A fault detection system and method for coal conveyor belt idlers based on inspection robots

Abstract

This invention belongs to the field of inspection robot technology and discloses a fault detection system and method for coal conveyor belt idlers based on an inspection robot. The method includes the following steps: establishing an idler fault image detection model and an idler fault noise detection model; acquiring real-time idler video data and real-time idler noise data; performing corresponding preprocessing on the real-time idler video data and real-time idler noise data to obtain preprocessed idler image data and preprocessed noise frequency band data of consecutive frames; inputting the preprocessed idler image data of consecutive frames into the idler fault image detection model for explicit idler fault detection to obtain explicit fault detection results; inputting the preprocessed noise frequency band data of consecutive frames into the idler fault noise detection model for implicit idler fault detection to obtain implicit fault detection results. This invention solves the problems of large workload, high labor cost, and poor automatic detection effect of existing technologies.

Description

Technical Field

[0001] This invention belongs to the field of inspection robot technology, specifically relating to a fault detection system and method for coal conveyor belt idlers based on an inspection robot. Background Technology

[0002] Idler rollers are one of the main components of belt conveyors, accounting for a significant portion of the overall cost. They are also vulnerable parts, and their quality directly affects the stability of the conveyor. Damaged idlers can even scratch the belt, causing spontaneous combustion and resulting in greater economic losses. Therefore, fault detection of idlers and timely replacement and repair of faulty rollers have become a major research focus. Currently, the main method is manual inspection, which is labor-intensive and costly. Even with some automated detection technologies, only obvious damage to idlers can be detected, failing to identify latent faults, leading to poor detection results. Summary of the Invention

[0003] To address the problems of high workload, high labor costs, and poor automatic detection results in existing technologies, the present invention aims to provide a fault detection system and method for coal conveyor belt idlers based on an inspection robot.

[0004] The technical solution adopted in this invention is as follows:

[0005] A fault detection system for coal conveyor belt idlers based on an inspection robot includes an inspection robot, an inspection track, an edge computing gateway, and a monitoring center. The inspection track is located directly below the idler, and the inspection robot is slidably connected to the inspection track. The inspection robot is equipped with an idler video acquisition unit and an idler noise acquisition unit, both of which face the idler. The edge computing gateway is communicatively connected to both the inspection robot and the monitoring center.

[0006] Furthermore, the inspection robot includes a robot body, a moving unit, a roller video acquisition unit, a roller noise acquisition unit, a running detection unit, a robot main control unit, and a rechargeable battery. The moving unit is located at the bottom of the robot body and is slidably connected to the inspection track. The roller video acquisition unit and the roller noise acquisition unit are both located outside the robot body and face the roller. The running detection unit and the rechargeable battery are both located inside the robot body. The robot main control unit is electrically connected to the moving unit, the roller video acquisition unit, the roller noise acquisition unit, and the running detection unit, and is communicatively connected to the edge computing gateway. The rechargeable battery is electrically connected to the robot main control unit, the moving unit, the roller video acquisition unit, the roller noise acquisition unit, and the running detection unit.

[0007] Furthermore, the edge computing gateway includes an edge computing unit and a network unit. The edge computing unit and the network unit are electrically connected, and the network unit is communicatively connected to the wireless communication module of the inspection robot and the monitoring center, respectively.

[0008] Furthermore, the monitoring center is equipped with a data server, which is connected to the edge computing gateway and also connected to an external cloud data center.

[0009] A method for detecting faults in coal conveyor belt idlers based on an inspection robot, and a coal conveyor belt idler fault detection system, includes the following steps:

[0010] Establish image detection models and noise detection models for idler roller faults;

[0011] Collect real-time video data and real-time noise data of the idler rollers;

[0012] The real-time idler video data and real-time idler noise data are preprocessed respectively to obtain preprocessed idler image data and preprocessed noise frequency band data of consecutive frames.

[0013] The preprocessed idler image data of consecutive frames are input into the idler fault image detection model to detect explicit idler faults and obtain explicit fault detection results.

