Bending detection method and system for cable in fast moving drag chain machine

Abstract

The invention provides a bending detection method and system for a cable in a fast moving drag chain machine. The system comprises a data acquisition and transmission module, a target detection model and an intelligent detection and early warning module. The data acquisition and transmission module comprises tow chain machine monitoring equipment and is mainly used for acquiring real-time videos of a tow chain machine and performing data transmission; the target detection model comprises picture clearness / fuzzy labeling, training and verification of a cable clearness model, cable bending / straight labeling and training and verification of a cable bending detection model, and is mainly used for detecting actual coordinate information of a target cable; and the intelligent detection and early warning module comprises a target detection module and an alarm, and is mainly used for detecting the bending condition of the cable and returning a final detection result. The bending condition of the cable in the fast moving drag chain machine is monitored in real time through the cable bending detection system, the cable bending detection efficiency can be improved, the manual dependence is reduced, and the factory intelligent requirement is met.

Description

Technical Field

[0001] The invention belongs to the field of machine vision and is a bending detection system for cables in a fast-moving drag chain machine based on machine vision. Background Art

[0002] In the cable production process, especially in the quality inspection before leaving the factory, it is crucial to ensure that the cable has good bending tolerance. Cable bending tolerance testing based on drag chain machines is a standard procedure for qualified cables to leave the factory, but the drag chain machine needs to run 24 hours a day during the testing process. During this period, severe bending of the cable will cause irreversible damage to the fixed track of the drag chain machine, causing serious losses. Therefore, it is very important to detect the bending condition of the experimental cable in real time during the operation of the drag chain machine, warn of risks, and replace the bent cable in time. Traditional detection methods mostly use manual monitoring combined with experience judgment, which has problems such as low efficiency, poor timeliness, and low consistency of judgment standards. It is difficult to adapt to the needs of modern large-scale production.

[0003] YOLO (You Only Look Once) is a cutting-edge machine vision technology with many outstanding advantages. First, YOLO innovatively adopts a single prediction mechanism, which can traverse the entire image and quickly locate all targets through a single forward propagation process. This design brings excellent real-time performance and is particularly suitable for scenarios that require high-speed response. Secondly, YOLO optimizes the traditional two-stage detection algorithm, greatly reducing the computational complexity and running time, and improving processing efficiency. Especially when paired with GPU hardware acceleration, it can achieve stable target detection at an extremely high frame rate, taking into account both accuracy and speed, and providing strong support for many artificial intelligence application scenarios. At present, significant results have been achieved in fields with strong real-time requirements such as autonomous driving vehicles, drone flight control, and real-time video surveillance. However, no examples of its application in the field of cable bending detection have been seen in existing patents.

[0004] The invention is based on machine vision technology. The system uses a high-resolution camera to capture the dynamic bending state of the cable in the application scenario of the drag chain machine, and uses image processing algorithms and deep learning technology to analyze the image in real time to accurately determine whether the cable is bent. This helps manufacturers improve product quality and reduce the risk of equipment downtime and maintenance costs caused by cable bending failures. Summary of the invention

[0005] The present invention aims to provide a method and system for detecting the bending of cables in a fast-moving drag chain machine based on machine vision, so as to realize a system for automatically monitoring and detecting whether the cables are bent, thereby reducing dependence on human resources and improving the accuracy and efficiency of detection.

[0006] To achieve this purpose, the present invention adopts the following technical solution: a method and system for detecting the bending of a cable in a fast moving drag chain machine, the method comprising the following steps:

[0007] Step 1) Building a cable clarity detection model: The video data collected by the drag chain machine monitoring is frame-sliced ​​to build a cable clarity detection dataset, and the dataset is annotated for YOLO model training and verification to achieve cable image clarity detection;

[0008] Step 2) Building a cable bending detection model: Annotate the pictures predicted to be clear in step 1 to build a cable bending detection dataset for training and verification of the YOLO model to achieve automatic detection of whether the cable is bent;

[0009] Step 3) Fast cascade detection of cable bending: Cascade the models trained in step 1 and step 2, and input only the clear picture output by the cable clear detection model into the cable bending detection model for detection;

[0010] Step 4) Intelligent detection and early warning: Deploy the trained cable clarity detection model and cable bending detection model to the detection module in the monitoring system to detect whether the cable is bent and return the final detection result;

[0011] Furthermore, the video data annotation of the cable clarity detection model described in step 1) constructing the target detection model specifically includes:

[0012] Divide the video data transmitted by the drag chain machine monitoring device into multiple pictures frame by frame;

[0013] Use the Labelme tool to draw two kinds of rectangular frame labels, clear and vague, respectively. That is, the drag chain machine in the picture is clear and the drag chain machine in the picture is fuzzy.

