Fault monitoring method and device for carrier roller and electronic equipment

Abstract

The embodiment of the invention provides a fault monitoring method and device for a carrier roller and electronic equipment, and relates to the technical field of distributed optical fiber sensing. The fault monitoring method for the carrier roller comprises the steps that a target audio signal collected by an optical fiber pickup in the motion process of a belt conveyor is obtained, the target audio signal is converted into a first spectrogram, the first spectrogram is input into a pre-trained generative network model, and a second spectrogram is obtained, based on whether the first spectrogram and the second spectrogram are similar or not, whether the target audio signal belongs to a normal audio signal or not is determined, and if the target audio signal does not belong to the normal audio signal, the target carrier roller is determined to be a fault carrier roller. According to the embodiment of the invention, the accuracy of carrier roller fault monitoring can be improved.

Description

Technical Field

[0001] The present application relates to the field of distributed optical fiber sensing technology, and in particular to a fault monitoring method, device and electronic equipment for rollers. Background Art

[0002] With the development of industrialization, belt conveyors are used in more and more scenarios, such as factories, mines, and construction sites. As the belt conveyor works, the rollers in the belt conveyor are usually the first to be worn out. Therefore, timely fault monitoring of the rollers can ensure the normal operation of the belt conveyor.

[0003] The existing fault monitoring method for rollers is generally to manually knock on the rollers or the winding reel in the fiber optic pickup when the belt conveyor stops working, so that the fiber optic pickup collects the audio signal emitted by the knocking, and the fault monitoring equipment analyzes the collected audio signal to monitor the roller fault. However, the frequency and weight of the manual knocking cannot be unified and are unstable, so the accuracy of roller fault monitoring is low.

[0004] How to improve the accuracy of roller fault monitoring is an urgent problem to be solved. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide a method, device and electronic device for monitoring roller faults, so as to improve the accuracy of monitoring roller faults. The specific technical solution is as follows:

[0006] In a first aspect, an embodiment of the present application provides a fault monitoring method for a roller, the method comprising:

[0007] Acquire a target audio signal collected by the optical fiber pickup during the movement of the belt conveyor; wherein the optical fiber pickup is used to pick up the vibration of the target roller of the belt conveyor;

[0008] Converting the target audio signal into a first spectrogram;

[0009] The first spectrogram is input into a pre-trained generative network model to obtain a second spectrogram; wherein the generative network model is used to reconstruct the first spectrogram into a spectrogram obtained by converting a normal audio signal, wherein the normal audio signal is an audio signal that can be collected when the target roller is operating normally;

[0010] determining whether the target audio signal belongs to the normal audio signal based on whether the first spectrogram and the second spectrogram are similar;

[0011] If the target audio signal does not belong to the normal audio signal, it is determined that the target roller is a faulty roller.

[0012] In a second aspect, an embodiment of the present application provides a fault monitoring device for a roller, the device comprising:

[0013] An acquisition module, used to acquire a target audio signal collected by the optical fiber pickup during the movement of the belt conveyor; wherein the optical fiber pickup is used to pick up the vibration of the target roller of the belt conveyor;

[0014] A conversion module, used for converting the target audio signal into a first spectrogram;

[0015] An input module, used for inputting the first spectrogram into a pre-trained generative network model to obtain a second spectrogram; wherein the generative network model is used for reconstructing the first spectrogram into a spectrogram obtained by converting a normal audio signal, wherein the normal audio signal is an audio signal that can be collected when the target roller is operating normally;

[0016] A first determination module, configured to determine whether the target audio signal belongs to the normal audio signal based on whether the first spectrogram and the second spectrogram are similar;

[0017] The second determination module is configured to determine that the target roller is a faulty roller if the target audio signal does not belong to the normal audio signal.

[0018] In a third aspect, an embodiment of the present application provides an electronic device, including:

[0019] Memory, used to store computer programs;

[0020] The processor is used to implement any of the above-mentioned fault monitoring methods for the roller when executing the program stored in the memory.

[0021] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, any of the above-mentioned fault monitoring methods for rollers is implemented.

[0022] Beneficial effects of the embodiments of the present application:

[0023] The fault monitoring method for the idler provided in the embodiment of the present application can obtain the target audio signal collected by the optical fiber pickup during the movement of the belt conveyor, convert the target audio signal into a first spectrum graph, input the first spectrum graph into a pre-trained generation network model, and obtain a second spectrum graph, wherein the generation network model is used to reconstruct the first spectrum graph into a spectrum graph obtained by converting a normal audio signal, and the normal audio signal is an audio signal that can be collected when the target idler is running normally. Based on whether the first spectrum graph and the second spectrum graph are similar, it is determined whether the target audio signal belongs to a normal audio signal. If the target audio signal does not belong to a normal audio signal, it is determined that the target idler is a faulty idler. It can be seen that the embodiment of the present application can determine the similarity relationship between the first spectrum graph converted by the target audio signal and the reconstructed second spectrum graph, so as to determine whether the target audio signal belongs to a normal audio signal, and then monitor whether the target idler is a faulty idler, and the embodiment of the present application does not need to use the audio signal emitted by knocking on the idler or the winding disk in the optical fiber pickup, but directly analyzes the target audio signal emitted by the idler when it is working, so the embodiment of the present application can improve the accuracy of monitoring the idler fault.

[0024] In addition, since the second spectrum diagram is a spectrum diagram obtained by converting a normal audio signal, if the target audio signal is a normal audio signal, the second spectrum diagram is similar to the first spectrum diagram, and if the target audio signal is not a normal audio signal, the second spectrum diagram is not similar to the first spectrum diagram. This makes it unnecessary for the embodiment of the present application to pay attention to whether the collected target audio signal is an audio signal that can be collected when the target roller is operating normally. The collected target audio signal can be effectively analyzed to monitor the fault of the target roller. Therefore, the embodiment of the present application can improve the versatility of roller fault monitoring. In addition, compared with the related art, the embodiment of the present application does not need to knock on the roller or the winding reel in the fiber optic pickup, so it will not cause secondary damage to the roller, and when determining the abnormal roller, it is not only compared through a single amplitude value, so the embodiment of the present application can improve the accuracy of roller fault monitoring.

[0025] Of course, implementing any product or method of the present application does not necessarily require achieving all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application, and for ordinary technicians in this field, other embodiments can also be obtained based on these drawings.

[0027] Figure 1A structural diagram of a DAS system provided in an embodiment of the present application;

[0028] Figure 2 A schematic flow chart of a fault monitoring method for a roller provided in an embodiment of the present application;

[0029] Figure 3 A flowchart of a training method for generating a network model provided in an embodiment of the present application;

[0030] FIG4( a) is a schematic diagram showing the effect of a reconstructed spectrum graph obtained after a spectrum graph converted from a normal audio signal provided in an embodiment of the present application is input into a generative network model;

[0031] FIG4( b ) is a schematic diagram showing the effect of a reconstructed spectrum graph obtained after the spectrum graph converted from the abnormal audio signal provided in an embodiment of the present application is input into a generation network model;

[0032] Figure 5 A schematic diagram of the principle of a fault monitoring method for a roller provided in an embodiment of the present application;

[0033] Figure 6 A schematic diagram of a process for performing denoising processing on a target audio signal provided in an embodiment of the present application;

[0034] FIG. 7( a ) is a high-frequency waterfall diagram provided in an embodiment of the present application;

[0035] FIG7( b ) is an amplitude diagram for an external optical fiber provided in an embodiment of the present application;

[0036] FIG. 7( c ) is an amplitude diagram for a fiber optic pickup provided in an embodiment of the present application;

[0037] FIG7( d ) is an amplitude diagram along an optical fiber provided in an embodiment of the present application;

[0038] FIG8( a ) is a schematic diagram of the structure of a fiber optic pickup provided in an embodiment of the present application;

[0039] FIG8( b ) is a schematic diagram of an application scenario of a fiber optic pickup provided in an embodiment of the present application;

[0040] Fig. 9 A schematic structural diagram of a fault monitoring device for a roller provided in an embodiment of the present application;

[0041] Fig.10 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field based on the present application belong to the scope of protection of the present application.

[0043] Below, the professional terms involved in the embodiments of the present application are first introduced:

[0044] Rollers: important parts of belt conveyors, with many types and large quantities, which can support the weight of conveyor belts and materials. It accounts for 35% of the total cost of a belt conveyor and produces more than 70% of the resistance, so the quality of rollers is particularly important.

[0045] DAS system: Distributed Optical Fiber Audio Sensing System, a distributed optical fiber acoustic wave sensing system, is an optical instrument that uses optical fiber as a sensor to sense acoustic wave vibrations. The system uses a single optical fiber to simultaneously monitor acoustic wave vibrations and transmit signals.

[0046] Fiber optic pickup: It can also be called a monitoring head, auscultation probe or fiber optic sensor, etc. It is a device used to collect environmental sounds on site and transmit them to the back-end equipment.

[0047] Furthermore, the present application embodiment can be based on the DAS system when performing fault monitoring for the roller. Therefore, before introducing the method provided by the present application embodiment, the DAS system will be briefly introduced in conjunction with the accompanying drawings. Figure 1 As shown:

[0048] The DAS system comprises: a seed laser (1), a beam splitter / isolator (2), a radio frequency driver (3), an acousto-optic modulator (4), a laser amplification module (5), an optical fiber circulator (6), a photodetector (7), a data acquisition module (8), a signal processing module (9) and a pickup group (10).

[0049] After the seed laser (1) is connected to the beam splitter / isolator (2), the output laser light is divided into two paths of light with a specific ratio for output, the first path is connected to the input end of the acousto-optic modulator (4), and the second path is connected to the photodetector (7) as the first path input;

[0050] The data acquisition module (8) outputs the radio frequency signal to the radio frequency driver (3), and the radio frequency driver (3) outputs the modulated signal to the acousto-optic modulator (4);

[0051] The output of the acousto-optic modulator (4) is connected to one port of a fiber optic circulator (6) through a laser amplification module (5); the output of the second port of the fiber optic circulator (6) is connected to a pickup group (10); the third port of the fiber optic circulator (6) receives the return light signal as a second input of the photoelectric detector (7); wherein the pickup group (10) includes a plurality of fiber optic pickups.