[0014] The preprocessed noise frequency band data of consecutive frames are input into the idler fault noise detection model to detect hidden idler faults and obtain the hidden fault detection results.

[0015] Furthermore, an image detection model for idler roller faults was established based on the TPH-YOLOv5 algorithm, and a noise detection model for idler roller faults was established based on the GRU-FICN algorithm.

[0016] Furthermore, the network structure of the idler roller fault image detection model includes an input terminal, a Backbone module, a DetectionNeck module, a TPH module, and a Prediction module;

[0017] The network structure of the idler roller fault noise detection model includes an input terminal, a DimensionShuffle module, a two-layer GRU module, a two-layer Inception module, a global pooling layer, a Concat module, and a Softmax module.

[0018] Furthermore, the real-time idler video data is preprocessed accordingly, including frame extraction, Gaussian denoising, grayscale, and normalization.

[0019] The real-time idler noise data is preprocessed by filtering, pre-emphasis, frame extraction, and windowing.

[0020] Furthermore, the preprocessed idler image data of consecutive frames are input into the idler fault image detection model for explicit idler fault detection, including the following steps:

[0021] The preprocessed roller image data of all frames is divided into grids, and the prior bounding box of each grid is obtained.

[0022] The initial prediction box corresponding to each prior box is obtained based on the offset between the prior box and the preset prediction box;

[0023] The initial prediction boxes are filtered by non-maximum suppression based on the preset intersection-union ratio and preset confidence level to obtain the final prediction boxes of all frames of preprocessed idler roller image data.

[0024] The target tracking algorithm is used to track and mark the same idler roller target within the final prediction box;

[0025] Explicit idler fault detection is performed on the idler targets within the final prediction frame to obtain explicit fault detection results.

[0026] Furthermore, the preprocessed noise frequency band data of consecutive frames are input into the idler fault noise detection model for implicit idler fault detection, including the following steps:

[0027] The preprocessed noise frequency band data of consecutive frames are input into the idler fault noise detection model to perform fuzzy idler fault diagnosis, and the noise characteristic data of the current idler and the corresponding fuzzy fault diagnosis results are obtained.

[0028] If the fuzzy fault diagnosis result indicates that the current idler roller has a fault, then extract the explicit fault detection result of the current idler roller.

[0029] If the explicit fault detection result is that there is no explicit fault in the current idler, then extract the idler temperature data of the current idler.

[0030] Based on the fusion and weighted fault diagnosis of idler roller temperature data and noise characteristic data, the results of hidden fault detection are obtained.

[0031] The beneficial effects of this invention are as follows:

[0032] 1) The coal conveyor belt idler fault detection system based on inspection robot provided by the present invention uses inspection robot and inspection track to automatically detect coal conveyor belt idlers, avoiding manual inspection, reducing workload and labor costs. Furthermore, the idler video acquisition unit and idler noise acquisition unit are used to detect idler faults simultaneously, improving the accuracy and efficiency of fault detection. Finally, the edge computing gateway and monitoring center are used for remote data transmission and control, improving the practicality of the system.

[0033] 2) The coal conveyor belt idler fault detection method based on inspection robot provided by the present invention establishes an idler fault image detection model and an idler fault noise detection model based on machine learning, which further improves the accuracy and efficiency of fault detection. Furthermore, it performs explicit fault detection and implicit fault detection on the idler, which can discover the latent faults of the idler and prevent them from developing into serious faults or complete damage to the idler, thus improving the practicality and timeliness of the method.

[0034] Other beneficial effects of the present invention will be further explained in the specific embodiments. Attached Figure Description

[0035] Figure 1 This is a structural block diagram of the coal conveyor belt idler roller fault detection system based on an inspection robot in this invention.

[0036] Figure 2 This is a cross-sectional view of the inspection robot in this invention.

[0037] Figure 3 This is a flowchart of the coal conveyor belt idler fault detection method based on an inspection robot in this invention.