[0014] The labeled image data is divided into training set, validation set and test set according to the ratio of 7:2:1.

[0015] Furthermore, the cable clarity detection model training and verification described in step 1) constructing the target detection model specifically includes:

[0016] Load the cable clarity detection model configuration file, set the number of training times to 300, the image resolution to 640, and the IDs corresponding to the clear and vague labels to 0 and 1 respectively;

[0017] Train the YOLOv8 object detection model, verify the model's mAP50, mAP50-95, precision, and recall indicators at the end of each round of training, and select the model weight with the highest mAP50 indicator to retain.

[0018] Furthermore, the image data annotation of the cable bending detection model described in step 2) building the target detection model specifically includes:

[0019] Use the Labelme tool to draw two rectangular box labels, straight and curved, for the clear data set predicted by the cable clear detection model, i.e., the two labels of the straight cable and the curved cable in the picture;

[0020] The labeled image data is divided into training set, validation set and test set according to the ratio of 7:2:1.

[0021] Furthermore, step 2) constructing the cable bending detection model training and verification in the target detection model specifically includes:

[0022] Load the configuration file of the cable bending detection model, set the number of training times to 300, the image resolution to 640, and the IDs corresponding to the straight and curved labels to 0 and 1 respectively;

[0023] Train the YOLOv8 object detection model, verify the model's mAP50, mAP50-95, precision, and recall indicators at the end of each round of training, and select the model weight with the highest mAP50 indicator to retain.

[0024] Furthermore, in step 3), the cascade steps of the cable clarity detection model and the cable bending detection model are as follows:

[0025] First, load the trained cable clarity detection model and cable bending detection model. The cable clarity model cuts the input video frame and identifies the image label. If the image label is "clear", the image is positioned and cropped in a rectangular frame and then input into the cable bending detection model to complete the subsequent detection.

[0026] Furthermore, the specific process of step 4) building an intelligent detection and early warning module is as follows:

[0027] Deploy the trained cable clarity detection model and cable bending detection model to the intelligence and early warning module;

[0028] Count is periodically initialized to 0

[0029] Load the cable clarity detection model to detect the clarity of the cut frame images in real time;

[0030] If the cable clarity detection model detects that the label is clear, the image size is cropped to 550×1024 and then input into the cable bending detection model, otherwise the output is "image blur";

[0031] Load the cable bending detection model to detect the cable bending condition in real time;

[0032] If the cable bending detection model detects a label of straight, then the output is "the cable is not bent", otherwise Count is incremented by 1. If Count / number of clear images in the clip is > 30%, the alarm starts to sound, otherwise the output is "the cable in this video clip is normal".

[0033] Compared with the prior art, the present invention has the following advantages:

[0034] Improve detection efficiency and accuracy: The present invention can use machine vision to automatically identify whether the cable has entered the identification area, and perform real-time bending detection on the cables that have entered the detection area, and finally perform cable bending assessment. The application of this technology significantly improves the accuracy and efficiency of the detection process, ensuring real-time and accurate assessment of cable bending detection.

[0035] Reduced reliance on human resources: The present invention reduces the need for manual monitoring through an automated detection mechanism, thereby reducing human resource consumption and related costs. This is of great significance for improving factory operating efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of a bending detection system for a cable in a fast moving drag chain machine provided by a specific embodiment of the present invention;

[0037] Figure 2 It is an example of the cable clarity / model picture predicted by the cable clarity detection model;

[0038] Figure 3 is an example of a straight / blurred picture of a cable predicted by the cable bending model; DETAILED DESCRIPTION

[0039] The technical solution of the present invention is further described below with reference to the accompanying drawings and through specific implementation methods.

