Belt conveyor system health monitoring method, system, device and medium based on distributed fiber optic acoustic sensing

Abstract

This invention discloses a method, system, device, and medium for health monitoring of belt conveyor systems based on distributed fiber optic acoustic sensing. The method involves deploying a sensing optical cable along the belt conveyor system, maintaining mechanical coupling with the support of key rotating components; emitting optical pulses into the optical cable using phase-sensitive optical time-domain reflectometry and receiving backscattered Rayleigh light signals; demodulating and acquiring distributed dynamic strain signals from multiple spatial channels distributed along the optical cable; extracting energy feature values ​​to filter out abnormal spatial channels exceeding a preset baseline; extracting signals from the abnormal spatial channel and its adjacent spatial channels to construct a multi-channel spatiotemporal feature matrix; inputting this matrix into a pre-trained neural network model for fusion analysis, and outputting fault type and location information. This invention utilizes optical fiber to achieve long-distance, full-area, blind-spot-free monitoring in harsh industrial environments, enabling accurate location and intelligent diagnosis of early, subtle faults.

Description

Technical Field

[0001] This invention relates to the field of material conveying equipment technology, specifically to a health monitoring method, system, equipment, and medium for belt conveyor systems based on distributed optical fiber acoustic wave sensing. Background Technology

[0002] Belt conveyor systems are core equipment for transporting bulk materials in modern industries such as steel, mining, and ports. These systems contain numerous critical rotating components, such as idlers and rollers, which are widely distributed and operate under harsh environments with high temperatures, high dust levels, and strong electromagnetic interference, making them highly susceptible to wear, jamming, and other malfunctions. If these critical rotating components fail to malfunction in time, it can lead to catastrophic accidents such as belt tearing or even fires, resulting in unplanned downtime and significant economic losses. Therefore, long-distance, comprehensive, and highly reliable early health monitoring of the critical rotating components of belt conveyor systems is crucial.

[0003] Existing health monitoring of belt conveyor systems mainly relies on two methods: one is point sensor monitoring, which involves installing point vibration accelerometers or acoustic sensors at some key equipment locations; the other is regular manual inspection, which relies on the experience of inspection personnel to make sensory judgments on the operating status of the equipment.

[0004] However, existing monitoring technologies have significant drawbacks: First, point sensors require individual installation, resulting in extremely complex and costly wiring. They cannot achieve full coverage on long-distance conveyor lines, often several kilometers long, leaving significant blind spots. Furthermore, traditional electronic sensors are easily damaged by strong electromagnetic interference and harsh environments, resulting in short lifespans and high maintenance workloads. Second, manual inspections are inefficient and subjective, failing to achieve continuous, 24 / 7 monitoring. Introducing fiber optic sensing technology for long-distance monitoring faces insurmountable technical bottlenecks in practical engineering applications: First, mechanical attenuation of weak signals; high-frequency vibration signals from early, minor faults in rotating components are easily absorbed during mechanical transmission. Second, spatial crosstalk; the physical transmission of signals from the rigid supports of conveyor belts can cause severe resonance false alarms at adjacent monitoring points. Third, computational bottlenecks; long-distance, high-precision continuous sampling generates massive amounts of data, and existing centralized analysis architectures cannot meet the computational requirements for real-time inference, leading to significant time lags in early warning systems. Summary of the Invention

[0005] The main objective of this invention is to provide a method, system, device, and medium for health monitoring of belt conveyor systems based on distributed optical fiber acoustic sensing, in order to solve the technical problem of how to overcome the blind spots in the all-area monitoring of long-distance belt conveyor systems in harsh industrial environments, thereby achieving accurate location and intelligent diagnosis of early minor faults in rotating components.

[0006] To achieve the above objectives, the present invention provides a health monitoring method for a belt conveyor system based on distributed fiber optic acoustic sensing, comprising the following steps: S1. Lay a sensing optical cable along the belt conveyor system, so that the sensing optical cable is mechanically coupled to the support of the key rotating component of the belt conveyor system, so as to sense the vibration characteristics of the key rotating component. S2. Emit light pulses to the sensing optical cable and receive backscattered Rayleigh light signals, and obtain distributed dynamic strain signals of multiple spatial channels distributed along the sensing optical cable through phase demodulation. S3. Obtain the preset energy baseline of vibration intensity, extract the energy characteristic value representing vibration intensity from the distributed dynamic strain signal, and screen out the abnormal spatial channels where the energy characteristic value exceeds the preset energy baseline. S4. Obtain one upstream and one downstream adjacent spatial channel adjacent to the abnormal spatial channel, extract the distributed dynamic strain signal corresponding to the abnormal spatial channel and the adjacent spatial channel, and combine the distributed dynamic strain signal to construct a multi-channel spatiotemporal feature matrix. S5. Input the multi-channel spatiotemporal feature matrix into a pre-trained neural network model, and perform fusion analysis on the signal features of each spatial channel through the neural network model to output the fault type and location information corresponding to the abnormal spatial channel.