[0038] In the diagram, 1 is the robot body; 2 is the mobile unit; 3 is the roller video acquisition unit; 4 is the roller noise acquisition unit; 5 is the robot main control unit; 6 is the rechargeable battery; 7 is the track body; and 8 is the track support. Detailed Implementation

[0039] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0040] Example 1:

[0041] like Figure 1 As shown, this embodiment provides a fault detection system for coal conveyor belt idlers based on an inspection robot, including an inspection robot, an inspection track, an edge computing gateway, and a monitoring center. The inspection track is set directly below the idler, and the inspection robot is slidably connected to the inspection track. The inspection robot is equipped with an idler video acquisition unit 3 and an idler noise acquisition unit 4, both of which face the idler. The edge computing gateway is communicatively connected to the inspection robot and the monitoring center.

[0042] The inspection robot and the inspection track work together to collect video data and noise data of the idler rollers under the coal conveyor belt. The data is then input into the edge computing gateway for relevant fault detection. The edge computing gateway can process a large amount of edge computing and send the detection results to the monitoring center for display and analysis.

[0043] As a preferred option, such as Figure 2As shown, the inspection robot includes a robot body 1, a mobile unit 2, a roller video acquisition unit 3, a roller noise acquisition unit 4, a running detection unit, a robot main control unit 5, and a rechargeable battery 6. The mobile unit 2 is located at the bottom of the robot body 1 and is slidably connected to the inspection track. The roller video acquisition unit 3 and the roller noise acquisition unit 4 are both located outside the robot body 1 and face the roller. The running detection unit and the rechargeable battery 6 are both located inside the robot body 1. The robot main control unit 5 is electrically connected to the mobile unit 2, the roller video acquisition unit 3, the roller noise acquisition unit 4, and the running detection unit, and is also communicatively connected to an edge computing gateway. The rechargeable battery 6 is electrically connected to the robot main control unit 5, the mobile unit 2, the roller video acquisition unit 3, the roller noise acquisition unit 4, and the running detection unit, respectively.

[0044] The idler roller video acquisition unit 3 includes two thermal imaging cameras located on both sides of the top of the robot and a high-speed camera located in the center of the top of the robot. Both the thermal imaging cameras and the high-speed camera are electrically connected to the robot main control unit 5 and the rechargeable battery 6, respectively. The thermal imaging cameras are used to acquire thermal imaging images of the idler roller. Based on the thermal imaging images, temperature data of various parts of the idler roller can be obtained for subsequent detection of latent faults of the idler roller. The high-speed camera is used to acquire video data of the idler roller.

[0045] The roller noise acquisition unit 4 is a digital microphone, which is electrically connected to the robot main control unit 5 and the rechargeable battery 6 respectively; the digital microphone can extract digital audio data that the robot main control unit 5 can recognize.

[0046] The robot main control unit 5 includes a robot main control module, a first storage module, a motor drive module, and a wireless communication module. The robot main control module is electrically connected to the first storage module, the motor drive module, and the wireless communication module, respectively. The motor drive module is electrically connected to the mobile unit 2, and the wireless communication module is communicatively connected to the edge computing gateway.

[0047] The mobile unit 2 includes a mobile motor and a mobile component, both located on the bottom outer side of the robot body 1. The mobile motor is connected to the mobile component by a bearing, and the mobile motor is electrically connected to the motor drive module and the rechargeable battery 6 respectively. The mobile unit 2 is used to support the mobile task of the inspection robot, that is, to move back and forth on the inspection track body 7 and collect data from each roller.

[0048] The operation detection unit includes a temperature sensor, a humidity sensor, a power sensor, a speed sensor, a vibration sensor, a position sensor, and an A / D converter, all installed inside the robot body 1. The A / D converter is electrically connected to the robot main control module, the temperature sensor, the humidity sensor, the power sensor, the speed sensor, the vibration sensor, and the position sensor, respectively. The temperature sensor and the humidity sensor are both located at the rechargeable battery 6, and the power sensor is electrically connected to the output terminal of the rechargeable battery 6. The operation detection unit is used to detect the normal operating status of the inspection robot, including temperature, humidity, speed, vibration, position, and the power of the rechargeable battery 6, and to detect whether the inspection robot has malfunctioned.

