Intelligent safety management system of digital production line

Abstract

The invention discloses an intelligent safety management system for a digital production line, which comprises a business layer, an algorithm layer and a data layer, and is characterized in that the data layer acquires and stores real-time video data on the digital production line and concern points and button points pre-marked on the real-time video data; concern points and button points corresponding to different action sequences in the action information of the production line equipment are stored; the business layer is connected with production line equipment through a standardized interface, acquires real-time action information of the production line equipment, and classifies the action information of the production line equipment into a head action category or a hand action category; the algorithm layer comprises an intelligent security identification algorithm, and the intelligent security identification algorithm is used for obtaining a human body image according to the real-time video data. According to action classification, different identifications of the intelligent security identification algorithm can be realized, and the method is more targeted. Targeted compliance identification and safety identification are realized, the identification effect is improved, and the safety of a digital production line is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent management software for production lines, and particularly to an intelligent safety management system for a digital production line. Background Art

[0002] In a typical production line in a digital production line that includes automated loading and unloading, various manual workstations, and mobile AGVs, the current traditional SCADA system (Supervisory Control And Data Acquisition) can collect production data of the production line, but there are the following problems: there is a lack of judgment on the compliance rate and efficiency of the daily operations of the operators in the man-machine workstations and manual workstations, and it is necessary to rely on post-event calculation; it is impossible to achieve personnel action recognition, dangerous object recognition, and preventive supervision that combine security and production management. In the recognition of some operators, only some dangerous human motion information can be recognized, and it is impossible to perform targeted compliance recognition on the motion information of the production line equipment. In particular, there is no effective means for attention recognition. That is, the prior art has the problem of poor recognition efficiency and effect in the compliance recognition of digital production lines. Summary of the Invention

[0003] Therefore, it is necessary to provide an intelligent safety management system for a digital production line to solve the problem of poor recognition efficiency and effect in the compliance recognition of digital production lines in the prior art.

[0004] To achieve the above object, the present invention provides an intelligent safety management system for a digital production line, including a service layer, an algorithm layer, and a data layer. The data layer acquires and stores real-time video data on the digital production line, as well as the focus points and button points pre-marked in the real-time video data, and stores the focus points and button points corresponding to different action sequences in the production line equipment action information. The service layer is connected to the production line equipment through a standardized interface and acquires real-time production line equipment action information, and classifies it into a head action category or a hand action category according to the production line equipment action information. The algorithm layer includes an intelligent safety recognition algorithm, which is used to acquire a human body image according to the real-time video data, and recognize a torso node, an arm node, and a head node according to the human body image. When classified as a head action category, calculate the head direction according to the head node, and then judge whether the head direction is within the preset range of the focus points corresponding to the action sequence in the production line equipment action information. If it is not within the preset range, an abnormal alarm is given. When classified as a hand action category, calculate the arm angle information according to the torso node and the arm node and the button points corresponding to the action sequence in the production line equipment action information. If the arm angle information exceeds the preset range, an abnormal alarm is given.

[0005] Further, the intelligent security recognition algorithm is also used to identify the current action sequence based on the action information of the production line equipment, determine whether the action sequence meets the preset standard action sequence, and issue an abnormal alarm if it does not meet the standard action sequence.

[0006] Further, the intelligent security recognition algorithm includes the YOLOv5n detection algorithm and the Lite-HRNet detection algorithm. The backbone of the YOLOv5n detection algorithm is embedded with a SimAM module. The embedded YOLOv5n detection algorithm is used to implement the recognition of human images. Then, the Conv2D in LiteHRNet is replaced with GSConv, and the FPN upsampling layer of the Lite-HRNet detection algorithm is replaced by CARAFE. The Lite-HRNet detection algorithm is used to implement the recognition of torso nodes, arm nodes, and head nodes, and return the names and coordinates of each node.

[0007] Further, different dangerous areas of different nodes are pre-stored in the data layer. The intelligent security recognition algorithm is also used to determine whether the recognized torso nodes, arm nodes, and head nodes are in the dangerous areas respectively. If they are in the dangerous areas, an abnormal alarm is issued.

[0008] Further, the intelligent security recognition algorithm is also used to calculate the angle between the torso and the horizontal based on the recognized torso nodes. If the angle is greater than the preset value, an abnormal alarm is issued.

[0009] Further, the abnormal alarms are divided into different levels, including system record prompts, sound and light alarms. Different alarm modules are driven according to the abnormal alarm levels for alarming. The alarm modules include voice broadcasts, red light warnings, display screen prompts, etc.

[0010] Further, when an abnormal alarm occurs, the service layer records the product ID information of the current production and establishes a correspondence between the abnormal alarm and the product ID information in the data layer.

[0011] Further, a detection module is also included, which is used to obtain the product ID information of the data layer and detect the product corresponding to the product ID information.

[0012] Further, the production line equipment is lithium battery processing production line equipment.

[0013] Further, a web interaction layer is also included. The web interaction layer obtains the user's login information and displays the video screen. The web interaction layer is also used to obtain the user's marking information to update the operation and re-mark the focus points and button points.

[0014] Different from the prior art, the above technical solution can identify the production line actions through the production line equipment action information, and then perform different action classifications and the corresponding focus points and button points for the actions. According to the action classifications, different identifications of the intelligent safety recognition algorithm can be realized, which is more targeted. The intelligent safety recognition algorithm is used to identify the human body image, torso nodes, arm nodes and head nodes. Through these nodes, the head direction or arm angle can be calculated specifically, so as to realize the recognition of the operator's attention and the recognition of operation safety. Finally, targeted compliance recognition and safety recognition are realized, the recognition effect is improved, and the safety of the digital production line is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the system architecture of the present invention; Figure 2 It is an algorithm architecture diagram of the intelligent safety recognition algorithm of the present invention; Figure 3 It is a system flow logic diagram of the present invention; Figure 4 It is a schematic diagram of the process of the intelligent safety recognition algorithm of the present invention for video image recognition. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] In order to describe in detail the technical content, structural features, achieved purposes and effects of the technical solution, the following will be described in detail in conjunction with specific embodiments and with reference to the accompanying drawings.