[0007] Furthermore, the step S1 of maintaining mechanical coupling between the sensing optical cable and the support of the key rotating component of the belt conveyor system specifically includes the following steps: The sensing optical cable is laid close to the main longitudinal beam of the belt conveyor system and is fixed by rigid metal clamps at the support connection points of each key rotating component to suppress the mechanical attenuation of high-frequency vibration signals.

[0008] Furthermore, before constructing the multi-channel spatiotemporal feature matrix in step S4, the following steps are also included: A filter is used to remove low-frequency ambient temperature drift interference from the distributed dynamic strain signal, and the filtered signal is then standardized.

[0009] Furthermore, the neural network model includes a spatial attention mechanism module, and the fusion analysis of signal features from each spatial channel using the neural network model specifically includes: The spatial attention mechanism module performs global average pooling on the time series dimension of the multi-channel spatiotemporal feature matrix to extract the channel weight vector representing the global features of each spatial channel. Generate the incentive weight vector corresponding to each spatial channel based on the channel weight vector; The excitation weight vector is multiplied element-wise with the multi-channel spatiotemporal feature matrix to amplify the signal weight of the spatial channel where the real vibration source is located and suppress the crosstalk feature weight of the adjacent spatial channels caused by the physical transmission of the support.

[0010] Furthermore, the neural network model also includes a multi-scale one-dimensional dilated convolution module, which specifically includes three parallel feature extraction branches: The first branch uses one-dimensional convolution with a first expansion rate to extract high-frequency transient impact features; wherein, the high-frequency transient impact features are caused by microscopic damage to the bearings of key rotating components; The second branch uses one-dimensional dilated convolution with a second dilation rate to extract mid-frequency vibration features, where the second dilation rate is greater than the first dilation rate; wherein the mid-frequency vibration features are caused by abnormal friction in the belt conveyor system. The third branch uses one-dimensional dilated convolution with a third dilation rate to extract long-period low-frequency degradation features, wherein the third dilation rate is greater than the second dilation rate; wherein the long-period low-frequency degradation features are caused by macroscopic structural imbalance of the rotating component; The feature maps output from the first branch, the second branch, and the third branch are fused and stitched together, and then input into the fully connected layer to classify the fault type.

[0011] Furthermore, the training steps for a neural network model include: The error of the neural network model is calculated using a focus loss function. The gradient update weights of normally operating samples during the training process are reduced by adjusting the focus loss function, so that the computing resources of the neural network model are focused on feature learning of minority class samples containing early and weak fault characteristics.

[0012] This invention also provides a health monitoring system for a belt conveyor system based on distributed fiber optic acoustic sensing, comprising: A sensing optical cable is laid along the belt conveyor system and mechanically coupled to the support of the key rotating component of the belt conveyor system to sense the vibration characteristics of the key rotating component. An optical fiber demodulation device, connected to the sensing optical cable, is used to transmit optical pulses to the sensing optical cable and receive backscattered Rayleigh light signals, and to obtain distributed dynamic strain signals of multiple spatial channels distributed along the sensing optical cable through phase demodulation. The data processing device is communicatively connected to the fiber optic demodulation device, receives the distributed dynamic strain signal, and executes steps S3 to S5 in the health monitoring method for belt conveyor systems based on distributed fiber optic acoustic wave sensing as described above.

[0013] The present invention also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the data analysis step in the health monitoring method for a belt conveyor system based on distributed optical fiber acoustic wave sensing as described above. The data analysis step specifically includes steps S3 to S5.

[0014] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the data analysis step in the health monitoring method for a belt conveyor system based on distributed optical fiber acoustic wave sensing as described above; wherein the data analysis step specifically includes steps S3 to S5.

