Belt health monitoring method based on inspection robot

Abstract

The present invention discloses a belt health monitoring method based on an inspection robot, which belongs to the field of industrial detection and artificial intelligence. By deploying an inspection robot, multiple monitoring cameras and sensors, on-site video and sensor data are acquired, and data preprocessing is performed. The belt status is detected by real-time analysis using an AI algorithm, and the real-time start / stop status, offset coefficient, and material flow ratio of the belt are sent to a background server for storage, and abnormal situations detected are pushed in a timely manner. The real-time data and alarm data stored in the database are read, and the historical data of the belt are modeled and analyzed using an LSTM model and statistical methods to determine the current health status of the belt.

Description

Technical Field

[0001] The present invention belongs to the fields of industrial detection and artificial intelligence, and in particular relates to a belt health monitoring method based on a patrol robot. Background Art

[0002] In the field of belt monitoring system research and development, a variety of different solutions have been proposed both domestically and internationally. Existing belt health monitoring technologies mainly focus on belt anomaly detection, which occurs after the fact; such as foreign object alarms, belt deviation, belt tearing and other anomaly detection. Related sensors and detection systems are mainly divided into two categories: fixed installation and deployment based on inspection robots. At present, research on belt conveyor fault detection mainly includes: foreign object identification, belt spillage, belt slippage, belt deviation, belt tearing, etc. Most of the intelligent belt conveyor detection products on the market can be divided into three categories: monitoring systems based on traditional sensors, detection systems based on machine vision, and intelligent diagnostic systems based on vibration analysis.

[0003] 1. Temperature monitoring device: This device installs infrared temperature sensors or contact thermocouples at key locations on the belt to monitor the belt's operating temperature in real time. It triggers an alarm when the temperature exceeds a preset threshold. This device is simple and inexpensive, but it only reflects localized overheating and cannot fully assess the overall health of the belt.

[0004] 2. Tension Monitoring System: This system uses strain gauges or load cells to measure belt tension during operation. Abnormal changes in tension can be used to identify belt damage or deviation. A typical implementation involves installing a force sensor on the roller bearing seat to measure the dynamic belt tension. This system has high mechanical installation requirements and is susceptible to interference from environmental vibration.

[0005] 3. Visual inspection of surface defects: Industrial cameras are used to capture images of the belt surface and image processing algorithms are used to identify defects such as cracks and wear. Recent advances include the use of high-resolution line scan cameras for continuous scanning, combined with deep learning for defect classification. This method offers high accuracy but is significantly affected by lighting conditions and can be difficult to detect internal damage.

[0006] 4. Spectrum Analysis System: This system uses an accelerometer to collect vibration signals from the belt during operation and performs an FFT transform to obtain a characteristic spectrum. This system can identify characteristic frequencies of faults such as bearing damage and belt joint anomalies. This method requires professional analysts to interpret the spectrum and has limited automation.

[0007] However, belt conveyor systems present many potential safety hazards that cannot be detected or addressed by conventional belt conveyor five-prevention systems. Currently, common belt health monitoring systems primarily focus on post-event belt anomaly detection, such as foreign object alarms, belt deviation, and belt tears. However, these monitoring systems still have technical flaws, leaving many potential safety hazards undetected and addressed by conventional belt conveyor five-prevention systems.

[0008] In response to the existing problems, a belt health monitoring method based on inspection robots is urgently needed. Summary of the Invention

[0009] The technical problem to be solved by the present invention is to provide a belt health monitoring method based on a patrol robot to address the shortcomings of the background technology. The method uses AI algorithm to perform real-time analysis to detect the belt status, sends the real-time start and stop status, offset coefficient, and material flow ratio data of the belt to the background server for storage, and pushes the monitored abnormal conditions in a timely manner; reads the real-time data and alarm data stored in the database, uses the LSTM model and statistical methods to model and analyze the historical data of the belt, and judges the current health status of the belt.

[0010] The present invention adopts the following technical solutions to solve the above technical problems:

[0011] A belt health monitoring method based on a patrol robot specifically comprises the following steps:

[0012] Step 1: Deploy patrol robots, multiple surveillance cameras, and sensors to acquire on-site video and sensor data, and perform data preprocessing.