[0040] like Figure 1 As shown, the present invention proposes a bending detection method and system for cables in a fast-moving drag chain machine based on machine vision, which specifically includes the following steps:

[0041] Step 1) Building a cable clarity detection model: The video data collected by the drag chain machine monitoring is frame-sliced ​​to build a cable clarity detection dataset, and the dataset is annotated for YOLO model training and verification to achieve cable image clarity detection;

[0042] Step 2) Building a cable bending detection model: Annotate the pictures predicted to be clear in step 1 to build a cable bending detection dataset for training and verification of the YOLO model to achieve automatic detection of whether the cable is bent;

[0043] Step 3) Fast cascade detection of cable bending: Cascade the models trained in step 1 and step 2, and input only the clear picture output by the cable clear detection model into the cable bending detection model for detection;

[0044] Step 4) Intelligent detection and early warning module: deploy the trained cable clarity detection model and cable bending detection model to the detection module in the monitoring system to detect whether the cable is bent and return the final detection result;

[0045] Specific as Figure 1 As shown above, step 1) constructing the target detection model, the cable clarity detection model video data annotation specifically includes:

[0046] Divide the video data transmitted by the drag chain machine monitoring device into multiple pictures frame by frame;

[0047] Use the Labelme tool to draw two kinds of rectangular frame labels, clear and vague, respectively. That is, the drag chain machine in the picture is clear and the drag chain machine in the picture is fuzzy.

[0048] The labeled image data is divided into training set, validation set and test set according to the ratio of 7:2:1.

[0049] Load the configuration file of the cable clarity detection model, set the number of training times to 300, the image resolution to 640, the confidence to 0.5, and the IDs corresponding to the clear and vague labels to 0 and 1 respectively;

[0050] Train the cable clarity detection model, verify the model's mAP50, mAP50-95, precision, and recall indicators at the end of each round of training, and select the model weight with the highest mAP50 indicator to retain.

[0051] Load the trained model with the highest mAP50 weight and make a simple prediction on the cable image. Figure 2 For the predicted result, the image includes a rectangular box, label and confidence. Figure 2 The above is a clear picture, labeled "clear", with a confidence level of 0.81. Figure 2 The following is a blurred image with a confidence level of 0.82.

[0052] Specifically, the image data annotation operation of the cable bending detection model in step 2) building the target detection model specifically includes:

[0053] Use the Labelme tool to draw two types of rectangular frame labels, straight and curved, respectively. That is, the two labels for the straight cable and the curved cable in the picture. The pictures with cable bending radius exceeding the factory setting value are marked as "curved".

[0054] The labeled image data is divided into training set, validation set and test set according to the ratio of 7:2:1.

[0055] Load the configuration file of the cable bending detection model, set the number of training times to 300, the image resolution to 640, the confidence level to 0.5, and the IDs corresponding to the straight and curved labels to 0 and 1 respectively;

[0056] Train the cable bending detection model, verify the model's mAP50, mAP50-95, precision, and recall indicators at the end of each round of training, and select the model weight with the highest mAP50 indicator to retain.

[0057] Load the trained model weight with the highest mAP50 and make a simple prediction on the clear cable picture. Figure 3 For the predicted result, the image includes a rectangular box, label and confidence. Figure 3 The image above shows a straight cable with the label "straight" and a confidence level of 0.92. Figure 2 Below is a picture of a bent cable with a confidence level of 0.91

[0058] Specifically, in step 3), the cascade steps of the cable clarity detection model and the cable bending detection model are as follows:

[0059] First, load the trained cable clarity detection model and cable bending detection model. The cable clarity model cuts the input video frame and identifies the image label. If the image label is "clear", the image is located and extracted by rectangular frame coordinates and converted into RGB image and cropped (the cropped image ratio is 550×1024) and then input into the cable bending detection model to complete the subsequent detection.

[0060] like Figure 1 As shown in the lower right corner, step 4) builds the intelligent detection and early warning module process as follows:

[0061] Deploy the trained cable clarity detection model and cable bending detection model to the intelligence and early warning module;

[0062] Take a video clip for detection in real-time video;

[0063] Count is periodically initialized to 0

[0064] Load the cable clarity detection model to detect the clarity of the cut frame images in real time;

[0065] If the cable clarity detection model detects that the label is clear, it obtains the coordinate information of the rectangular box in the image and converts the video into an RGB image. The image size is then converted to 550×1024 and input into the cable bending detection model. Otherwise, the output is "image blur".