[0015] Compared with the prior art, the present invention has the following beneficial effects: Overcoming signal attenuation: By laying sensing optical cables along the belt conveyor system and maintaining mechanical coupling with the support of key rotating components, and combining the back Rayleigh scattering light signal of the optical pulse for phase demodulation, the sensing optical cables are transformed into a continuous sensing medium for acquiring distributed dynamic strain signals. The inherent electromagnetic interference resistance of optical fibers enables long-distance, full-area, blind-spot-free monitoring in harsh industrial environments.

[0016] Breaking through the computing power bottleneck: By extracting the energy feature values ​​of the signal and combining them with the preset energy baseline to screen out abnormal spatial channels, lightweight and precise positioning of potential fault areas in massive data has been achieved.

[0017] Eliminating spatial crosstalk: By extracting the distributed dynamic strain signals corresponding to the anomalous spatial channel and its adjacent spatial channels, a multi-channel spatiotemporal feature matrix is ​​constructed. This multi-channel spatiotemporal feature matrix is ​​then input into a pre-trained neural network model for fusion analysis. By utilizing the physical transmission correlation between adjacent spatial channels and the deep feature fusion capability of the neural network, spatial signal crosstalk caused by mechanical rigid connection is effectively suppressed, and weak early degradation features are accurately extracted from strong background noise.

[0018] Precise positioning and intelligent diagnosis: The system accurately outputs the fault type and positioning information, enabling precise positioning and intelligent diagnosis of early-stage minor faults in rotating components. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating a health monitoring method for a belt conveyor system based on distributed optical fiber acoustic sensing, according to an embodiment of the present invention.

[0021] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0024] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0025] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0026] This embodiment takes a raw material conveyor belt system of approximately 2.2 kilometers in length in a steel plant as an example. The system has tens of thousands of idlers and dozens of drive drums, and is in an environment of high dust, strong vibration, and strong electromagnetic interference from the motor frequency converter for a long time.

[0027] Please see Figure 1 This embodiment provides a health monitoring method for a belt conveyor system based on distributed fiber optic acoustic sensing, including the following steps: S1. Lay a sensing optical cable along the belt conveyor system, so that the sensing optical cable is mechanically coupled to the support of the key rotating component of the belt conveyor system, so as to sense the vibration characteristics of the key rotating component. Specifically, to ensure the transmission quality of optical signals and withstand harsh environments, the preferred sensing optical cable is the G.657.A2 standard bending loss resistant single-mode sensing optical cable. In terms of installation, the conventional loose laying method using cable trays has been completely abandoned. Instead, the sensing optical cable is laid tightly against the main longitudinal beam of the belt conveyor system, and rigid metal clamps are used for hard coupling and fixation at the support connection points of each critical rotating component. Through rigid hard coupling, the high-frequency vibration signal generated by the early wear of the idler roller bearing can be transmitted to the outer sheath of the optical cable with minimal mechanical impedance across the mechanical interface, thereby causing a slight change in the refractive index of the fiber core. This effectively overcomes the serious attenuation problem of high-frequency signals during long-distance mechanical transmission, laying a solid physical foundation for subsequent feature extraction.

[0028] S2. Emit light pulses to the sensing optical cable and receive backscattered Rayleigh light signals, and obtain distributed dynamic strain signals of multiple spatial channels distributed along the sensing optical cable through phase demodulation. Specifically, a commercially available Φ-OTDR type fiber optic demodulation device is used, based on phase-sensitive optical time-domain reflectometry (OTDR). In this embodiment, the optical pulse emission and sampling frequency of the demodulation device is set to 1000Hz, and the spatial resolution of the demodulation is set to 1 meter. This means that the 2.2-kilometer sensing optical cable is virtually divided into 2200 independent spatial channels, each channel generating 1000 strain data points per second. Through high-frequency phase demodulation, the ordinary communication optical cable is transformed into a continuously distributed ultra-high-density vibration sensor array, achieving comprehensive state perception of the belt conveyor line without blind spots.

[0029] S3. Obtain the preset energy baseline of vibration intensity, extract the energy characteristic value representing vibration intensity from the distributed dynamic strain signal, and screen out the abnormal spatial channels where the energy characteristic value exceeds the preset energy baseline. Specifically, due to the massive amount of data generated by 2200 channels, which is prone to "data avalanche," the edge computing server in this embodiment performs lightweight pre-detection at the front end. It calculates the root mean square (RMS) value or short-time energy of the signal within a 1-second sliding time window for each spatial channel as an energy characteristic value. When the belt conveyor is running normally, the energy characteristic value of each channel fluctuates around a baseline as the belt load changes. Once a jam or severe friction occurs at a certain point, the energy characteristic value of the corresponding spatial channel will instantly break through the preset energy baseline, and the system immediately marks it as an "abnormal spatial channel." Through this extremely lightweight sliding window energy pre-detection mechanism, the system can filter out 99% of normal redundant data, perfectly solving the real-time inference computing power bottleneck problem caused by the continuous input of massive high-frequency time-series data into the neural network model.