[0052] The first optical signal of the photoelectric detector (7) interferes with the second input return light signal, and the internal detector converts the optical signal into an electrical signal and outputs it to the data acquisition module (8) and the signal processing module (9).

[0053] It can be understood that the signal processing module (9) in the above-mentioned DAS system can restore the converted electrical signal to an audio signal, and the audio signal is the audio signal collected by the pickup group (10). Therefore, the above-mentioned DAS system can transmit the audio signal collected by the optical fiber pickup to the back-end device, so that the back-end device can perform specific processing based on the audio signal collected by the optical fiber pickup, for example: fault monitoring and analysis of the roller in this embodiment.

[0054] The following first introduces a fault monitoring method for rollers provided in an embodiment of the present application.

[0055] The fault monitoring method for the roller provided in the embodiment of the present application can be applied to an electronic device. In a specific application, the electronic device can be a terminal device or a server, and the embodiment of the present application does not limit the specific form of the electronic device.

[0056] Specifically, the execution subject of the roller fault monitoring method may be a roller fault monitoring device. Exemplarily, when the roller fault monitoring method is applied to a terminal device, the roller fault monitoring device may be a client running in the terminal device for roller fault monitoring. Exemplarily, when the roller fault monitoring method is applied to a server, the roller fault monitoring device may be a computer program running in the server, and the computer program may be used for roller fault monitoring.

[0057] Among them, an embodiment of the present application provides a fault monitoring method for a roller, the method comprising:

[0058] Acquire a target audio signal collected by the optical fiber pickup during the movement of the belt conveyor; wherein the optical fiber pickup is used to pick up the vibration of the target roller of the belt conveyor;

[0059] Converting the target audio signal into a first spectrogram;

[0060] The first spectrogram is input into a pre-trained generative network model to obtain a second spectrogram; wherein the generative network model is used to reconstruct the first spectrogram into a spectrogram obtained by converting a normal audio signal, wherein the normal audio signal is an audio signal that can be collected when the target roller is operating normally;

[0061] determining whether the target audio signal belongs to the normal audio signal based on whether the first spectrogram and the second spectrogram are similar;

[0062] If the target audio signal does not belong to the normal audio signal, it is determined that the target roller is a faulty roller.

[0063] It can be seen that the embodiment of the present application can determine the similarity relationship between the first spectrum graph converted from the target audio signal and the reconstructed second spectrum graph, thereby determining whether the target audio signal belongs to a normal audio signal, and then monitoring whether the target roller is a faulty roller. In addition, the embodiment of the present application does not need to utilize the audio signal emitted by knocking on the roller or the winding reel in the optical fiber pickup, but directly analyzes the target audio signal emitted by the roller when it is working. Therefore, the embodiment of the present application can improve the accuracy of roller fault monitoring.

[0064] In addition, since the second spectrum diagram is a spectrum diagram obtained by converting a normal audio signal, if the target audio signal is a normal audio signal, the second spectrum diagram is similar to the first spectrum diagram, and if the target audio signal is not a normal audio signal, the second spectrum diagram is not similar to the first spectrum diagram. This makes it unnecessary for the embodiment of the present application to pay attention to whether the collected target audio signal is an audio signal that can be collected when the target roller is operating normally. The collected target audio signal can be effectively analyzed to monitor the fault of the target roller. Therefore, the embodiment of the present application can improve the versatility of roller fault monitoring. In addition, compared with the related art, the embodiment of the present application does not need to knock on the roller or the winding reel in the fiber optic pickup, so it will not cause secondary damage to the roller, and when determining the abnormal roller, it is not only compared through a single amplitude value, so the embodiment of the present application can improve the accuracy of roller fault monitoring.

[0065] A fault monitoring method for a roller provided in an embodiment of the present application is introduced below in conjunction with the accompanying drawings.

[0066] like Figure 2 As shown, an embodiment of the present application provides a fault monitoring method for a roller, and the method may include the following steps:

[0067] S201, obtaining a target audio signal collected by the optical fiber pickup during the movement of the belt conveyor;

[0068] Wherein, the optical fiber pickup is used to pick up the vibration of the target roller of the belt conveyor;

[0069] It is understandable that during the movement of the belt conveyor, the roller of the belt conveyor will vibrate and make a sound. The sound made by the vibration of the target roller can be picked up by the fiber optic pickup, that is, audio signal collection; wherein, the fiber optic pickup can be set in the vibration sensing area of ​​the target roller of the belt conveyor, such as: on the target roller, so as to collect the target audio signal by picking up the sound, or it can be set at a position close to the target roller of the belt conveyor. It should be emphasized that the number of fiber optic pickups can be multiple during use, so each fiber optic pickup can collect the target audio signal during the operation of the belt conveyor. In order to implement this solution, the target audio signal collected by each fiber optic pickup can be identified. For the sake of clear layout, the method for determining the target audio signal collected by the fiber optic pickup will be introduced in subsequent embodiments and will not be repeated here.

[0070] It should be emphasized that a single fiber optic pickup can be used to collect target audio signals emitted by the vibration of multiple target rollers, or a single fiber optic pickup can be used only to collect target audio signals emitted by the vibration of a single target roller. Of course, multiple fiber optic pickups can also be used to collect target audio signals emitted by the vibration of a single target roller, which is not specifically limited in the embodiments of the present application. For example, fiber optic pickup a is used to collect target audio signals emitted by the vibration of rollers 1, 2, and 3 during the movement of the belt conveyor.

[0071] In addition, since the fiber optic pickup can be set in the vibration sensing area of ​​the target roller of the belt conveyor, or can be set near the target roller of the belt conveyor, then, when a single fiber optic pickup is used to collect multiple target rollers, the fiber optic pickup can be set in the vibration sensing area that can collect the target audio signals emitted by the vibration of multiple target rollers. For example, the collection range of fiber optic pickup a is 6m, that is, it can collect audio signals within 3m on the left and audio signals within 3m on the right. Rollers 1, 2 and 3 are three adjacent rollers, and each roller is 2m apart. Among them, the vibration sensing areas of rollers 1, 2 and 3 have an overlapping area, which is located near roller 2. Then, fiber optic pickup a can be set in the overlapping area. In this way, fiber optic pickup a can collect the target audio signals emitted by the vibration of rollers 1, 2 and 3.

[0072] S202, converting the target audio signal into a first spectrogram;

[0073] It is understandable that when the target audio signal is multi-segment, each target audio signal can be converted into a first spectrogram. It should be emphasized that the method of converting the target audio signal into the first spectrogram can be short time Fourier transform (STFT). Of course, other methods can also be used to convert the target audio signal into the first spectrogram, and the embodiment of the present application does not specifically limit this.

[0074] S203, inputting the first spectrogram into a pre-trained generative network model to obtain a second spectrogram;

[0075] The generative network model is used to reconstruct the first spectrum graph into a spectrum graph obtained by converting a normal audio signal, wherein the normal audio signal is an audio signal that can be collected when the target roller operates normally;

[0076] It can be understood that after the first spectrum graph is input into the pre-trained generative network model, the generative network model can use the autoencoder contained in the generative network model to encode the first spectrum graph to obtain an intermediate spectrum graph, and then use the autodecoder contained in the generative network model to decode the intermediate spectrum graph to generate a second spectrum graph, which can also be called a reconstructed spectrum graph.

[0077] It can be understood that the structure of the generated network model includes but is not limited to: autoencoder (AutoEncoder, AE), autodecoder (AutoDecoder, AD), variational autoencoder (Variational AutoenEonder, VAE), diffusion model (Diffusion model), etc.; wherein, the generated network model is used to make the reconstructed second spectrum graph a spectrum graph obtained by converting a normal audio signal, and the variational autoencoder and diffusion model can be used to generate images, and the autoencoder and autodecoder structures can be directly replaced to achieve the same image generation effect. The embodiment of the present application does not specifically limit the structure of the generated network model.

[0078] In addition, the audio signal that can be collected when the target roller is running abnormally can be called an abnormal audio signal, so the normal audio signal and the abnormal audio signal are both audio signals collected for the running condition of the target roller. It can be understood that the first spectrum graph can be a spectrum graph converted from a normal audio signal, or a spectrum graph converted from an abnormal audio signal, but after the above two first spectrum graphs are input into the generation network model, the obtained second spectrum graphs are both audio signals that can be collected when the target roller is running normally. Exemplarily, the first spectrum graph A is a spectrum graph converted from an audio signal that can be collected when the roller 1 is operating normally. The first spectrum graph A is input into the generative network model to obtain the second spectrum graph B. The second spectrum graph B is a spectrum graph obtained by reconstructing the first spectrum graph A into a spectrum graph obtained by converting a normal audio signal. Since the first spectrum graph A is converted from a normal audio signal, the first spectrum graph A is similar to the second spectrum graph B. The first spectrum graph C is a spectrum graph converted from an audio signal that can be collected when the roller 2 is operating abnormally. The first spectrum graph C is input into the generative network model to obtain the second spectrum graph D. The second spectrum graph D is a spectrum graph obtained by reconstructing the first spectrum graph C into a spectrum graph obtained by converting a normal audio signal. Since the first spectrum graph C is converted from an abnormal audio signal, the first spectrum graph C and the second spectrum graph D are very different.

[0079] For a clearer understanding, the effect of the reconstructed spectrum graph obtained by inputting the spectrum graph converted by the normal audio signal and the abnormal audio signal into the generation network model will be introduced in conjunction with Figure 4(a) and Figure 4(b), as shown in Figure 4(a) and Figure 4(b):

[0080] Figure 4(a) is an effect diagram of the reconstructed spectrum graph obtained after the spectrum graph converted from a normal audio signal is input into the generative network model. It can be understood that when the input spectrum graph is a spectrum graph converted from a normal audio signal, the input spectrum graph and the reconstructed spectrum graph have a high similarity.