[0049] The inspection track includes a track body 7 and a track support 8. The track support 8 is located directly below the idler roller and is fixedly connected to the bottom end of the coal conveyor frame. The track body 7 is located inside the track support 8 and is an I-shaped track.

[0050] Preferably, the edge computing gateway includes an edge computing unit and a network unit, the edge computing unit and the network unit are electrically connected, and the network unit is communicatively connected to the wireless communication module of the inspection robot and the monitoring center, respectively.

[0051] The edge computing unit includes an edge computing main control module, a second storage module, a preprocessing module, a roller fault image detection module, a roller fault noise detection module, and an encryption module. The edge computing main control module is connected to the second storage module, the preprocessing module, the roller fault image detection module, the roller fault noise detection module, the encryption module, and the network unit. The roller fault image detection module is equipped with a roller fault image detection model, and the roller fault noise detection module is equipped with a roller fault noise detection model.

[0052] The edge computing gateway undertakes most of the computing and remote data transmission functions, enabling control of inspection robots in environments with poor signal. Preprocessing, roller fault image detection, and roller fault noise detection are performed on the edge computing unit, reducing the data processing pressure on the monitoring center and saving storage space in the second storage module. The monitoring center broadcasts the public key to all edge computing units and keeps the corresponding private key locally. The encryption module encrypts the data to be uploaded by the edge computing units based on the public key, and the monitoring center decrypts the encrypted data using the private key, ensuring the security of data transmission.

[0053] As a preferred option, the monitoring center is equipped with a data server, which is connected to the edge computing gateway and is also connected to an external cloud data center.

[0054] The data server includes a parallel data receiving module, a decryption module, a cache database module, and a parallel data uploading module. The parallel data receiving module is connected to the edge computing gateway. The parallel data receiving module, decryption module, cache database module, and parallel data uploading module are connected in sequence. The parallel data uploading module is connected to an external cloud data center.

[0055] The coal conveyor belt idler fault detection system based on inspection robot provided by this invention uses inspection robot and inspection track to automatically detect coal conveyor belt idlers, avoiding manual inspection, reducing workload and labor costs. Furthermore, it uses idler video acquisition unit and idler noise acquisition unit to detect idler faults simultaneously, improving the accuracy and efficiency of fault detection. Finally, it uses edge computing gateway and monitoring center for remote data transmission and control, improving the practicality of the system.

[0056] Example 2:

[0057] like Figure 3 As shown, this embodiment provides a method for detecting faults in coal conveyor belt idlers based on an inspection robot. The coal conveyor belt idler fault detection system includes the following steps:

[0058] Establish image detection models and noise detection models for idler roller faults;

[0059] A roller fault image detection model is established based on the TPH-YOLOv5 algorithm. The network structure of the roller fault image detection model includes an input terminal, a backbone module, a detection neck module, a TPH module, and a prediction module. The first input terminal uses the Mosaic data augmentation method to process the input image. The backbone module includes a focus structure and a CSP structure. The first detection neck module has an FPN+PAN structure. The first prediction module uses the GIOU_Loss function to calculate the loss. A TPH module is added to YOLOv5 to detect roller fault targets of different sizes.

