to ensure reproducibility. Generate a training set, validation set, and test set using the data of fault signals and normal signals in a ratio of 7:2:1. Generate a dataset and save it as a joblib file.

[0099] Step 2: Wavelet transform feature extraction

[0100] Because this model is an image recognition model, it is necessary to regenerate a time-frequency image dataset based on the generated joblib file. Using a continuous wavelet transform, the one-dimensional vibration signal is converted into a two-dimensional time-frequency graph. For each dataset (i.e., the previously generated training, validation, and test sets), a 224*224 pixel image is generated for both the faulty and normal signals. This image is adapted to the input size of the model network and batched into the training, validation, and test set directories using the {fault / normal}_{serial number}.png format (e.g., normal_1.png). All images are then converted to RGB mode.

[0101] Next, convert the image dataset into a PyTorch dataset, convert each image into a PyTorch tensor (shape: [3, 224, 224]), and then normalize each channel of the RGB image. Generate labels based on the file name, with normal representing '0' and fault representing '1'. Save the tensorized dataset again (including training set, validation set, and test set).

[0102] Step 3: Fault detection model construction:

[0103] Define the network architecture of the fault detection model, see Figure 3 , Figure 3A network architecture diagram of a fault detection model provided by an embodiment of the present invention. The model is instantiated and moved to the GPU before initialization.

[0104] Step 4: Model training:

[0105] When training data, the samples were divided into batches of 32 samples each. The training set was shuffled, but the validation set was not. The processed data was fed into a pre-built binary classification convolutional neural network model. The Adam optimizer was used for training, and the loss function used was a binary cross-entropy loss for classification tasks combined with a sigmoid activation loss function. 30 training epochs were set, with a learning rate of 0.00001. During training, the average loss and accuracy of each epoch were recorded. The model's performance on the validation set was monitored, and the optimal parameter model obtained from each epoch was saved in a pre-specified path. The parameter model with the lowest validation loss was ultimately obtained.

[0106] It should be noted that, in this document, terms such as "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements that are inherent to such process, method, article or apparatus.

[0107] The embodiments of the present invention are described in detail above, but the contents described are only preferred embodiments of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. The belt conveyor roller fault monitoring system based on DAS technology is characterized by: The system comprises: The data acquisition module is used to obtain vibration signals of multiple monitoring points through distributed optical fiber sensing technology; A preprocessing module is used to preprocess the first target signal to obtain a second target signal; the first target signal is a vibration signal of any monitoring point; a signal image generating 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 input to a pre-trained fault detection model to obtain a fault detection result; a fault alarm module, configured to determine the fault location based on the monitoring point corresponding to the first target signal and issue an alarm message if the fault detection result indicates that a fault exists; The pre-processing module comprises: A modal decomposition module, configured to decompose the first target signal into multiple modal components using a variational modal decomposition algorithm; a component discrimination module, configured to determine, based on the kurtosis of the first target modal component and the distance between the first target modal component and the first target signal, whether the first target modal component is a valid component, a noisy component, or a noise component; the first target modal component is any one of a plurality of modal components; a denoising module, configured to perform wavelet threshold denoising on the first target modal component to obtain a second target modal component if the first target modal component is a noisy component; A signal reconstruction module is used to reconstruct the signal according to the effective component and the denoised second target modal component to obtain a second target signal; The modal decomposition module includes: The parameter optimization module is used to optimize the parameters of the variational mode decomposition using the particle swarm optimization algorithm and the fitness function with the maximum kurtosis as the goal, and 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 a plurality of modal components; 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 a Euclidean distance ED between the first target modal component and the first target signal; An index calculation module is used to calculate the effective index value EV of the first target modal component according to the kurtosis and the Euclidean distance: Score calculation module, used to calculate the Z score of the effective index value of each modal component; a noise determination module, configured to determine a modal component having a Z score less than a preset threshold as a noise component; The clustering module is used to perform binary clustering on the remaining modal components, classifying the one with a higher effective index value as the effective component and the other as the noisy component; The denoising module includes: a wavelet decomposition module, configured to perform wavelet multi-layer 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; The threshold processing module is used to perform threshold processing on multiple high-frequency detail coefficients using a preset threshold function to obtain multiple target detail coefficients; the threshold function is: in, is the target detail coefficient after threshold processing; W j,k is the high-frequency detail coefficient of the j-th scale, λ j is the corresponding wavelet threshold; sgn() is the sign function; exp is the exponential function with the natural constant e as the base; The component reconstruction module is used to perform inverse wavelet transform according to the low-frequency approximation coefficient and multiple target detail coefficients to obtain a denoised second target modal component.

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

3. The belt conveyor roller fault monitoring system based on DAS technology according to claim 1 is characterized in that: The fault detection model includes a feature extraction network and a classifier; the feature extraction network includes a first convolution block, a second convolution block, a third convolution block, a fourth convolution block, and a fifth convolution 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 multidimensional features output by the fifth convolution block into a one-dimensional vector as the input of the first feature compressor; specifically: Each convolution block consists of multiple convolutional layers and a pooling layer; the convolution kernel size of the convolution layer is 3×3, the stride is 1, and the activation function is ReLU; the pooling layer kernel size is 2×2, and the stride is 2; the number of convolutional layers in the first convolution block, the second convolution block, the third convolution block, the fourth convolution block, and the fifth convolution block are 2, 2, 3, 3, and 3 respectively; Each feature compressor consists of a fully connected layer, ReLU activation, and Dropout regularization. During training, each feature compressor randomly loses 50% of its neurons to prevent fitting from solidification. The output layer consists of a fully connected layer and a Sigmoid activation.

Images