[0081] FIG4(b) is a diagram showing the effect of the reconstructed spectrum graph obtained after the spectrum graph converted from the abnormal audio signal is input into the generative network model. It can be understood that when the input spectrum graph is a spectrum graph converted from the abnormal audio signal, there are vertical lines on the input spectrum graph, while the reconstructed spectrum graph has no vertical lines, so the similarity between the two is relatively low.

[0082] In addition, the training method for generating the network model will be introduced in other embodiments and will not be described in detail here.

[0083] S204, determining whether the target audio signal belongs to the normal audio signal based on whether the first spectrogram and the second spectrogram are similar;

[0084] It can be understood that based on the first spectrogram and the second spectrogram reconstructed by the generative network model, it can be determined whether the two are similar. If the first spectrogram is a spectrogram converted from an audio signal that can be collected when the target roller is running normally, and the second spectrogram is a spectrogram obtained by reconstructing the first spectrogram into a spectrogram obtained by converting a normal audio signal, then the first spectrogram and the second spectrogram are similar, and it can be determined that the target audio signal belongs to a normal audio signal; similarly, if the first spectrogram is a spectrogram converted from an audio signal that can be collected when the target roller is running abnormally, and the second spectrogram is a spectrogram obtained by reconstructing the first spectrogram into a spectrogram obtained by converting a normal audio signal, then the first spectrogram and the second spectrogram are dissimilar, and it can be determined that the target audio signal does not belong to a normal audio signal. Exemplarily, based on the similarity between the first spectrogram A and the second spectrogram B, it can be determined that the target audio signal a corresponding to the first spectrogram A belongs to a normal audio signal; based on the dissimilarity between the first spectrogram C and the second spectrogram D, it can be determined that the target audio signal c corresponding to the first spectrogram C does not belong to a normal audio signal.

[0085] Optionally, in an implementation manner, determining whether the target audio signal belongs to the normal audio signal based on whether the first spectrogram and the second spectrogram are similar comprises steps A1-A2:

[0086] Step A1, inputting the first spectrogram and the second spectrogram into a pre-trained discriminator network model to obtain an output result; wherein the discriminator network model is used to analyze whether the first spectrogram and the second spectrogram are similar;

[0087] It is understandable that the implementation method for determining whether the target audio signal is a normal audio signal can be to input the first spectrogram and the second spectrogram into a pre-trained discriminator network model to obtain an output result. It should be emphasized that the output result can be the similarity itself, for example: the similarity is 10%, or the similarity is 96%, etc., or it can also represent the result of similarity, for example: the output result is similar or dissimilar, etc.

[0088] It should be noted that the discriminator network model can use vector distance to discriminate the similarity between the first spectrum image and the second spectrum image. The calculation method includes but is not limited to: cosine distance, Euclidean distance, Manhattan distance, etc. The embodiment of the present application does not specifically limit the various calculation methods.

[0089] The following uses Euclidean distance as a calculation method to determine the similarity of vectors:

[0090] Exemplarily, the features extracted by the discriminator network model are n-dimensional row vectors, so assuming that the first spectrum Figure X The vector A extracted by the discriminator network model is [a1, a2, …, an ], the second spectrum Figure X hat The vector B extracted by the discriminator network model is [b1,b2,…,b n ], the cosine distance between vector A and vector B is calculated as follows:

[0091]

[0092] It can be understood that the cosine distance value calculated based on the vector A and the vector B can be used to determine the similarity between the vectors, that is, the first spectrum Figure X With the second spectrum Figure X hat The similarity between .

[0093] Step A2: determining whether the target audio signal belongs to the normal audio signal based on the output result.

[0094] It can be understood that when the output result is similarity, it is possible to determine whether the target audio signal belongs to a normal audio signal based on the audio threshold as a dividing line. For example, the similarity between the first spectrum graph A and the second spectrum graph B is 60%, and the audio threshold is 80%. The similarity above the audio threshold indicates that the corresponding target audio signal belongs to a normal audio signal. Since the similarity between the first spectrum graph A and the second spectrum graph B is lower than the audio threshold, it is considered that the corresponding target audio signal a does not belong to a normal audio signal. When the output result is a result content representing whether the first spectrum graph and the second spectrum graph are similar, it is possible to directly determine whether the target audio signal belongs to a normal audio signal. For example, when the output result represents that the first spectrum graph A and the second spectrum graph B are similar, it is possible to determine that the target audio signal a belongs to a normal audio signal. When the output result represents that the first spectrum graph A and the second spectrum graph B are similar, it is possible to determine that the target audio signal a belongs to a normal audio signal. When the output result represents that the first spectrum graph C and the second spectrum graph D are dissimilar, it is possible to determine that the target audio signal c does not belong to a normal audio signal.

[0095] It can be understood that in one implementation, with the help of the sample spectrogram used to train the generative network model, the sample spectrogram is input into the generative network model to obtain a reconstructed spectrogram, so that the similarity between the sample spectrogram and the reconstructed spectrogram can be calculated, and the similarities are sorted to obtain the audio threshold.

[0096] Exemplarily, during the training process, multiple sample spectrograms [X_1, X_2, ..., X_n] converted from normal audio signals for testing can be sent to the generative network model to obtain multiple reconstructed spectrograms [X_hat1, X_hat2, ..., X_hatn], and then discriminated by the discriminator network model to obtain a set of cosine distance values ​​[dist_1, dist_2, ..., dist_n]. The cosine distance values ​​are sorted from small to large and the 95% quantile is taken. 95% can be understood as 100 similarities, and the similarity ranked at 95. Of course, other quantiles can also be taken, and 95% is only used as an empirical value. Then, the cosine distance value of dist_(95%) can be used as the classification boundary between normal audio signals and abnormal audio signals, that is, the audio threshold.

[0097] It can be seen that the embodiment of the present application inputs the first spectrum graph and the second spectrum graph into the pre-trained discriminator network model to obtain an output result, and based on the output result, determines whether the target audio signal is a normal audio signal. Therefore, the embodiment of the present application determines whether the target audio signal is a normal audio signal based on the first spectrum graph and the reconstructed second spectrum graph, and then monitors the target roller. Therefore, the embodiment of the present application can improve the accuracy of roller fault monitoring.

[0098] S205: If the target audio signal does not belong to the normal audio signal, determine that the target roller is a faulty roller.

[0099] It can be understood that when the target audio signal belongs to a normal audio signal, it can be determined that the target roller is not a faulty roller. Therefore, whether the target audio signal belongs to a normal audio signal can be used as a basis for judging whether the target roller is a faulty roller. Exemplarily, when the target audio signal a does not belong to a normal audio signal, it can be determined that the target roller 1 is a faulty roller.

[0100] It can be seen that the embodiment of the present application can determine the similarity relationship between the first spectrum graph converted from the target audio signal and the reconstructed second spectrum graph, thereby determining whether the target audio signal belongs to a normal audio signal, and then monitoring whether the target roller is a faulty roller. In addition, the embodiment of the present application does not need to utilize the audio signal emitted by knocking on the roller or the winding reel in the optical fiber pickup, but directly analyzes the target audio signal emitted by the roller when it is working. Therefore, the embodiment of the present application can improve the accuracy of roller fault monitoring.

[0101] In addition, since the second spectrum diagram is a spectrum diagram obtained by converting a normal audio signal, if the target audio signal is a normal audio signal, the second spectrum diagram is similar to the first spectrum diagram, and if the target audio signal is not a normal audio signal, the second spectrum diagram is not similar to the first spectrum diagram. This makes it unnecessary for the embodiment of the present application to pay attention to whether the collected target audio signal is an audio signal that can be collected when the target roller is operating normally. The collected target audio signal can be effectively analyzed to monitor the fault of the target roller. Therefore, the embodiment of the present application can improve the versatility of roller fault monitoring. In addition, compared with the related art, the embodiment of the present application does not need to knock on the roller or the winding reel in the fiber optic pickup, so it will not cause secondary damage to the roller, and when determining the abnormal roller, it is not only compared through a single amplitude value, so the embodiment of the present application can improve the accuracy of roller fault monitoring.

[0102] In order to better understand the above method of using the generative network model and the discriminator network model to monitor the fault of the target roller, the following will be briefly introduced in conjunction with the accompanying drawings. Figure 5 As shown:

[0103] S501, obtaining a target audio signal;

[0104] The target audio signal is the audio signal collected by the optical fiber pickup during the movement of the belt conveyor.

[0105] S502, performing short-time Fourier transform on the target audio signal to obtain a first spectrum diagram;

[0106] It can be understood that short-time Fourier transform (STFT) is a type of Fourier transform, and the target audio signal can be converted into a first spectrogram through short-time Fourier transform.

[0107] S503, inputting the first spectrum graph into a generation network model;

[0108] It can be understood that the generation network model includes an encoder and a decoder. The encoder can encode the high-dimensional variable in the first spectrum graph into a low-dimensional variable h, and the decoder can decode the low-dimensional variable h into the second spectrum graph.

[0109] S504, outputting a second spectrum diagram;

[0110] It can be understood that the second spectrogram is a spectrogram reconstructed from the first spectrogram through conversion of a normal audio signal.

[0111] It should be emphasized that by inputting both the first spectrum graph and the second spectrum graph into the discriminator network model, the discriminator network model can analyze whether the first spectrum graph and the second spectrum graph are similar, and obtain an output result, which represents whether the target audio signal is a normal audio signal.

[0112] S505, determining whether the target roller is a faulty roller based on the output result of the discriminator network model;

[0113] It can be understood that, when the output result indicates that the target audio signal is a normal audio signal, it indicates that the target roller is not a faulty roller; when the output result indicates that the target audio signal is not a normal audio signal, it indicates that the target roller is a faulty roller.

[0114] It can be seen that the embodiment of the present application can determine the similarity relationship between the first spectrum graph converted from the target audio signal and the reconstructed second spectrum graph, thereby determining whether the target audio signal belongs to a normal audio signal, and then monitoring whether the target roller is a faulty roller. In addition, the embodiment of the present application does not need to utilize the audio signal emitted by knocking on the roller or the winding reel in the optical fiber pickup, but directly analyzes the target audio signal emitted by the roller when it is working. Therefore, the embodiment of the present application can improve the accuracy of roller fault monitoring.