[0030] S4. Obtain one upstream and one downstream adjacent spatial channel adjacent to the abnormal spatial channel, extract the distributed dynamic strain signal corresponding to the abnormal spatial channel and the adjacent spatial channel, and combine the distributed dynamic strain signal to construct a multi-channel spatiotemporal feature matrix. Before constructing the feature matrix, signal preprocessing is performed. Because slow changes in ambient temperature can cause significant low-frequency phase drift in the raw fiber optic signal, directly inputting it into the neural network would lead to instantaneous saturation of the activation function. Therefore, a high-pass filter with a cutoff frequency of 0.5Hz is first used to filter out low-frequency ambient temperature drift interference in the phase signal, followed by normalization using the Z-score algorithm. Since the frequency of ambient temperature changes is typically below 0.1Hz, a 0.5Hz cutoff frequency will not filter out effective vibration signals.

[0031] After preprocessing, channel combination is performed. Assume the coordinates of the abnormal spatial channel triggering the alarm are z. Considering that the physical spacing of the idlers is typically between 1 and 1.5 meters, and highly compatible with the 1-meter spatial resolution of the demodulation equipment, the system extracts z along with the adjacent upstream and downstream channels z-1 and z+1, totaling three spatial channels. Since the sampling rate is 1000Hz, the system extracts data for a total of 2 seconds before and after the trigger, i.e., a time window of 2000 sampling points. This 2-second window can cover a complete rotation cycle of a typical roller (approximately 60 rpm), ensuring that periodic fault characteristics are fully captured. Finally, a multi-channel spatiotemporal feature matrix with dimensions (3, 2000) is constructed. This step ensures the numerical stability of the network training through temperature drift filtering; and extracting z-1, z, and z+1 together to construct the (3, 2000) tensor essentially forcibly injects the physical spatial transmission information of the conveyor belt support into the data structure, providing a complete contextual data source for subsequent model elimination of spatial crosstalk.

[0032] S5. Input the multi-channel spatiotemporal feature matrix into a pre-trained neural network model, and perform fusion analysis on the signal features of each spatial channel through the neural network model to output the fault type and location information corresponding to the abnormal spatial channel.

[0033] Specifically, the neural network model comprises two core modules: The first module is the spatial attention mechanism: it performs global average pooling on the time dimension of the (3, 2000) multi-channel spatiotemporal feature matrix, compressing it into a channel weight vector of size (3, 1). This vector is then passed through a dimension reduction fully connected layer, a dimension increase fully connected layer, and a Sigmoid mapping function to generate the excitation weight vector corresponding to each spatial channel. Finally, the excitation weight vector is multiplied element-wise with the original feature tensor along the channel dimension. When the idler roller at z is actually damaged, the strong mechanical vibration will be transmitted along the rigid support to z-1, causing resonance crosstalk. The spatial attention mechanism can automatically learn this causal relationship, amplify the signal weight of the real vibration source z channel, and actively attenuate the signal weight of the passive resonance channel z-1, thereby completely solving the problem of spatial false alarms caused by rigid structure transmission.

[0034] The second component is a multi-scale one-dimensional dilated convolution module: This module has three parallel feature extraction branches. The first branch uses a small-size one-dimensional convolution with a dilation rate of 1 to extract high-frequency transient impact features caused by microscopic damage to rotating bearing components, such as ball bearing breakage, with high temporal resolution. The second branch uses a one-dimensional dilated convolution with a dilation rate of 4 to extract mid-frequency vibration features caused by abnormal friction in belt conveyor systems. The third branch uses a one-dimensional dilated convolution with a dilation rate of 16 or higher to cover long-period waveforms containing nearly 2000 sampling points with a large receptive field, specifically for extracting low-frequency degradation features caused by macroscopic structural imbalances in key rotating components, such as roller eccentricity, whose period is often 1 to 2 seconds. The feature maps output from the three branches are concatenated and input into a fully connected layer for classification, such as wear on the outer ring of idler bearings and roller imbalance.