[0066] Load the cable bending detection model to detect the cable bending condition in real time;

[0067] If the cable bending detection model detects a label of straight, then the output is "the cable is not bent", otherwise Count is incremented by 1. If Count / number of clear images in the clip is > 30%, the alarm starts to sound, otherwise the output is "the cable in this video clip is normal".

Claims

1. A method and system for detecting cable bending in a fast moving drag chain machine, characterized in that: The following steps are involved: Step 1) Building a cable clarity detection model: The video data collected by the drag chain machine monitoring equipment is frame-cut to build a cable clarity detection dataset, and the dataset is annotated for YOLO model training and verification to achieve cable image clarity detection; Step 2) Building a cable bending detection model: Annotate the pictures predicted to be clear in step 1 to build a cable bending detection dataset for training and verification of the YOLO model to achieve automatic detection of whether the cable is bent; Step 3) Fast cascade detection of cable bending: Cascade the models trained in step 1 and step 2, and input only the clear picture output by the cable clear detection model into the cable bending detection model for detection; Step 4) Intelligent detection and early warning: The trained cable clarity detection model and cable bending detection model are deployed to the detection module in the intelligent detection and early warning module to detect whether the cable is bent and return the final detection result.

2. A method and system for detecting cable bending in a fast moving drag chain machine according to claim 1, characterized in that: The video data annotation operation of step 1) constructing the cable clarity detection model in the target detection model specifically includes: Divide the video data transmitted by the drag chain machine monitoring device into multiple pictures frame by frame; Use the Labelme tool to draw two kinds of rectangular frame labels, clear and vague, respectively. That is, the drag chain machine in the picture is clear and the drag chain machine in the picture is fuzzy. The labeled image data is divided into training set, validation set and test set according to a certain ratio.

3. A method and system for detecting cable bending in a fast moving drag chain machine as claimed in claim 1, characterized in that: The step 1) constructing the cable clarity detection model training and verification step in the target detection model specifically includes: Load the cable clarity detection model configuration file, set the number of training times to 300, and the IDs corresponding to the clear and vague labels to 0 and 1 respectively; Train the cable clarity detection model, verify the model's mAP50, mAP50-95, precision, and recall indicators at the end of each round of training, and select the model weight with the highest mAP50 indicator to retain.

4. A method and system for detecting cable bending in a fast moving drag chain machine as claimed in claim 1, characterized in that: The image data annotation operation of step 2) constructing the cable bending detection model specifically includes: Use the Labelme tool to draw two rectangular frame labels, straight and curved, for the clear images predicted by the cable clarity model, i.e., two labels for the straight cable and the curved cable in the image. The images with a cable bending radius exceeding the factory setting value are marked as "curved"; The labeled image data is divided into training set, validation set and test set according to a certain ratio.

5. A method and system for detecting cable bending in a fast moving drag chain machine as claimed in claim 1, characterized in that: The step 2) constructing the cable bending detection model training and verification in the target detection model specifically includes: Load the cable bending detection model configuration file, set the number of training times to 300, and the IDs corresponding to the two labels straight and curved are 0 and 1 respectively; Train the cable bending detection model, verify the model's mAP50, mAP50-95, precision, and recall indicators at the end of each round of training, and select the model weight with the highest mAP50 indicator to retain.

6. A method and system for detecting cable bending in a fast moving drag chain machine as claimed in claim 1, characterized in that: Step 3) First, load the trained cable clarity detection model and cable bend detection model. The cable clarity model cuts the input video frames and identifies the image labels. If the image label is "clear", the image is used as input to the cable bend detection model to complete subsequent detection.

7. A method and system for detecting cable bending in a fast moving drag chain machine as claimed in claim 1, characterized in that: The specific process of step 4) constructing the intelligent detection and early warning module is as follows: Deploy the trained cable clarity detection model and cable bending detection model to the intelligence and early warning module; Count is periodically initialized to 0; Load the cable clarity detection model to detect the clarity of the cut frame images in real time; If the cable clarity detection model detects that the label is clear, the image is input into the cable bending detection model, otherwise the output is "image blur"; Load the cable bending detection model to detect the cable bending in real time; If the cable bending detection model detects the label as straight, then the output is "the cable is not bent", otherwise Count is incremented by 1, and if Count / number of clear images in the clip is > 30%, the alarm starts to sound, otherwise the output is "the cable in this video clip is normal".

Images