[0115] In addition, since the second spectrum graph is a spectrum graph obtained by converting a normal audio signal, if the target audio signal is a normal audio signal, the second spectrum graph is similar to the first spectrum graph; if the target audio signal is not a normal audio signal, the second spectrum graph is not similar to the first spectrum graph. This makes it unnecessary for the embodiment of the present application to pay attention to whether the collected target audio signal is an audio signal that can be collected when the target roller is operating normally. The collected target audio signal can be effectively analyzed to monitor the fault of the target roller. Therefore, the embodiment of the present application can improve the versatility of roller fault monitoring.

[0116] Optionally, in another embodiment of the present application, before converting the target audio signal into a first spectrogram, the method further includes steps B1-B3:

[0117] Step B1, performing Fourier transform on the target audio signal to obtain a third spectrum graph;

[0118] It is understandable that the target audio signal may be subjected to a Fourier transform using a short-time Fourier transform method, thereby obtaining the third spectrum graph. Exemplarily, the target audio signal I_1 is subjected to a short-time Fourier transform to generate the third spectrum graph F_1.

[0119] Step B2, using a mask corresponding to a target noise signal, filtering the third spectrum graph to obtain a fourth spectrum graph; wherein the target noise signal is an audio signal generated and collected when the belt conveyor is in a silent state;

[0120] It is understandable that during the acquisition of the target audio signal, there is a corresponding target noise signal, which can also be called system noise I_noise. The target noise signal still exists when the belt conveyor is in a silent state, which will affect the subsequent audio signal analysis. Therefore, the target audio signal can be subjected to noise reduction processing.

[0121] It is understandable that by filtering using the mask corresponding to the target noise signal, the target noise signal in the third spectrum graph can be filtered, so that the fourth spectrum graph obtained is a clean, noise-free spectrum graph. Exemplarily, F_1 ​​is filtered according to mask to obtain F_1'.

[0122] In addition, the mask can be understood as using a selected image, graphic or object to block the processed image (all or part) to control the image processing area or processing process. Its main purpose is to extract the area of ​​interest, that is, a clean, noise-free spectrum map.

[0123] Optionally, in one implementation, a method for generating a mask corresponding to the target noise signal may include methods B21-B22:

[0124] Mode B21, performing Fourier transform on the target noise signal to generate a fifth spectrum diagram;

[0125] It is understandable that the fifth spectrum graph corresponding to the target noise signal can be obtained by performing a short-time Fourier transform on the target noise signal generated when the belt conveyor is in a silent state. Exemplarily, the target noise signal I_0 is short-time Fourier transformed to generate the fifth spectrum graph F_0.

[0126] Mode B22: determining a maximum value in the fifth spectrum graph as a segmentation threshold, and using the segmentation threshold to generate a mask corresponding to the target noise signal.

[0127] It can be understood that the maximum value in the fifth spectrum graph is counted as the subsequent segmentation threshold thred, and a mask mask corresponding to the target noise signal is generated according to the segmentation threshold. There are many ways to generate masks using the segmentation threshold, such as: using python to generate masks, which is not specifically limited in the embodiments of the present application.

[0128] Step B3: Perform inverse Fourier transform on the fourth spectrum graph to obtain a denoised target audio signal.

[0129] It can be understood that the clean, noise-free fourth spectrogram is subjected to an inverse Fourier transform to obtain a denoised target audio signal; wherein the Fourier transform is relative to the inverse Fourier transform, the Fourier transform can convert an audio signal into a spectrogram, and the inverse Fourier transform can convert a spectrogram into an audio signal. Exemplarily, the fourth spectrogram F_1' is subjected to an inverse short-time Fourier transform to obtain a denoised target audio signal I_signal.

[0130] In order to better understand the above content about filtering, the following will be combined with the accompanying drawings to briefly introduce it. Figure 6 As shown:

[0131] S601, collecting target noise signal I_0;

[0132] S602, performing short-time Fourier transform on the target noise signal to obtain a fifth spectrum diagram;

[0133] S603, setting a segmentation threshold thred, and generating a mask mask corresponding to the target noise signal;

[0134] It can be understood that the maximum value in the fifth spectrum graph can be used as the segmentation threshold thred.

[0135] S604, collecting target audio signal I_1;

[0136] S605, performing short-time Fourier transform on the target audio signal to obtain a third spectrum graph;

[0137] S606, filtering the third spectrum map using the mask to obtain a fourth spectrum map;

[0138] S607 , performing short-time inverse Fourier transform on the fourth spectrum graph to obtain a denoised target audio signal I_signal.

[0139] It is understandable that the above content has been introduced in the above embodiments and will not be repeated here.

[0140] It can be seen that the embodiment of the present application proposes a denoising method for audio signals, which filters and reduces noise on audio signals in the frequency domain, thereby effectively improving the signal-to-noise ratio and making the target audio signal clean and noise-free. The subsequent analysis result of whether the target audio signal is a normal audio signal is more accurate, thereby improving the accuracy of roller fault monitoring.

[0141] Optionally, in another embodiment of the present application, the method for determining the target audio signal collected by the optical fiber pickup during the movement of the belt conveyor may include steps C1-C2:

[0142] Step C1, obtaining an electrical signal converted from an optical signal transmitted by a target optical fiber; wherein the target optical fiber is an optical fiber to which the optical fiber pickup belongs, and the target optical fiber is used to transmit a modulated pulse optical signal sent by a designated laser during the movement of the belt conveyor;

[0143] It is understandable that the number of fiber optic pickups used for sound pickup can be multiple groups, then the optical fibers to which the multiple groups of fiber optic pickups belong can be used as target optical fibers. It should be emphasized that when the fiber optic pickup collects the target audio signal, the target audio signal can be converted into an optical signal, so that the modulated pulse optical signal sent by the designated laser transmitted in the target optical fiber can carry the optical signal converted from the target audio signal and transmit it to the DAS system, and the DAS system can obtain the optical signal transmitted by the target optical fiber and convert it into an electrical signal. In addition, the optical signal transmitted in the target optical fiber is a modulated pulse optical signal emitted by the designated laser in the DAS system, and the designated laser can also be called a seed laser.

[0144] It should be emphasized that the ambient sound waves in the environment where the fiber optic pickup is located can enter the internal sound cavity of the fiber optic pickup through the sound hole, and the ambient sound waves entering the internal sound cavity of the fiber optic pickup can cause the sound-sensitive winding shaft, that is, the sound-sensitive material in the sound-sensitive winding shaft or the sound-sensitive material attached to the shaft wall surface of the sound-sensitive winding shaft, to undergo local deformation.

[0145] Step C2, based on the pre-calibrated optical fiber position corresponding to the optical fiber pickup, restore the audio signal of the electrical signal to obtain the target audio signal collected by the optical fiber pickup during the movement of the belt conveyor;

[0146] The optical fiber position corresponding to the optical fiber pickup is used to characterize the position point of the optical fiber pickup in the target optical fiber.

[0147] It is understandable that the fiber position corresponding to the fiber pickup obtained by pre-calibration can represent the specific length of the fiber pickup on the target fiber, for example: the length of fiber pickup 1 on the target fiber is 10m, and the length of fiber pickup 2 on the target fiber is 20m. It should be emphasized that based on the fiber position corresponding to the fiber pickup, the electrical signal can be restored to obtain the target audio signal, and the target audio signal collected by the fiber pickup during the movement of the belt conveyor can be determined. In actual situations, the DAS system is generally used to restore the electrical signal to determine the target audio signal collected by the fiber pickup during the movement of the belt conveyor; when the number of target audio signals and fiber pickups is multiple groups, the target audio signal collected by each fiber pickup can be determined.

[0148] In addition, after the fiber optic pickups are fixed on the roller fixing steel frame in turn, the target audio signal can be collected in real time. Under ideal circumstances, there is a mapping relationship between the fiber optic position corresponding to the fiber optic pickup and the roller number. However, errors occur during the actual implementation process, and the errors will gradually increase with the increase in the number of monitored rollers. Therefore, the fiber optic position corresponding to the fiber optic pickup can be pre-calibrated. The method for calibrating the fiber optic position corresponding to the fiber optic pickup will be introduced in subsequent embodiments and will not be elaborated here.

[0149] It can be seen that based on the DAS system, the target audio signal collected by the fiber optic pickup during the movement of the belt conveyor can be determined, so as to facilitate the analysis of the target audio signal in the subsequent steps. Moreover, the accuracy of roller fault monitoring can be improved by matching the fiber optic pickup with the target audio signal.

[0150] Based on the above embodiment, the calibration method of the optical fiber position corresponding to the optical fiber pickup includes: one of the first calibration method D1 or the second calibration method D2;

[0151] The first calibration method D1 includes: obtaining a first audio signal, converting the first audio signal into a target spectrum diagram, and using the target spectrum diagram to determine the optical fiber position corresponding to the optical fiber pickup; wherein the first audio signal is: an audio signal along the optical fiber obtained when an audio playback device set at a specified position plays a high-frequency signal; the specified position is a position in the pickup area corresponding to the optical fiber pickup;

[0152] It can be understood that the first audio signal is an audio signal for the entire optical fiber. The first calibration method can be performed by manually playing a high-frequency signal at a specified position using an audio player. Then, when the first audio signal is obtained in real time, the first audio signal can be converted into a target spectrum diagram, so that the target spectrum diagram can be directly used to determine the optical fiber position corresponding to the optical fiber pickup. Exemplarily, the staff uses a player to play a 5000HZ high-frequency signal at the pickup area of ​​the optical fiber pickup 1. Then, when calibrating, the first audio signal can be obtained, the first audio signal can be converted into a target spectrum diagram, and the optical fiber position corresponding to the point near 5000HZ on the target spectrum diagram is found to be 10m, thereby determining that the optical fiber position corresponding to the optical fiber pickup 1 is 10m.