[0060] A noise detection model for idler roller faults is established based on the GRU-FICN algorithm. The network structure of the idler roller fault noise detection model includes an input terminal, a DimensionShuffle module, a two-layer GRU module, a two-layer Inception module, a global pooling layer, a Concat module, and a Softmax module. The convolutional layers in the GRU-FICN algorithm effectively extract local features of the sound signal. The GRU in the model structure is a variant of LSTM, integrating the forget gate and input gate of LSTM into a single update gate. Its structure also contains a reset gate, where the update gate determines the degree to which information from the previous state is passed into the current state. The gate determines how much information from the previous state is written into the current state candidate set. In this embodiment, a two-layer GRU network structure is adopted. The GRU network reduces the computational complexity of LSTM by introducing fewer gating units with fewer parameters. The local feature extraction of the sound signal uses two one-dimensional extended Inception modules and a global pooling layer. The two Inception modules can effectively extract local sound features, while the use of the global average pooling layer greatly reduces the feature value of the network output. The global average pooling operation averages each feature map into an output value. Compared with the commonly used fully connected layer, global pooling has advantages such as fewer parameters, a comprehensive receptive field, and spatial information.

[0061] Collect real-time video data and real-time noise data of the idler rollers;

[0062] The real-time idler video data and real-time idler noise data are preprocessed respectively to obtain preprocessed idler image data and preprocessed noise frequency band data of consecutive frames.

[0063] The real-time idler video data is preprocessed in sequence, including frame capture, Gaussian denoising, grayscale, and normalization. Gaussian denoising removes image noise from the idler image, converts the idler image into a grayscale image, and normalizes it to an idler image of the same size to facilitate subsequent fault detection.

[0064] The real-time idler noise data is preprocessed in sequence, including filtering, pre-emphasis, frame truncation, and windowing. Low-frequency, low-amplitude noise data is filtered to avoid the superposition of adjacent idler noise affecting the detection accuracy. Pre-emphasis enhances the high-frequency part, making the signal spectrum relatively flat and balanced. The selected frame length is 10ms-30ms to treat the sound signal as approximately stationary. In order to ensure continuity and correlation, a Hamming window is selected for windowing processing, that is, each frame of sound signal is multiplied by the window function, so that the two segments of the sound signal in this frame gradually approach zero.

[0065] The preprocessed idler image data of consecutive frames are input into the idler fault image detection model to detect explicit idler faults, and the explicit fault detection results are obtained. The steps include:

[0066] The preprocessed roller image data of all frames is divided into grids, and the prior bounding box of each grid is obtained.

[0067] The initial prediction box corresponding to each prior box is obtained based on the offset between the prior box and the preset prediction box;

[0068] The initial prediction boxes are filtered by non-maximum suppression based on the preset intersection-union ratio and preset confidence level to obtain the final prediction boxes of all frames of preprocessed idler roller image data.

[0069] The target tracking algorithm is used to track and mark the same idler roller target within the final prediction box. Since the idler roller video data is collected from multiple idler rollers during the inspection process, and each idler roller is rotating, tracking and numbering are performed on different idler rollers and different fault parts of the same idler roller, which improves the efficiency and accuracy of fault detection.

[0070] Visible idler fault detection is performed on the idler targets within the final prediction frame to obtain the visible fault detection results. The visible fault detection results indicate whether there are visible faults on the current idler and the type of visible fault, including easily identifiable faults such as idler cracks, idler scratches, and idler breakage that require repair and replacement. Visible faults will have a significant impact on the normal operation of the coal conveyor belt.

[0071] The preprocessed noise frequency band data of consecutive frames are input into the idler roller fault noise detection model to perform latent idler roller fault detection, and the latent fault detection results are obtained, including the following steps:

[0072] The preprocessed noise frequency band data of consecutive frames are input into the idler fault noise detection model to perform fuzzy idler fault diagnosis, and the noise characteristic data of the current idler and the corresponding fuzzy fault diagnosis results are obtained.

[0073] Fuzzy fault diagnosis can analyze noise frequency data to identify both overt and latent faults in idlers. Overt faults are detected using an idler fault image detection model, while the remaining abnormalities are latent faults, including idler tilting and idler deformation. Although these faults may not cause major safety incidents or accidents in the short term, they can gradually develop into serious faults that affect the normal operation of the idlers and coal conveyor belts, requiring early detection and handling.