[0035] In this embodiment, the output dimension of the dimensionality reduction fully connected layer of the spatial attention mechanism is set to 1, and the output dimension of the dimensionality increase fully connected layer is restored to 3. The three branches of the multi-scale one-dimensional dilated convolution all adopt one-dimensional convolution with a kernel size of 3, and the dilation rate is set to 1, 4 and 16 respectively. Each branch outputs 64 feature maps, which are concatenated to obtain a 192-dimensional feature vector. Finally, the fault category probability is output through two fully connected layers (128-dimensional and 64-dimensional).

[0036] This step utilizes dilated convolutions with different dilation rates, allowing the neural network model to simultaneously possess receptive fields at the millisecond and second levels within the same architecture without exponentially increasing the number of parameters. This accurately balances the extraction of multimodal mechanical fault features, including high-frequency instantaneous impacts and low-frequency slow cycles.

[0037] In this embodiment, the neural network model is trained using a focus loss function instead of the conventional cross-entropy loss function. Since the belt conveyor system operates normally most of the time, fault samples are extremely rare. An adjustment factor is introduced into the focus loss function. In this embodiment Set to 2.0. When the neural network model predicts a high probability of easily categorized normal operating samples, the adjustment factor will exponentially reduce its weight in the training error calculation. This mechanism forcibly blocks the inundation effect of massive normal samples on the model gradient, forcibly focusing the neural network's computing resources on the feature learning of extremely scarce early and minor fault samples, greatly improving the model's ability to generalize and identify early minor wear and tear hazards in industrial settings.

[0038] Specifically, over 8000 hours of field operation data were collected, including labeled samples of normal conditions and typical fault conditions such as wear of the inner / outer rings of the idler bearings, idler jamming, and roller imbalance. The ratio of normal to fault samples was approximately 100:1. The Adam optimizer was used with an initial learning rate of 0.001, trained for 100 epochs, and a batch size of 64. On the validation set, the model achieved an average recognition accuracy of 96.5% for various fault types, and its early warning sensitivity for early, subtle faults was improved by approximately 40% compared to the traditional vibration threshold method.

[0039] This invention also provides a health monitoring system for a belt conveyor system based on distributed fiber optic acoustic sensing, comprising: A sensing optical cable is laid along the belt conveyor system and mechanically coupled to the support of the key rotating component of the belt conveyor system to sense the vibration characteristics of the key rotating component. An optical fiber demodulation device, connected to the sensing optical cable, is used to transmit optical pulses to the sensing optical cable and receive backscattered Rayleigh light signals, and to obtain distributed dynamic strain signals of multiple spatial channels distributed along the sensing optical cable through phase demodulation. The data processing device is communicatively connected to the fiber optic demodulation device, receives the distributed dynamic strain signal, and executes steps S3 to S5 in the health monitoring method for belt conveyor systems based on distributed fiber optic acoustic wave sensing as described above.

[0040] The present invention also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the data analysis step in the health monitoring method for a belt conveyor system based on distributed optical fiber acoustic wave sensing as described above. The data analysis step specifically includes steps S3 to S5.

[0041] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the data analysis step in the health monitoring method for a belt conveyor system based on distributed optical fiber acoustic wave sensing as described above; wherein the data analysis step specifically includes steps S3 to S5.

[0042] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A health monitoring method for a belt conveyor system based on distributed fiber optic acoustic sensing, characterized in that, Includes the following steps: S1. Lay a sensing optical cable along the belt conveyor system, so that the sensing optical cable is mechanically coupled to the support of the key rotating component of the belt conveyor system, so as to sense the vibration characteristics of the key rotating component. S2. Emit light pulses to the sensing optical cable and receive backscattered Rayleigh light signals, and obtain distributed dynamic strain signals of multiple spatial channels distributed along the sensing optical cable through phase demodulation. S3. Obtain the preset energy baseline of vibration intensity, extract the energy characteristic value representing vibration intensity from the distributed dynamic strain signal, and screen out the abnormal spatial channels where the energy characteristic value exceeds the preset energy baseline. S4. Obtain one upstream and one downstream adjacent spatial channel adjacent to the abnormal spatial channel, extract the distributed dynamic strain signal corresponding to the abnormal spatial channel and the adjacent spatial channel, and combine the distributed dynamic strain signal to construct a multi-channel spatiotemporal feature matrix. S5. Input the multi-channel spatiotemporal feature matrix into a pre-trained neural network model, and perform fusion analysis on the signal features of each spatial channel through the neural network model to output the fault type and location information corresponding to the abnormal spatial channel.