[0153] In order to better understand the content of determining the optical fiber position corresponding to the optical fiber pickup in the target spectrum diagram, the following will be briefly introduced in conjunction with the accompanying drawings, as shown in Figure 7(a):

[0154] FIG7(a) shows the target spectrum diagram corresponding to the high-frequency signal playing 5000 Hz. The target spectrum diagram can also be called a high-frequency waterfall diagram. The target spectrum diagram in the figure has a horizontal axis and a vertical axis, wherein the horizontal axis is the unit, the unit is m, 1 unit can be 1m, or other lengths, and the specific length corresponding to the unit can be specifically set; the vertical axis is the frequency, the unit is HZ, and two points can be determined in the target spectrum diagram. The point close to the bottom has a horizontal coordinate x=81 and a vertical coordinate y=2906.25. Its vertical coordinate does not conform to the high-frequency signal of 5000 Hz, and the circled point above is closer to 5000 Hz on the vertical coordinate. Therefore, the horizontal coordinate of this point can be used as the optical fiber position corresponding to the optical fiber pickup. The horizontal coordinate of this point is approximately 87, so it can be obtained that the optical fiber position corresponding to the optical fiber pickup is 87m.

[0155] The second calibration method D2 includes: obtaining a second audio signal along the optical fiber of the target optical fiber within a predetermined time period, and generating a vibration amplitude diagram of the obtained second audio signal, performing amplitude averaging processing on the vibration amplitude diagram of the second audio signal to obtain a target amplitude diagram, counting the amplitudes of the target amplitude diagram at the same time within the predetermined time period to obtain an amplitude diagram along the optical fiber, and determining the optical fiber position corresponding to the optical fiber pickup based on the optical fiber position corresponding to the minimum value in the amplitude diagram along the optical fiber.

[0156] It can be understood that the optical fiber along the target optical fiber is composed of a fiber optic pickup and an external optical fiber, and the external optical fiber is the optical fiber connecting the various fiber optic pickups. Then, when obtaining the second audio signal along the optical fiber along the target optical fiber within a predetermined time period, the audio signals of the fiber optic pickup at various positions along the optical fiber within the predetermined time period and the audio signals of the external optical fiber can be obtained. In this way, a vibration amplitude diagram of the obtained second audio signal can be generated.

[0157] Then, after obtaining the vibration amplitude graph of the acquired second audio signal, the vibration amplitude graph of the second audio signal can be subjected to amplitude averaging processing to obtain a target amplitude graph; wherein the target amplitude graph includes the amplitude of the fiber optic pickup and the amplitude of the external optical fiber.

[0158] It can be understood that based on the target amplitude graph, the amplitude of the target amplitude graph at the same time in a predetermined time period can be counted to obtain an amplitude graph along the optical fiber. In the amplitude graph along the optical fiber, there is the amplitude of the external optical fiber, and there is also the amplitude of the optical fiber pickup. Since the amplitude at the external optical fiber is much larger than the amplitude at the optical fiber pickup, the minimum value in the amplitude graph along the optical fiber is the optical fiber position of the optical fiber pickup.

[0159] For ease of understanding, the amplitude diagrams along the optical fiber are introduced in conjunction with Figure 7(b), Figure 7(c) and Figure 7(d), as shown in Figure 7(b), Figure 7(c) and Figure 7(d):

[0160] Figure 7(b) is the amplitude diagram of the external optical fiber, with the horizontal axis being time and the vertical axis being amplitude. At the external optical fiber, the amplitude fluctuates greatly with the increase of time. Figure 7(c) is the amplitude diagram of the optical fiber pickup, with the horizontal axis being time and the vertical axis being amplitude. At the optical fiber pickup, the amplitude fluctuates less with the increase of time, and the amplitude at the external optical fiber is much larger than the amplitude at the optical fiber pickup.

[0161] The horizontal axis of Figure 7(d) is the unit, the unit is m, 1 unit can be 10m, and the vertical axis is the amplitude, the unit is HZ, then, the circled parts in the figure are the minimum values ​​of the amplitude diagram, and the horizontal axis of the minimum value is the fiber position of the fiber pickup, for example: the minimum value of the first circle corresponds to the fiber pickup 1, so the horizontal axis corresponding to the minimum value is about 4 units, that is, 40m, so the fiber position corresponding to the fiber pickup 1 is 40m.

[0162] It can be seen that there are two preset methods of calibrating the optical fiber position corresponding to the optical fiber pickup in the embodiment of the present application, so that errors in actual use can be reduced and the accurate optical fiber position corresponding to the optical fiber pickup can be obtained, so as to accurately determine the target audio signal collected by the optical fiber pickup during the movement of the belt conveyor, thereby improving the accuracy of roller fault monitoring.

[0163] Optionally, based on the generative network model introduced above, the training method of the generative network model is introduced below, such as Figure 3 As shown, the following steps may be included:

[0164] S301, obtaining a sample spectrum diagram obtained by converting a sample audio signal;

[0165] The sample audio signal is an audio signal collected by the sample optical fiber pickup during the movement of the sample belt conveyor, and the sample optical fiber pickup is used to pick up the vibration of the sample roller of the sample belt conveyor;

[0166] It is understandable that the sample audio signal can be a normal audio signal or an abnormal audio signal, and the embodiments of the present application do not specifically limit it. In addition, since abnormal audio signals are not easy to be simulated and generated, the sample audio signal can be a large number of normal audio signals mixed with a small amount of abnormal audio signals.

[0167] It should be emphasized that the sample belt conveyor can be the belt conveyor corresponding to the target roller, or any belt conveyor. Similarly, the sample roller can be the target roller or other rollers. Therefore, when training the generated network model, the sample roller in any belt conveyor can be trained. The embodiment of the present application does not specifically limit the sample belt conveyor and the sample roller. Then, the sample optical fiber pickup can be set in the vibration sensing area of ​​the sample roller of the sample belt conveyor, or it can be set near the sample roller of the sample belt conveyor.

[0168] It can be understood that the process of converting the sample audio signal into the sample spectrogram is consistent with step S202, so it will not be described in detail here.

[0169] S302, inputting the sample spectrogram into a generative network model to be trained to obtain a reconstructed spectrogram;

[0170] It can be understood that by inputting the sample spectrum graph into the generative network model to be trained, the sample spectrum graph can be reconstructed by the generative network model to be trained to obtain a reconstructed spectrum graph. The process of reconstructing the sample spectrum graph is consistent with step S203, so it will not be repeated here.

[0171] S303, determining a loss value of the generated network model based on a difference between the sample spectrogram and the reconstructed spectrogram, and a true value corresponding to the sample audio signal;

[0172] The true value corresponding to the sample audio signal is used to indicate whether the sample audio signal is: an audio signal that can be collected when the sample roller operates normally;

[0173] It can be understood that the true value can be divided into two types, one is used to characterize the sample audio signal as: the audio signal that can be collected when the sample roller is operating normally, that is, the sample audio signal belongs to a normal audio signal; the other is used to characterize the sample audio signal as: the audio signal that can be collected when the sample roller is operating normally, that is, the sample audio signal does not belong to a normal audio signal. Then, based on the determined true value and the difference between the sample spectrum graph and the reconstructed spectrum graph, the loss value of the generated network model can be determined. During the training process, the smaller the loss value of the generated network model is determined, the better the training effect of the generated network model is.

[0174] The following is a detailed description of the process of determining the loss value of the generated network model:

[0175] Optionally, determining the loss value of the generated network model based on the difference between the sample spectrogram and the reconstructed spectrogram and the true value corresponding to the sample audio signal may include step E1:

[0176] Step E1, substituting the sample spectrogram, the reconstructed spectrogram, and the true value corresponding to the sample audio signal into a predetermined loss function to obtain a loss value of the generated network model;

[0177] The predetermined loss function is used to make: when the true value of the sample audio signal is a first-class true value, the smaller the difference between the sample spectrogram and the reconstructed spectrogram, the smaller the function value, and when the true value of the sample audio signal is a second-class true value, the larger the difference between the sample spectrogram and the reconstructed spectrogram, the smaller the function value;

[0178] The first type of truth value representation: the sample audio signal is an audio signal that can be collected when the sample roller operates normally, and the second type of truth value representation: the sample audio signal is not an audio signal that can be collected when the sample roller operates normally.

[0179] It can be understood that the true value of the sample audio signal can be a first-class true value, that is, the sample audio signal is a normal audio signal, or a second-class true value, that is, the sample audio signal is not a normal audio signal. However, after substituting it into the loss function, it is expected that the determined function value is smaller, that is, the loss value of the generated network model is smaller. It should be emphasized that, when the true value of the sample audio signal is the first type of true value, the smaller the difference between the sample spectrum and the reconstructed spectrum, the smaller the function value, and the better the training effect of the generated network model. Therefore, the difference between the sample spectrum and the reconstructed spectrum is directly proportional to the function value; conversely, when the true value of the sample audio signal is the second type of true value, the greater the difference between the sample spectrum and the reconstructed spectrum, the smaller the function value, and the better the training effect of the generated network model. Therefore, the difference between the sample spectrum and the reconstructed spectrum is inversely proportional to the function value. That is, the ultimate goal of generating the network model is that the reconstructed spectrum is generated in the direction of the spectrum converted from the normal audio signal, and the spectrum reconstructed by the true value of the sample audio signal is consistent whether it is the first type of true value or the second type of true value.

[0180] Optionally, the predetermined loss function includes:

[0181] loss=(‖XX hat ‖ 2 +ε) -label

[0182] Wherein, loss is the loss value of the generated network model, X is the sample spectrum graph, and X hat is the reconstructed spectrum diagram, label is the true value of the sample audio signal, the first category true value is -1, the second category true value is 1, ε is a predetermined setting parameter, and the setting parameter is a parameter used to prevent the denominator from being zero.

[0183] It can be understood that when the sample audio signal is a normal audio signal, the true value of the sample audio signal is the first type of true value, that is, -1, then the subscript -label in the upper right corner of the loss function is 1, if the sample audio graph is similar to the reconstructed spectrum graph, then the loss value will be very small; similarly, when the sample audio signal is not a normal audio signal, the true value of the sample audio signal is the second type of true value, that is, 1, then the subscript -label in the upper right corner of the loss function is -1, at this time, loss is a fraction, if the sample audio graph is not similar to the reconstructed spectrum graph, then the loss value will be very small. In addition, ε can be understood as an empirical value. In order to prevent the denominator from being 0, ε = 1e in general. -9 It should be emphasized that when the sample audio signal is a normal audio signal, when the difference between the sample spectrum and the reconstructed spectrum is small, the loss value of the generated network model is smaller; conversely, when the sample audio signal is not a normal audio signal, when the difference between the sample spectrum and the reconstructed spectrum is large, the loss value of the generated network model is smaller.