[0013] Step 2, AI recognition and detection: Use AI algorithms to conduct real-time analysis to detect the belt status, send the belt's real-time start and stop status, offset coefficient, and material flow ratio data to the backend server for storage, and promptly push any abnormal conditions detected;

[0014] Step 3: Read the real-time data and alarm data stored in the database, use the LSTM model and statistical methods to model and analyze the historical data of the belt to determine the current health status of the belt.

[0015] As a further preferred embodiment of the belt health monitoring method based on the inspection robot of the present invention, in step 2, the AI ​​recognition detection includes a Mobile U-Net segmentation model, a YOLO recognition model, and a K-Means clustering;

[0016] Among them, the Mobile U-Net segmentation model is used for belt deviation detection and material flow ratio detection. The Mobile U-Net network model is used to achieve real-time detection of belt deviation and material flow ratio.

[0017] YOLO recognition model: used for foreign object recognition, personnel recognition, and fireworks recognition; uses YOLOV7 for anomaly recognition and detection;

[0018] K-Means clustering: used for tear detection. The K-Means clustering algorithm is used for tear detection. Cluster analysis is performed on the image to obtain a new image with connected domain recoloring. Connected domain marker extraction and analysis are performed to extract abnormal areas with tear characteristics. Combined with the belt area mask, calculations are performed to remove interference from areas outside the belt, thus realizing belt tear anomaly detection.

[0019] As a further preferred solution of the belt health monitoring method based on a patrol robot of the present invention, the Mobile-Unet segmentation model uses a lightweight MobileNetV3 network as the backbone network to achieve deep feature extraction and combines it with the Unet network to improve the overall segmentation performance; specifically, it includes:

[0020] Transplant the bneck module in MobileNetV3 to the encoding part of the Unet network;

[0021] According to the Unet encoding part, bneck modules of different depths are used to extract features. Generally, the deeper the position, the more bneck modules are required, and the more advanced the abstract features extracted.

[0022] The extracted features are sequentially connected to the corresponding positions of the decoder for feature fusion. During the connection process, it is necessary to ensure that the number of channels and size of the feature map and the corresponding position of the decoder are consistent.

[0023] As a further preferred solution of the belt health monitoring method based on the inspection robot of the present invention, tear recognition and detection are performed using the K-Means clustering algorithm, specifically including:

[0024] Randomly select from the dataset Initial cluster centers ,in, , calculate the remaining data objects and cluster centers Euclidean distance, find the cluster center closest to the target data object , and assign data objects to cluster centers The corresponding cluster is obtained; the average value of the data objects in each cluster is calculated as the new cluster center, and the next iteration is performed until the cluster center no longer changes or the maximum number of iterations is reached;

[0025] The Euclidean distance calculation formula between the data object and the cluster center in the space is:

[0026] ;

[0027] in, For data objects; For the cluster centers; is the dimension of the data object; and They are and No. attribute values;

[0028] The sum of squared errors for the entire dataset The calculation formula is:

[0029] ;

[0030] in, The size of indicates the quality of the clustering result; is the number of clusters.

[0031] As a further preferred embodiment of the belt health monitoring method based on the inspection robot of the present invention, in step 3, the belt data is analyzed using the LSTM model and statistical methods to monitor the belt health, which specifically includes the following steps:

[0032] Step 3.1, preprocess the historical data: use interpolation to fill in missing temperature or offset data to ensure the continuity of the time series;

[0033] Step 3.2: Eliminate obviously erroneous data;

[0034] Step 3.3, perform Z-score normalization on temperature and offset data of different dimensions;

[0035] In step 3.4, the LSTM model is used for data analysis and prediction. The belt health status is determined by combining analysis of multiple data dimensions, including the belt start / stop status, operating time, temperature, offset, and material flow ratio.

[0036] As a further preferred embodiment of the belt health monitoring method based on the inspection robot of the present invention, in step 3.2, obviously erroneous data are eliminated based on statistical methods and business rules; the statistical method includes In principle, the business rules include physical limit values.