[0074] If the fuzzy fault diagnosis result indicates that the current idler roller has a fault, then extract the explicit fault detection result of the current idler roller.

[0075] If the explicit fault detection result is that there is no explicit fault in the current idler, then extract the idler temperature data of the current idler.

[0076] A thermal imaging camera is used to capture thermal images of the idler roller, and temperature data at various points on the idler roller can be obtained from the thermal images.

[0077] Based on the fusion and weighted fault diagnosis of idler roller temperature data and noise characteristic data, the results of hidden fault detection are obtained;

[0078] Based on the temperature data distribution of different parts of the idler roller, the probability of failure is obtained. For example, if the temperature data of the non-load-bearing points of the idler roller is high, the idler roller may tilt and the probability of failure is high. Or, if the temperature data of the middle position of the idler roller differs too much, the idler roller may be deformed and about to break, and the probability of failure is high. The probability of failure can also be obtained from the noise characteristic data. The two are fused and weighted, and finally classified to obtain the corresponding hidden faults.

[0079] The present invention provides a method for detecting faults in coal conveyor idlers based on inspection robots. It establishes an image detection model and a noise detection model for idler faults based on machine learning, which further improves the accuracy and efficiency of fault detection. Furthermore, it can detect both explicit and implicit faults in idlers, thus discovering latent faults and preventing them from developing into serious faults or complete damage to the idlers. This improves the practicality and timeliness of the method.

[0080] This invention is not limited to the optional embodiments described above, and anyone can derive other various forms of products based on the teachings of this invention. The specific embodiments described above should not be construed as limiting the scope of protection of this invention; the scope of protection of this invention should be determined by the claims, and the specification can be used to interpret the claims.

Claims

1. A fault detection system for coal conveyor belt idlers based on an inspection robot, characterized in that: The system includes an inspection robot, an inspection track, an edge computing gateway, and a monitoring center. The inspection track is located directly below the idler roller. The inspection robot is slidably connected to the inspection track. The inspection robot is equipped with an idler roller video acquisition unit (3) and an idler roller noise acquisition unit (4). Both the idler roller video acquisition unit (3) and the idler roller noise acquisition unit (4) are facing the idler roller. The edge computing gateway is communicatively connected to the inspection robot and the monitoring center, respectively. The inspection robot includes a robot body (1), a moving unit (2), a roller video acquisition unit (3), a roller noise acquisition unit (4), a running detection unit, a robot main control unit (5), and a rechargeable battery (6). The moving unit (2) is located at the bottom of the robot body (1) and is slidably connected to the inspection track. The roller video acquisition unit (3) and the roller noise acquisition unit (4) are both located outside the robot body (1) and face the inspection track. The idler roller is set up, and the operation detection unit and the rechargeable battery (6) are both set inside the robot body (1). The robot main control unit (5) is electrically connected to the mobile unit (2), the idler roller video acquisition unit (3), the idler roller noise acquisition unit (4) and the operation detection unit respectively. The robot main control unit (5) is also connected to the edge computing gateway. The rechargeable battery (6) is electrically connected to the robot main control unit (5), the mobile unit (2), the idler roller video acquisition unit (3), the idler roller noise acquisition unit (4) and the operation detection unit respectively. The operation detection unit includes a temperature sensor, a humidity sensor, a power sensor, a speed sensor, a vibration sensor, a position sensor, and an A / D converter, all of which are installed inside the robot body (1). The A / D converter is electrically connected to the robot main control module, the temperature sensor, the humidity sensor, the power sensor, the speed sensor, the vibration sensor, and the position sensor, respectively. The temperature sensor and the humidity sensor are both located at the rechargeable battery (6), and the power sensor is electrically connected to the output end of the rechargeable battery (6). The edge computing gateway includes an edge computing unit and a network unit. The edge computing unit and the network unit are electrically connected. The network unit is communicatively connected to the wireless communication module of the inspection robot and the monitoring center, respectively. The edge computing unit includes an edge computing main control module, a second storage module, a preprocessing module, a roller fault image detection module, a roller fault noise detection module, and an encryption module. The edge computing main control module is connected to the second storage module, the preprocessing module, the roller fault image detection module, the roller fault noise detection module, the encryption module, and the network unit. The roller fault image detection module is equipped with a roller fault image detection model, and the roller fault noise detection module is equipped with a roller fault noise detection model. A roller fault image detection model is established based on the TPH-YOLOv5 algorithm. The network structure of the roller fault image detection model includes an input terminal, a Backbone module, a DetectionNeck module, a TPH module, and a Prediction module. A roller fault noise detection model is established based on the GRU-FICN algorithm. The network structure of the roller fault noise detection model includes an input terminal, a DimensionShuffle module, a two-layer GRU module, a two-layer Inception module, a global pooling layer, a Concat module, and a Softmax module.