[0184] S304: Adjust the network parameters of the generated network model based on the loss value.

[0185] It can be understood that, for the first type of true value sample audio signal, adjusting the network parameters of the generated network model makes the sample spectrum graph and the reconstructed spectrum graph extremely similar, while for the second type of true value sample audio signal, the adjustment makes the sample spectrum graph and the reconstructed spectrum graph quite different. However, the purpose of the adjustment is to make the loss value smaller. The smaller the loss value, the better the training effect of the generated network model.

[0186] In addition, in the prior art, the analysis of spectrograms is generally based on neural network models. Traditional neural network models are generally binary classification models, that is, a variety of sample spectrograms need to be input into the neural network model. Therefore, a large number of spectrograms converted from abnormal audio signals are required to enrich the sample library before a neural network model with high accuracy can be trained. In actual use, the neural network model cannot effectively identify fresh sample spectrograms that have not been trained, and it is not easy to simulate the collection of a large number of abnormal audio signals. Therefore, this method lacks versatility.

[0187] It can be seen that the embodiment of the present application can effectively train the generative network model, so that the generative network model can generate a spectrum converted from a normal audio signal regardless of the target audio signal. Compared with the traditional binary classification model, the generative network model used in the embodiment of the present application has better versatility and higher accuracy for roller fault monitoring. In addition, the embodiment of the present application also solves the problem of consuming effort to simulate and collect a large number of abnormal audio signals.

[0188] The following is a brief introduction to the optical fiber pickup used in the embodiments of the present application:

[0189] Optionally, there are multiple fiber optic pickups, and the multiple fiber optic pickups belong to at least one pickup group. In the same pickup group, the internal optical fiber of the fiber optic pickup and the external optical fiber between adjacent fiber optic pickups are composed of the same optical fiber, and different pickup groups are connected by a predetermined connection method.

[0190] It is understandable that a pickup assembly may include multiple fiber optic pickups, and each two fiber optic pickups are connected by an external optical fiber. A winding shaft is provided inside the fiber optic pickup, and the internal optical fiber of the fiber optic pickup is wound on the winding shaft. In the same pickup group, the internal optical fiber of the fiber optic pickup and the external optical fiber between adjacent fiber optic pickups are composed of the same optical fiber, and different pickup groups can be connected by fusion or by using fiber optic flanges. It should be emphasized that since fiber loss will occur when connecting by fusion or by using fiber optic flanges, the same fiber in the same pickup group can reduce fiber loss and achieve long-distance pickup.

[0191] It can be seen that the embodiment of the present application divides multiple fiber optic pickups into the same pickup group. In the same pickup group, the internal optical fiber of the fiber optic pickup and the external optical fiber between adjacent fiber optic pickups are composed of the same optical fiber, which can reduce fiber loss and achieve long-distance sound pickup.

[0192] Optionally, the external optical fibers between adjacent optical fiber pickups in the same group are provided with optical fiber protective covers.

[0193] It is understandable that providing an optical fiber protective sleeve on the external optical fiber between adjacent optical fiber pickups can prevent the optical fiber from stretching, impacting, and bending, thereby protecting the external optical fiber and ensuring that stable fault monitoring can be performed on the roller. In addition, the optical fiber protective sleeve can be an armored tube, which can also be other protective tubes, and the embodiments of the present application do not specifically limit this.

[0194] Optionally, the optical fiber pickup has a plurality of sound pickup holes, and the plurality of sound pickup holes are all arranged on the same surface of the optical fiber pickup, and the opening direction of the sound pickup holes is the same as the direction of gravity.

[0195] It is understandable that after the fiber optic pickup is installed, the opening direction of the pickup hole is downward, which is the same as the direction of gravity, so that dust or foreign matter can be placed there and it can also serve as a drainage hole.

[0196] Optionally, the internal optical fiber and the corresponding external optical fiber of the fiber optic pickup are both sound-sensitive optical fibers, and the sound-sensitive optical fibers are characterized by adding sensitizing materials to the bare optical fibers.

[0197] Among them, the sensitizing material can be PVC (Polyvinyl chloride), PU (polyurethane), TPU (thermoplastic polyurethane elastomer), PEBA (Polyeher block amide), TPEE (thermoplastic polyester elastomer), TPV (Thermoplastic Vulcanizate), EPDM (Ethylene-Propylene-Diene Monomer), EVA (Ethylene Vinyl Acetate Copolymer), space cotton, pearl cotton, etc. It can be understood that adding the above sensitizing materials to the bare optical fiber can make the optical fiber more sensitive to sound and improve the sound pickup effect.

[0198] Optionally, the internal optical fiber of the fiber pickup and the corresponding external optical fiber are arranged based on spatial resolution;

[0199] The length of the internal optical fiber is not less than twice the length represented by the spatial resolution, and the length of the external optical fiber is not less than the length represented by the spatial resolution.

[0200] It is understandable that the layout of the internal and external optical fibers of the fiber optic pickup is based on spatial resolution.

[0201] Exemplarily, the internal optical fiber length is A, the external optical fiber length is B, the spatial resolution is N, the internal optical fiber length is not less than 2*N, and the external optical fiber length is not less than N.

[0202] Exemplarily, the spatial resolution is 10 m, the length of the inner optical fiber wound on the winding shaft of the optical fiber pickup is not less than 20 m, and the length of the outer optical fiber is not less than 10 m.

[0203] It can be seen that the embodiments of the present application are arranged based on spatial resolution, which does not waste the length of the optical fiber, while also ensuring that the optical fiber pickup can effectively collect the target audio signal.

[0204] In one implementation, the embodiment of the present application can also use the spatial resolution to calculate the layable length of the actual belt line of the belt conveyor. The repetition frequency set by the DAS system is K, the speed of light is c, the refractive index of the optical fiber is n, the internal optical fiber length is A, and the external optical fiber length is B. Then, the maximum optical fiber length that the DAS system can connect is The actual laying length of the belt line is

[0205] For example, the repetition frequency of the DAS system is 10kHz, the maximum optical fiber length that the DAS system can connect is 10km, the internal optical fiber length is 20m, and the external optical fiber length is 10m. Then, the actual belt line laying length is 10000 / (10 / (10+20))≈3.3km.

[0206] It can be seen that the embodiment of the present application can estimate the actual laying length of the belt line, so that when the fiber optic pickup picks up the sound of other belt conveyors, it can be estimated whether the replaced belt conveyor is suitable for the fiber optic pickup.

[0207] For a better understanding, the fiber optic pickup and its application scenarios are briefly introduced below in conjunction with FIG8(a) and FIG8(b):

[0208] As shown in FIG8(a), the fiber optic pickup includes an upper shell 810 and a lower shell 820. The fiber optic pickup includes a winding shaft 830 in the middle, on which an internal optical fiber is wound. The fiber optic pickups are connected by an external optical fiber, and a fiber protection tube 840 is provided on the external optical fiber for protection.

[0209] As shown in FIG8(b), the belt conveyor includes a conveyor belt 801 and a roller 802. A bracket 803 is provided at the lower part of the belt conveyor. A connecting part 806 is provided between the belt conveyor and the bracket. The optical fiber pickup 804 can be provided near the roller bracket. A fusion point or optical fiber flange 805 is provided between each pickup group.

[0210] It should be emphasized that the roller bracket is blocked due to the viewing angle. It is understandable that when the fiber optic pickup 804 is installed, the pickup hole faces downward, and it can be installed near the roller bracket of the belt conveyor, or it can be installed on the roller bracket and installed on the roller bracket by clamps, screws, magnets, adhesives and other fastening methods. In addition, when installing the fiber optic pickup 804, try to avoid direct transmission of vibration to the winding shaft. For example, if the fiber optic pickup is installed on the roller bracket of the belt conveyor, the vibration generated by the roller may be directly transmitted to the winding shaft, and the product needs certain shock-absorbing measures, such as adding shock-absorbing pads, shock-absorbing brackets, dampers or hanging installation.

[0211] It can be seen that the embodiments of the present application can adopt a variety of methods when installing the fiber optic pickup to avoid direct transmission to the winding shaft, thereby preventing inaccurate pickup effects and further improving the accuracy of roller fault monitoring.

[0212] Based on the above method embodiment, Fig. 9As shown, an embodiment of the present application provides a fault monitoring device for a roller, the device comprising:

[0213] An acquisition module 910 is used to acquire a target audio signal collected by the optical fiber pickup during the movement of the belt conveyor; wherein the optical fiber pickup is used to pick up the vibration of the target roller of the belt conveyor;

[0214] A conversion module 920, configured to convert the target audio signal into a first spectrogram;

[0215] An input module 930 is used to input the first spectrogram into a pre-trained generative network model to obtain a second spectrogram; wherein the generative network model is used to reconstruct the first spectrogram into a spectrogram obtained by converting a normal audio signal, and the normal audio signal is an audio signal that can be collected when the target roller is operating normally;

[0216] A first determination module 940, configured to determine whether the target audio signal belongs to the normal audio signal based on whether the first spectrogram and the second spectrogram are similar;

[0217] The second determination module 950 is configured to determine that the target roller is a faulty roller if the target audio signal does not belong to the normal audio signal.

[0218] Optionally, the first determining module includes:

[0219] An input submodule, used to input the first spectrogram and the second spectrogram into a pre-trained discriminator network model to obtain an output result; wherein the discriminator network model is used to analyze whether the first spectrogram and the second spectrogram are similar;

[0220] A determination submodule is used to determine whether the target audio signal belongs to the normal audio signal based on the output result.