[0037] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0038] The present invention provides a belt health monitoring method based on an inspection robot; by deploying an inspection robot, multiple monitoring cameras and sensors, on-site video and sensor data are acquired, and data preprocessing is performed; an AI algorithm is used for real-time analysis to detect the belt status, and the real-time start and stop status, offset coefficient, and material flow ratio data of the belt are sent to a background server for storage, and abnormal situations detected are pushed in a timely manner; the real-time data and alarm data stored in the database are read, and the historical data of the belt are modeled and analyzed using an LSTM model and statistical methods to determine the current health status of the belt; the real-time data and abnormal alarm information identified by the algorithm are analyzed using monitoring means such as vision and sensors in combination with data analysis and data mining technology; through data analysis, possible faults and losses are predicted, and effective maintenance suggestions are given, annotations or explanations are provided for certain data or anomalies to help users understand the reasons behind them, and suggestions for improving the performance of the belt conveyor or extending its service life are put forward, thereby solving the problem that the ordinary belt five-protection system cannot clearly show the system effect in a short time. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of a belt health monitoring method based on a patrol robot according to the present invention;

[0040] Figure 2 is a schematic diagram of the Mobile-Unet network model of the present invention; DETAILED DESCRIPTION

[0041] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings:

[0042] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. The present invention is described in detail below based on the drawings and preferred embodiments. The purpose and effect of the present invention will become more clear. It should be understood that the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0043] A belt health monitoring method based on inspection robots, such as Figure 1 As shown, the specific steps include:

[0044] Step 1: Deploy patrol robots, multiple surveillance cameras, and sensors to acquire on-site video and sensor data, and perform data preprocessing.

[0045] Step 2, AI recognition and detection: Use AI algorithms to conduct real-time analysis to detect the belt status, send the belt's real-time start and stop status, offset coefficient, and material flow ratio data to the backend server for storage, and promptly push any abnormal conditions detected;

[0046] Step 3: Read the real-time data and alarm data stored in the database, use the LSTM model and statistical methods to model and analyze the historical data of the belt to determine the current health status of the belt.

[0047] The most critical core algorithms are AI recognition and detection and historical data analysis; AI recognition and detection include the Mobile U-Net segmentation model, the YOLO recognition model, and K-Means clustering; data analysis is implemented using the LSTM model and statistical methods.

[0048] Belt deviation and material flow ratio detection: Utilizing the Mobile U-Net network model, real-time detection of belt deviation and material flow ratio is achieved. Belt and material flow areas are accurately extracted, and belt area masks are created to reduce false alarms caused by environmental and lighting issues outside the belt area.

[0049] To balance recognition accuracy and computational efficiency, the Mobile-Unet network model uses the lightweight MobileNetV3 network as its backbone network for deep feature extraction, combined with the Unet network to improve overall segmentation performance. Designed specifically to address computational constraints in real-world applications, the MobileNetV3 network offers advantages such as a small number of parameters, high speed, and moderate depth, making it a common backbone network for medical image segmentation tasks.

[0050] like Figure 2 As shown in the figure, first, the bneck module in MobileNetV3 is transplanted to the encoding part of the Unet network;

[0051] Secondly, based on the Unet encoding part position, bneck modules of different depths are used to extract features. Generally, the deeper the position, the more bneck modules are required, and the more advanced the abstract features extracted.

[0052] Finally, the extracted features are connected to the corresponding positions of the decoder in sequence for feature fusion. During the connection process, it is necessary to ensure that the number of channels and size of the feature map and the corresponding position of the decoder are consistent.

[0053] Foreign object recognition, personnel recognition, fireworks recognition, etc.: Use YOLOV7 for anomaly recognition and detection.

[0054] Yolov7 uses the same preprocessing methods for data input as Yolov5, including mosaic data augmentation, adaptive anchor box calculation, and adaptive image scaling. For example, the mosaic hyperparameter in Yolov7 is 1.0. The code randomly generates a random value between 0 and 1. If the random value is less than the hyperparameter, mosaic is enabled. If the generated random value is less than 0.8, mosaic augmentation is performed on four images. If it is between 0.8 and 1.0, mosaic augmentation is performed on nine images.

[0055] YoloV7 is a further improvement based on yoloV5. Compared with yoloV5, yoloV7 has been optimized in terms of network structure, data enhancement and activation function, which can improve the model's expressiveness and detection accuracy.

[0056] Network structure: yoloV7 has a faster and more efficient network architecture; it adopts a deeper network structure, including more convolutional layers and residual blocks, which can improve the model's expressive power and detection accuracy.

[0057] Data enhancement: YoloV7 has a more effective label assignment method; it introduces more data enhancement methods, such as random cropping and random rotation, which can increase the diversity of training data and improve the robustness of the model.