2. The coal conveyor belt idler fault detection system based on an inspection robot according to claim 1, characterized in that: The monitoring center is equipped with a data server, which is connected to an edge computing gateway and an external cloud data center.

3. A method for detecting faults in coal conveyor belt idlers based on an inspection robot, based on the coal conveyor belt idler fault detection system as described in claim 2, characterized in that: Includes the following steps: Establish image detection models and noise detection models for idler roller faults; A roller fault image detection model was established based on the TPH-YOLOv5 algorithm, and a roller fault noise detection model was established based on the GRU-FICN algorithm. The network structure of the idler roller fault image detection model includes an input terminal, a Backbone module, a DetectionNeck module, a TPH module, and a Prediction module; The network structure of the idler roller fault noise detection model includes an input terminal, a DimensionShuffle module, a two-layer GRU module, a two-layer Inception module, a global pooling layer, a Concat module, and a Softmax module. Collect real-time video data and real-time noise data of the idler rollers; The real-time idler video data and real-time idler noise data are preprocessed respectively to obtain preprocessed idler image data and preprocessed noise frequency band data of consecutive frames. The preprocessed idler image data of consecutive frames are input into the idler fault image detection model to detect explicit idler faults and obtain explicit fault detection results. The preprocessed noise frequency band data of consecutive frames are input into the idler roller fault noise detection model to perform latent idler roller fault detection, and the latent fault detection results are obtained, including the following steps: The preprocessed noise frequency band data of consecutive frames are input into the idler fault noise detection model to perform fuzzy idler fault diagnosis, and the noise characteristic data of the current idler and the corresponding fuzzy fault diagnosis results are obtained. If the fuzzy fault diagnosis result indicates that the current idler roller has a fault, then extract the explicit fault detection result of the current idler roller. If the explicit fault detection result is that there is no explicit fault in the current idler, then extract the idler temperature data of the current idler. Based on the fusion and weighted fault diagnosis of idler roller temperature data and noise characteristic data, the results of hidden fault detection are obtained.

4. The method for detecting faults in coal conveyor belt idlers based on an inspection robot according to claim 3, characterized in that: The real-time idler video data is preprocessed in sequence, including frame extraction, Gaussian denoising, grayscale, and normalization. The real-time idler noise data is preprocessed by filtering, pre-emphasis, frame extraction, and windowing.

5. The method for detecting faults in coal conveyor belt idlers based on an inspection robot according to claim 4, characterized in that: The preprocessed idler image data of consecutive frames are input into the idler fault image detection model for explicit idler fault detection, including the following steps: The preprocessed roller image data of all frames is divided into grids, and the prior bounding box of each grid is obtained. The initial prediction box corresponding to each prior box is obtained based on the offset between the prior box and the preset prediction box; The initial prediction boxes are filtered by non-maximum suppression based on the preset intersection-union ratio and preset confidence level to obtain the final prediction boxes of all frames of preprocessed idler roller image data. The target tracking algorithm is used to track and mark the same idler roller target within the final prediction box; Explicit idler fault detection is performed on the idler targets within the final prediction frame to obtain explicit fault detection results.

Images