[0221] Optionally, the training method of generating the network model includes:

[0222] Obtaining a sample spectrum diagram obtained by converting a sample audio signal; wherein the sample audio signal is an audio signal collected by a sample optical fiber pickup during the movement of a sample belt conveyor, and the sample optical fiber pickup is used to pick up the vibration of a sample roller of the sample belt conveyor;

[0223] Inputting the sample spectrogram into a generative network model to be trained to obtain a reconstructed spectrogram;

[0224] Based on the difference between the sample spectrum graph and the reconstructed spectrum graph, and the true value corresponding to the sample audio signal, determining the loss value of the generated network model; wherein the true value corresponding to the sample audio signal is used to characterize whether the sample audio signal is: an audio signal that can be collected when the sample roller operates normally;

[0225] Based on the loss value, the network parameters of the generated network model are adjusted.

[0226] Optionally, determining the loss value of the generated network model based on the difference between the sample spectrogram and the reconstructed spectrogram, and the true value corresponding to the sample audio signal, includes:

[0227] Substituting the sample spectrogram, the reconstructed spectrogram, and the true value corresponding to the sample audio signal into a predetermined loss function to obtain a loss value of the generated network model;

[0228] The predetermined loss function is used to make: when the true value of the sample audio signal is a first-class true value, the smaller the difference between the sample spectrogram and the reconstructed spectrogram, the smaller the function value, and when the true value of the sample audio signal is a second-class true value, the larger the difference between the sample spectrogram and the reconstructed spectrogram, the smaller the function value;

[0229] The first type of truth value representation: the sample audio signal is an audio signal that can be collected when the sample roller operates normally, and the second type of truth value representation: the sample audio signal is not an audio signal that can be collected when the sample roller operates normally.

[0230] Optionally, the predetermined loss function includes:

[0231] loss=(‖XX hat ‖ 2 +ε) -label

[0232] Wherein, loss is the loss value of the generated network model, X is the sample spectrum graph, and X hat is the reconstructed spectrum diagram, label is the true value of the sample audio signal, the first category true value is -1, the second category true value is 1, ε is a predetermined setting parameter, and the setting parameter is a parameter used to prevent the denominator from being zero.

[0233] Optionally, the device further comprises:

[0234] a transform module, configured to perform Fourier transform on the target audio signal to obtain a third spectrogram before converting the target audio signal into the first spectrogram;

[0235] A filtering module, used to filter the third spectrum map using a mask corresponding to a target noise signal to obtain a fourth spectrum map; wherein the target noise signal is an audio signal generated and collected when the belt conveyor is in a silent state;

[0236] The inverse transform module is used to perform inverse Fourier transform on the fourth spectrum graph to obtain a denoised target audio signal.

[0237] Optionally, a method of generating a mask corresponding to the target noise signal includes:

[0238] Performing Fourier transform on the target noise signal to generate a fifth frequency spectrum;

[0239] A maximum value in the fifth spectrum graph is determined as a segmentation threshold, and a mask corresponding to the target noise signal is generated using the segmentation threshold.

[0240] Optionally, the method for determining the target audio signal collected by the optical fiber pickup during the movement of the belt conveyor includes:

[0241] Obtaining an electrical signal converted from an optical signal transmitted by a target optical fiber; wherein the target optical fiber is an optical fiber to which the optical fiber pickup belongs, and the target optical fiber is used to transmit a modulated pulse optical signal sent by a designated laser during the movement of the belt conveyor;

[0242] Based on the pre-calibrated optical fiber position corresponding to the optical fiber pickup, the audio signal is restored to the electrical signal to obtain the target audio signal collected by the optical fiber pickup during the movement of the belt conveyor;

[0243] The optical fiber position corresponding to the optical fiber pickup is used to characterize the position point of the optical fiber pickup in the target optical fiber.

[0244] Optionally, the calibration method of the optical fiber position corresponding to the optical fiber pickup includes: a first calibration method or a second calibration method;

[0245] The first calibration method includes: obtaining a first audio signal, converting the first audio signal into a target spectrum diagram, and using the target spectrum diagram to determine the optical fiber position corresponding to the optical fiber pickup; wherein the first audio signal is: an audio signal along the optical fiber obtained when an audio playback device set at a specified position plays a high-frequency signal; the specified position is a position in the sound pickup area corresponding to the optical fiber pickup;

[0246] The second calibration method includes: obtaining a second audio signal along the optical fiber of the target optical fiber within a predetermined time period, and generating a vibration amplitude diagram of the obtained second audio signal, performing amplitude averaging processing on the vibration amplitude diagram of the second audio signal to obtain a target amplitude diagram, and counting the amplitudes of the target amplitude diagram at the same time within the predetermined time period to obtain an amplitude diagram along the optical fiber, and determining the optical fiber position corresponding to the optical fiber pickup based on the optical fiber position corresponding to the minimum value in the amplitude diagram along the optical fiber.

[0247] Optionally, there are multiple fiber optic pickups, and the multiple fiber optic pickups belong to at least one pickup group. In the same pickup group, the internal optical fiber of the fiber optic pickup and the external optical fiber between adjacent fiber optic pickups are composed of the same optical fiber, and different pickup groups are connected by a predetermined connection method.

[0248] Optionally, the external optical fibers between adjacent optical fiber pickups in the same group are provided with optical fiber protective covers.

[0249] Optionally, the optical fiber pickup has a plurality of sound pickup holes, and the plurality of sound pickup holes are all arranged on the same surface of the optical fiber pickup, and the opening direction of the sound pickup holes is the same as the direction of gravity.

[0250] Optionally, the internal optical fiber and the corresponding external optical fiber of the fiber optic pickup are both sound-sensitive optical fibers, and the sound-sensitive optical fibers are characterized by adding sensitizing materials to the bare optical fibers.

[0251] Optionally, the internal optical fiber of the fiber pickup and the corresponding external optical fiber are arranged based on spatial resolution;

[0252] The length of the internal optical fiber is not less than twice the length represented by the spatial resolution, and the length of the external optical fiber is not less than the length represented by the spatial resolution.

[0253] In the technical solution of this application, the operations involved in obtaining, storing, using, processing, transmitting, providing and disclosing user personal information are all carried out with the user's authorization.

[0254] The present application also provides an electronic device, such as Fig.10 As shown, including:

[0255] Memory 1001, used for storing computer programs;

[0256] The processor 1002 is used to implement the above-mentioned fault monitoring method for the roller when executing the program stored in the memory 1001.

[0257] Furthermore, the electronic device may further include a communication bus and / or a communication interface, and the processor 1002, the communication interface, and the memory 1001 communicate with each other via the communication bus.

[0258] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0259] The communication interface is used for communication between the above electronic device and other devices.

[0260] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0261] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0262] In another embodiment provided in the present application, a computer-readable storage medium is further provided, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, any of the above-mentioned fault monitoring methods for rollers is implemented.

[0263] In another embodiment provided in the present application, a computer program product including instructions is also provided, which, when executed on a computer, enables the computer to execute any fault monitoring method for a roller in the above embodiments.

[0264] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a solid-state hard disk (SSD), etc.

[0265] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" 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 includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0266] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0267] The above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application are included in the protection scope of the present application.

Claims

1. A fault monitoring method for a roller, characterized in that: The method comprises: Acquire a target audio signal collected by the optical fiber pickup during the movement of the belt conveyor; wherein the optical fiber pickup is used to pick up the vibration of the target roller of the belt conveyor; Converting the target audio signal into a first spectrogram; The first spectrogram is input into a pre-trained generative network model to obtain a second spectrogram; wherein the generative network model is used to reconstruct the first spectrogram into a spectrogram obtained by converting a normal audio signal, wherein the normal audio signal is an audio signal that can be collected when the target roller is operating normally; determining whether the target audio signal belongs to the normal audio signal based on whether the first spectrogram and the second spectrogram are similar; If the target audio signal does not belong to the normal audio signal, it is determined that the target roller is a faulty roller.

2. The method according to claim 1, characterized in that The training method of generating the network model includes: Obtaining a sample spectrum diagram obtained by converting a sample audio signal; wherein the sample audio signal is an audio signal collected by a sample optical fiber pickup during the movement of a sample belt conveyor, and the sample optical fiber pickup is used to pick up the vibration of a sample roller of the sample belt conveyor; Inputting the sample spectrogram into a generative network model to be trained to obtain a reconstructed spectrogram; Based on the difference between the sample spectrum graph and the reconstructed spectrum graph, and the true value corresponding to the sample audio signal, determining the loss value of the generated network model; wherein the true value corresponding to the sample audio signal is used to characterize whether the sample audio signal is: an audio signal that can be collected when the sample roller operates normally; Based on the loss value, the network parameters of the generated network model are adjusted.

3. The method according to claim 2, characterized in that The determining the loss value of the generated network model based on the difference between the sample spectrogram and the reconstructed spectrogram and the true value corresponding to the sample audio signal comprises: Substituting the sample spectrogram, the reconstructed spectrogram, and the true value corresponding to the sample audio signal into a predetermined loss function to obtain a loss value of the generated network model; The predetermined loss function is used to make: when the true value of the sample audio signal is a first-class true value, the smaller the difference between the sample spectrogram and the reconstructed spectrogram, the smaller the function value, and when the true value of the sample audio signal is a second-class true value, the larger the difference between the sample spectrogram and the reconstructed spectrogram, the smaller the function value; The first type of truth value representation: the sample audio signal is an audio signal that can be collected when the sample roller operates normally, and the second type of truth value representation: the sample audio signal is not an audio signal that can be collected when the sample roller operates normally.

4. The method according to claim 3, characterized in that The predetermined loss function includes: loss=(‖X-X hat ‖ 2 +ε) -label Wherein, loss is the loss value of the generated network model, X is the sample spectrum graph, and X hat is the reconstructed spectrum diagram, label is the true value of the sample audio signal, the first category true value is -1, the second category true value is 1, ε is a predetermined setting parameter, and the setting parameter is a parameter used to prevent the denominator from being zero.

5. The method according to claim 1, characterized in that Before converting the target audio signal into a first spectrogram, the method further includes: Performing Fourier transform on the target audio signal to obtain a third spectrum graph; The third spectrum graph is filtered using a mask corresponding to a target noise signal to obtain a fourth spectrum graph; wherein the target noise signal is an audio signal generated and collected when the belt conveyor is in a silent state; Perform inverse Fourier transform on the fourth spectrum graph to obtain a denoised target audio signal.