[0058] Activation function: yoloV7 has a more effective feature integration method; it uses the swish activation function, which can improve the nonlinear expression ability of the model and further improve the detection accuracy.

[0059] Tear detection: K-Means clustering algorithm is used for tear detection. Cluster analysis is performed on the image to obtain a new image with connected domain recoloring. Connected domain marker extraction and analysis are performed to extract abnormal areas with tear characteristics. Combined with the belt area mask, calculations are performed to remove interference from areas outside the belt and realize belt tear anomaly detection.

[0060] The idea of ​​the K-Means algorithm is very simple. For a given sample set, Euclidean distance is used as an indicator to measure the similarity between data objects. The similarity is inversely proportional to the distance between data objects. The greater the similarity, the smaller the distance.

[0061] The core idea of ​​the K-Means algorithm: first randomly select Initial cluster centers ,in, , calculate the remaining data objects and cluster centers Euclidean distance, find the cluster center closest to the target data object , and assign data objects to cluster centers The corresponding cluster is obtained; the average value of the data objects in each cluster is calculated as the new cluster center, and the next iteration is performed until the cluster center no longer changes or the maximum number of iterations is reached;

[0062] The Euclidean distance calculation formula between the data object and the cluster center in the space is:

[0063] ;

[0064] in, For data objects; For the cluster centers; is the dimension of the data object; and They are and No. attribute values;

[0065] The sum of squared errors for the entire dataset The calculation formula is:

[0066] ;

[0067] in, The size of indicates the quality of the clustering result; is the number of clusters.

[0068] The steps of the K-Means clustering algorithm are essentially the model optimization process of the EM algorithm (Expectation-Maximization algorithm, EM). The specific steps are as follows:

[0069] (1) Randomly select k samples as the mean vector of the initial cluster;

[0070] (2) Divide each sample data set into the cluster with the closest distance;

[0071] (3) Update the mean vector of the cluster class according to the cluster to which each sample belongs;

[0072] (4) Repeat steps (2) and (3). When the set number of iterations is reached or the mean vector of the cluster no longer changes, the model is built and the clustering algorithm results are output.

[0073] LSTM models and statistical methods are used to analyze belt data and monitor belt health;

[0074] First, the historical data is preprocessed: interpolation methods (such as time interpolation or forward filling) are used to fill in missing temperature or offset data to ensure the continuity of the time series. Obvious errors are eliminated based on principles (e.g., physical limits) or business rules. Z-score normalization is performed on data of different dimensions, such as temperature and offset. An LSTM model is then used for data analysis and prediction, combining multiple data dimensions, such as belt start / stop status, operating time, temperature, offset, and material flow ratio, to determine the belt's health status.

[0075] LSTM (Long Short-Term Memory) is a special type of recurrent neural network (RNN) designed specifically for processing time series data. Compared to traditional methods, LSTM has the following advantages:

[0076] Memory for long-term dependencies: Through gating mechanisms (input gate, forget gate, output gate), information is selectively retained or discarded to effectively capture long-term change patterns in conveyor belt data.

[0077] Handling nonlinear relationships: Automatically learn complex nonlinear relationships between parameters such as temperature, offset, and flow rate, without the need for manual feature construction.

[0078] Multivariate time series modeling: Supports simultaneous analysis of multiple sensor data (such as temperature, offset, and material flow) to explore the coordinated changes between parameters.

[0079] The present invention is based on a belt health system of a patrol robot; (1) patrol robots, multiple monitoring cameras, sensors and other hardware equipment are deployed on site, and AI algorithms are used to perform real-time analysis to monitor the belt status. The real-time start and stop status, offset coefficient, material flow ratio and other data of the belt are sent to the background server for storage, and abnormal conditions detected are pushed in a timely manner; (2) the real-time data and alarm data stored in the database are read, and the historical data of the belt is modeled and analyzed using LSTM and statistical methods to determine the current health status of the belt. Based on the results of AI recognition detection and data analysis, the present invention provides annotations or explanations for certain data or anomalies to help users understand the reasons behind them and put forward suggestions for improving the performance of the belt conveyor or extending its service life. It solves the problem that the ordinary belt five-protection system cannot clearly show the system effect in a short time.