2. The method for health monitoring of a belt conveyor system based on distributed fiber optic acoustic sensing according to claim 1, characterized in that, Step S1, which involves maintaining mechanical coupling between the sensing optical cable and the support of the key rotating component of the belt conveyor system, specifically includes the following steps: The sensing optical cable is laid close to the main longitudinal beam of the belt conveyor system and is fixed by rigid metal clamps at the support connection points of each key rotating component to suppress the mechanical attenuation of high-frequency vibration signals.

3. The method for health monitoring of a belt conveyor system based on distributed fiber optic acoustic sensing according to claim 1, characterized in that, Before constructing the multi-channel spatiotemporal feature matrix in step S4, the following steps are also included: A filter is used to remove low-frequency ambient temperature drift interference from the distributed dynamic strain signal, and the filtered signal is then standardized.

4. The method for health monitoring of a belt conveyor system based on distributed optical fiber acoustic sensing according to claim 1, characterized in that, The neural network model includes a spatial attention mechanism module. The specific steps of fusing and analyzing the signal features of each spatial channel using the neural network model include: The spatial attention mechanism module performs global average pooling on the time series dimension of the multi-channel spatiotemporal feature matrix to extract the channel weight vector representing the global features of each spatial channel. Generate the incentive weight vector corresponding to each spatial channel based on the channel weight vector; The excitation weight vector is multiplied element-wise with the multi-channel spatiotemporal feature matrix to amplify the signal weight of the spatial channel where the real vibration source is located and suppress the crosstalk feature weight of the adjacent spatial channels caused by the physical transmission of the support.

5. The method for health monitoring of a belt conveyor system based on distributed optical fiber acoustic sensing according to claim 1, characterized in that, The neural network model also includes a multi-scale one-dimensional dilated convolution module, which specifically includes three parallel feature extraction branches: The first branch uses one-dimensional convolution with a first expansion rate to extract high-frequency transient impact features; wherein, the high-frequency transient impact features are caused by microscopic damage to the bearings of key rotating components; The second branch uses one-dimensional dilated convolution with a second dilation rate to extract mid-frequency vibration features, where the second dilation rate is greater than the first dilation rate; wherein the mid-frequency vibration features are caused by abnormal friction in the belt conveyor system. The third branch uses one-dimensional dilated convolution with a third dilation rate to extract long-period low-frequency degradation features, wherein the third dilation rate is greater than the second dilation rate; wherein the long-period low-frequency degradation features are caused by macroscopic structural imbalance of the rotating component; The feature maps output from the first branch, the second branch, and the third branch are fused and stitched together, and then input into the fully connected layer to classify the fault type.

6. The health monitoring method for a belt conveyor system based on distributed fiber optic acoustic sensing according to claim 1, characterized in that, The training steps for a neural network model include: The error of the neural network model is calculated using a focus loss function. The gradient update weights of normally operating samples during the training process are reduced by adjusting the focus loss function, so that the computing resources of the neural network model are focused on feature learning of minority class samples containing early and weak fault characteristics.

7. A health monitoring system for a belt conveyor system based on distributed fiber optic acoustic sensing, characterized in that, include: A sensing optical cable is laid along the belt conveyor system and mechanically coupled to the support of the key rotating component of the belt conveyor system to sense the vibration characteristics of the key rotating component. An optical fiber demodulation device, connected to the sensing optical cable, is used to transmit optical pulses to the sensing optical cable and receive backscattered Rayleigh light signals, and to obtain distributed dynamic strain signals of multiple spatial channels distributed along the sensing optical cable through phase demodulation. The data processing device is communicatively connected to the optical fiber demodulation device, receives the distributed dynamic strain signal, and executes steps S3 to S5 in the health monitoring method for belt conveyor system based on distributed optical fiber acoustic wave sensing as described in any one of claims 1 to 6.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the data analysis step in the health monitoring method for belt conveyor system based on distributed optical fiber acoustic wave sensing as described in any one of claims 1 to 6; wherein, the data analysis step specifically includes steps S3 to S5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the data analysis step in the health monitoring method for belt conveyor system based on distributed optical fiber acoustic wave sensing as described in any one of claims 1 to 6; wherein, the data analysis step specifically includes steps S3 to S5.

Images