6. The method according to claim 5, characterized in that The generation method of the mask corresponding to the target noise signal includes: Performing Fourier transform on the target noise signal to generate a fifth frequency spectrum; A maximum value in the fifth spectrum graph is determined as a segmentation threshold, and a mask corresponding to the target noise signal is generated using the segmentation threshold.

7. The method according to claim 1, characterized in that The method for determining the target audio signal collected by the optical fiber pickup during the movement of the belt conveyor includes: Obtaining an electrical signal converted from an optical signal transmitted by a target optical fiber; wherein the target optical fiber is an optical fiber to which the optical fiber pickup belongs, and the target optical fiber is used to transmit a modulated pulse optical signal sent by a designated laser during the movement of the belt conveyor; Based on the pre-calibrated optical fiber position corresponding to the optical fiber pickup, the audio signal is restored to the electrical signal to obtain the target audio signal collected by the optical fiber pickup during the movement of the belt conveyor; The optical fiber position corresponding to the optical fiber pickup is used to characterize the position point of the optical fiber pickup in the target optical fiber.

8. The method according to claim 7, characterized in that The calibration method of the optical fiber position corresponding to the optical fiber pickup includes: a first calibration method or a second calibration method; The first calibration method includes: obtaining a first audio signal, converting the first audio signal into a target spectrum diagram, and using the target spectrum diagram to determine the optical fiber position corresponding to the optical fiber pickup; wherein the first audio signal is: an audio signal along the optical fiber obtained when an audio playback device set at a specified position plays a high-frequency signal; the specified position is a position in the sound pickup area corresponding to the optical fiber pickup; The second calibration method includes: obtaining a second audio signal along the optical fiber of the target optical fiber within a predetermined time period, and generating a vibration amplitude diagram of the obtained second audio signal, performing amplitude averaging processing on the vibration amplitude diagram of the second audio signal to obtain a target amplitude diagram, and counting the amplitudes of the target amplitude diagram at the same time within the predetermined time period to obtain an amplitude diagram along the optical fiber, and determining the optical fiber position corresponding to the optical fiber pickup based on the optical fiber position corresponding to the minimum value in the amplitude diagram along the optical fiber.

9. The method according to claim 1, characterized in that: The number of the fiber optic pickups is multiple, and the multiple fiber optic pickups belong to at least one pickup group. In the same pickup group, the internal optical fiber of the fiber optic pickup and the external optical fiber between adjacent fiber optic pickups are composed of the same optical fiber, and different pickup groups are connected to each other through a predetermined connection method; the external optical fiber between adjacent fiber optic pickups in the same group is provided with an optical fiber protective cover; and / or, The optical fiber pickup has a plurality of pickup holes, the plurality of pickup holes are arranged on the same surface of the optical fiber pickup, and the opening direction of the pickup holes is the same as the direction of gravity; and / or, The internal optical fiber and the corresponding external optical fiber of the optical fiber pickup are both sound-sensitive optical fibers, and the sound-sensitive optical fiber refers to an optical fiber with a sensitizing material added to a bare optical fiber; and / or, The internal optical fiber and the corresponding external optical fiber of the fiber pickup are arranged based on the spatial resolution; wherein the length of the internal optical fiber is not less than twice the length represented by the spatial resolution, and the length of the external optical fiber is not less than the length represented by the spatial resolution.

10. A fault monitoring device for a roller, characterized in that: The device comprises: An acquisition module, used to acquire a target audio signal collected by the optical fiber pickup during the movement of the belt conveyor; wherein the optical fiber pickup is used to pick up the vibration of the target roller of the belt conveyor; A conversion module, used for converting the target audio signal into a first spectrogram; An input module, used for inputting the first spectrogram into a pre-trained generative network model to obtain a second spectrogram; wherein the generative network model is used for reconstructing the first spectrogram into a spectrogram obtained by converting a normal audio signal, wherein the normal audio signal is an audio signal that can be collected when the target roller is operating normally; A first determination module, configured to determine whether the target audio signal belongs to the normal audio signal based on whether the first spectrogram and the second spectrogram are similar; The second determination module is configured to determine that the target roller is a faulty roller if the target audio signal does not belong to the normal audio signal.

11. The device according to claim 10, characterized in that The first determining module includes: An input submodule, used to input the first spectrogram and the second spectrogram into a pre-trained discriminator network model to obtain an output result; wherein the discriminator network model is used to analyze whether the first spectrogram and the second spectrogram are similar; A determination submodule, configured to determine whether the target audio signal belongs to the normal audio signal based on the output result; The training method of generating the network model includes: Obtaining a sample spectrum diagram obtained by converting a sample audio signal; wherein the sample audio signal is an audio signal collected by a sample optical fiber pickup during the movement of a sample belt conveyor, and the sample optical fiber pickup is used to pick up the vibration of a sample roller of the sample belt conveyor; Inputting the sample spectrogram into a generative network model to be trained to obtain a reconstructed spectrogram; Based on the difference between the sample spectrum graph and the reconstructed spectrum graph, and the true value corresponding to the sample audio signal, determining the loss value of the generated network model; wherein the true value corresponding to the sample audio signal is used to characterize whether the sample audio signal is: an audio signal that can be collected when the sample roller operates normally; Based on the loss value, adjusting the network parameters of the generated network model; The determining the loss value of the generated network model based on the difference between the sample spectrogram and the reconstructed spectrogram and the true value corresponding to the sample audio signal comprises: Substituting the sample spectrogram, the reconstructed spectrogram, and the true value corresponding to the sample audio signal into a predetermined loss function to obtain a loss value of the generated network model; The predetermined loss function is used to make: when the true value of the sample audio signal is a first-class true value, the smaller the difference between the sample spectrogram and the reconstructed spectrogram, the smaller the function value, and when the true value of the sample audio signal is a second-class true value, the larger the difference between the sample spectrogram and the reconstructed spectrogram, the smaller the function value; The first type of truth value indicates that: the sample audio signal is an audio signal that can be collected when the sample roller operates normally; the second type of truth value indicates that: the sample audio signal is not an audio signal that can be collected when the sample roller operates normally; The predetermined loss function includes: loss=(‖X-X hat ‖ 2 +ε) -label Wherein, loss is the loss value of the generated network model, X is the sample spectrum graph, and X hat is a reconstructed spectrum graph, label is the true value of the sample audio signal, the first category true value is -1, the second category true value is 1, ε is a predetermined setting parameter, and the setting parameter is a parameter used to prevent the denominator from being 0; The device also includes: a transform module, configured to perform Fourier transform on the target audio signal to obtain a third spectrogram before converting the target audio signal into the first spectrogram; A filtering module, used to filter the third spectrum map using a mask corresponding to a target noise signal to obtain a fourth spectrum map; wherein the target noise signal is an audio signal generated and collected when the belt conveyor is in a silent state; An inverse transform module, used for performing an inverse Fourier transform on the fourth spectrum graph to obtain a denoised target audio signal; The generation method of the mask corresponding to the target noise signal includes: Performing Fourier transform on the target noise signal to generate a fifth frequency spectrum; Determine a maximum value in the fifth spectrum graph as a segmentation threshold, and use the segmentation threshold to generate a mask corresponding to the target noise signal; The method for determining the target audio signal collected by the optical fiber pickup during the movement of the belt conveyor includes: Obtaining an electrical signal converted from an optical signal transmitted by a target optical fiber; wherein the target optical fiber is an optical fiber to which the optical fiber pickup belongs, and the target optical fiber is used to transmit a modulated pulse optical signal sent by a designated laser during the movement of the belt conveyor; Based on the pre-calibrated optical fiber position corresponding to the optical fiber pickup, the audio signal is restored to the electrical signal to obtain the target audio signal collected by the optical fiber pickup during the movement of the belt conveyor; The optical fiber position corresponding to the optical fiber pickup is used to characterize the position point of the optical fiber pickup in the target optical fiber; The calibration method of the optical fiber position corresponding to the optical fiber pickup includes: a first calibration method or a second calibration method; The first calibration method includes: obtaining a first audio signal, converting the first audio signal into a target spectrum diagram, and using the target spectrum diagram to determine the optical fiber position corresponding to the optical fiber pickup; wherein the first audio signal is: an audio signal along the optical fiber obtained when an audio playback device set at a specified position plays a high-frequency signal; the specified position is a position in the sound pickup area corresponding to the optical fiber pickup; The second calibration method includes: obtaining a second audio signal along the optical fiber of the target optical fiber within a predetermined time period, and generating a vibration amplitude diagram of the obtained second audio signal, performing amplitude averaging processing on the vibration amplitude diagram of the second audio signal to obtain a target amplitude diagram, and counting the amplitude of the target amplitude diagram at the same time within the predetermined time period to obtain an amplitude diagram along the optical fiber, and determining the optical fiber position corresponding to the optical fiber pickup based on the optical fiber position corresponding to the minimum value in the amplitude diagram along the optical fiber; The number of the optical fiber pickups is multiple, and the multiple optical fiber pickups belong to at least one pickup group. In the same pickup group, the internal optical fiber of the optical fiber pickup and the external optical fiber between adjacent optical fiber pickups are composed of the same optical fiber, and different pickup groups are connected by a predetermined connection method. The external optical fibers between adjacent optical fiber pickups in the same group are provided with optical fiber protective sleeves; The optical fiber pickup has a plurality of pickup holes, which are all arranged on the same surface of the optical fiber pickup, and the opening direction of the pickup holes is the same as the direction of gravity; The internal optical fiber and the corresponding external optical fiber of the optical fiber pickup are both sound-sensitive optical fibers, and the sound-sensitive optical fiber refers to an optical fiber with a sensitizing material added to a bare optical fiber; The internal optical fiber of the optical fiber pickup and the corresponding external optical fiber are arranged based on spatial resolution; The length of the internal optical fiber is not less than twice the length represented by the spatial resolution, and the length of the external optical fiber is not less than the length represented by the spatial resolution.

12. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, for implementing any of the methods described in claims 1-9 when executing a program stored in a memory.

Images