[0080] Those skilled in the art will understand that the above description is merely a preferred embodiment of the invention and is not intended to limit the invention. Although the invention has been described in detail with reference to the above examples, those skilled in the art can still modify the technical solutions described in the above examples or replace some of the technical features therein with equivalents. Any modifications, equivalent replacements, etc. made within the spirit and principles of the invention shall be included in the scope of protection of the invention. All technical features in this embodiment can be freely combined according to actual needs.

[0081] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A belt health monitoring method based on a patrol robot, characterized by: The specific steps include: Step 1: Deploy patrol robots, multiple surveillance cameras, and sensors to acquire on-site video and sensor data, and perform data preprocessing. Step 2, AI recognition and detection: Use AI algorithms to conduct real-time analysis to detect the belt status, send the belt's real-time start and stop status, offset coefficient, and material flow ratio data to the backend server for storage, and promptly push any abnormal conditions detected; Step 3: Read the real-time data and alarm data stored in the database, use the LSTM model and statistical methods to model and analyze the belt's historical data to determine the current health status of the belt; In step 2, AI recognition and detection includes the Mobile U-Net segmentation model, the YOLO recognition model, and the K-Means clustering; Among them, the Mobile U-Net segmentation model is used for belt deviation detection and material flow ratio detection. The Mobile U-Net network model is used to achieve real-time detection of belt deviation and material flow ratio. YOLO recognition model: used for foreign object recognition, personnel recognition, and fireworks recognition; uses YOLOV7 for anomaly recognition and detection; K-Means clustering: used for tear detection; the K-Means clustering algorithm is used for tear detection and cluster analysis of the image to obtain a new image with recolored connected domains. Connected domain marker extraction and analysis are performed to extract abnormal areas with tear characteristics. Combined with the belt area mask, calculations are performed to remove interference from areas outside the belt and realize belt tear anomaly detection.

2. The belt health monitoring method based on a patrol robot according to claim 1, characterized in that: The Mobile-Unet segmentation model uses a lightweight MobileNetV3 network as its backbone to achieve deep feature extraction and combines it with the Unet network to improve overall segmentation performance. Specifically, it includes: Transplant the bneck module in MobileNetV3 to the encoding part of the Unet network; According to the Unet encoding part, bneck modules of different depths are used to extract features. Generally, the deeper the position, the more bneck modules are required, and the more advanced the abstract features extracted. The extracted features are sequentially connected to the corresponding positions of the decoder for feature fusion. During the connection process, it is necessary to ensure that the number of channels and size of the feature map and the corresponding position of the decoder are consistent.

3. The belt health monitoring method based on a patrol robot according to claim 1, characterized in that: Use K-Means clustering algorithm for tear recognition and detection, specifically including; Randomly select k initial cluster centers C from the data set i , where i≤1≤k, calculate the remaining data objects and the cluster center C i Euclidean distance, find the cluster center C closest to the target data object i , and assign the data objects to the cluster center C i The corresponding cluster; Calculate the average value of the data objects in each cluster as the new cluster center and perform the next iteration until the cluster center no longer changes or the maximum number of iterations is reached; The Euclidean distance calculation formula between the data object and the cluster center in the space is: Among them, X is the data object; C i is the i-th cluster center; m is the dimension of the data object; X j and C ij X and C respectively i The j-th attribute value of ; The formula for calculating the sum of squared errors (SSE) for the entire data set is: Among them, the size of SSE indicates the quality of clustering results; k is the number of clusters.

4. The belt health monitoring method based on a patrol robot according to claim 1, characterized in that: In step 3, the LSTM model and statistical methods are used to analyze the belt data and monitor the belt health. The specific steps include the following: Step 3.1, preprocess the historical data: use interpolation to fill in missing temperature or offset data to ensure the continuity of the time series; Step 3.2: Eliminate obviously erroneous data; Step 3.3, perform Z-score normalization on temperature and offset data of different dimensions; In step 3.4, the LSTM model is used for data analysis and prediction. The belt health status is determined by combining analysis of multiple data dimensions, including the belt start / stop status, operating time, temperature, offset, and material flow ratio.

5. The belt health monitoring method based on a patrol robot according to claim 1, characterized in that: In step 3.2, obviously erroneous data are eliminated based on statistical methods and business rules; the statistical methods include the 3σ principle, and the business rules include physical limit values.

Images