[0017] Please refer to Figures 1 to 4 , the present invention provides an intelligent safety management system for a digital production line, including a service layer, an algorithm layer and a data layer. The data layer acquires and stores the real-time video data on the digital production line and the focus points and button points pre-marked in the real-time video data. It should be noted that the real-time video data is collected by a camera with a fixed angle, and the camera is generally fixed beside the production line (abbreviated as the production line) equipment. The focus points here are the points that need the operator to pay attention to, such as Figure 4 as shown in Figure 4 , if the display screen in the upper middle of Figure 4There are multiple buttons of different colors in the exact middle, which are marked as button points. And it stores the focus points and button points corresponding to different action sequences in the production line equipment action information; here, the correspondence means that different action steps in the production line correspond to different focus points or button points. For example, the focus point corresponding to visual inspection is the green dot on the display screen. For example, in the action step of pressing the red button, the corresponding button point is the red button point. The business layer is connected to the production line equipment through a standardized interface and obtains real-time production line equipment action information. Here, the standardized interface means a unified interface, and a certain standard interface (such as CAN bus) is used for connection. Here, the production line equipment action information refers to the action information triggered by the production line equipment. For example, when the red button is triggered, the action information that the red button is triggered will be generated. For example, when visual inspection displays a picture on the display screen, the action information of visual inspection will be generated. It is classified according to the production line equipment action information into a head action category or a hand action category. Here, the classification can be pre-classified. For example, pressing buttons, screwing screws, etc. are hand action categories, while visual inspection is a head action category. This category is subsequently used for identification and differentiation in the algorithm layer.

[0018] The algorithm layer includes an intelligent safety recognition algorithm. A neural network recognition algorithm can be preset in the system. The intelligent safety recognition algorithm is used to obtain a human body image according to real-time video data, such as Figure 4 shown, the yellow square corresponds to the recognized human body image. Then, according to the human body image, the torso node, arm nodes, and head node are recognized, such as Figure 4 shown by the blue dots on the human body image above. When classified as the head action category, the head direction is calculated according to the head node. Specifically, the forward direction pointed by the head can be calculated according to the connection of multiple head nodes, and then it is judged whether the head direction is within the preset range of the focus point corresponding to the action sequence in the production line equipment action information. For example, Figure 4 in, when the production line equipment is in the visual inspection action sequence, it is judged whether the head direction is within the preset range of the green dot on the display screen. If it is not within the preset range, an abnormal alarm is given; here, the preset range can be the vertical distance from the blue focus point to the forward direction of the head. If it is within the preset range, it is considered that visual inspection has been carried out. Otherwise, it is considered that visual inspection has not been carried out, and an abnormal alarm is given. When classified as the hand action category, the button point corresponding to the action sequence in the production line equipment action information is calculated according to the torso node and arm nodes, and the arm angle information is calculated. If the arm angle information exceeds the preset range, an abnormal alarm is given. The arm angle includes the upper arm angle and the lower arm angle. If it is not within the reasonable range, it means that the arm movement is not standard and may cause safety problems.

[0019] The present invention can identify the production line actions through the action information of the production line equipment, and then perform different action classifications and the corresponding focus points and button points for the actions. Different identifications can be achieved according to the action classifications by the intelligent safety identification algorithm, which is more targeted. The intelligent safety identification algorithm is used to identify the human body image, torso nodes, arm nodes and head nodes. Through these nodes, the head direction or arm angle can be calculated specifically, so as to realize the identification of the operator's attention and the identification of operation safety, and finally achieve targeted compliance identification and safety identification, improve the identification effect, and improve the safety of the digital production line.

[0020] Furthermore, the intelligent safety identification algorithm is also used to identify the current action sequence according to the action information of the production line equipment, and judge whether the action sequence meets the preset standard action sequence. If it does not meet the standard action sequence, an abnormal alarm will be issued. The system of the present invention is also used to identify the action sequence and compare it with the preset standard action sequence. If the sequence does not conform to the standard, an abnormal alarm will be triggered. By combining the current state of the equipment with the human body action data, it is judged whether the operation process is compliant. If there are steps out of order, such as incorrect button order before and after or pressing buttons multiple times and other behaviors that violate the operation specifications, the system can detect it in time. This embodiment solves the problem that it is difficult to detect in real time the non-standard operation process in the production line, and improves the monitoring ability of the process integrity and operation sequence.

[0021] To improve the detection effect, as Figure 2 shown, the intelligent safety identification algorithm includes the YOLOv5n detection algorithm and the Lite-HRNet detection algorithm. The backbone of the YOLOv5n detection algorithm is embedded with a SimAM module. The embedded YOLOv5n detection algorithm is used to realize the identification of the human body image, and then the Conv2D in LiteHRNet is replaced with GSConv and the FPN upsampling layer of the Lite-HRNet detection algorithm is replaced by CARAFE. The Lite-HRNet detection algorithm is used to realize the identification of the torso nodes, arm nodes and head nodes, and return the names and coordinates of each node. The present invention adopts the improved YOLOv5n and Lite-HRNet detection algorithms. The SimAM module is embedded in YOLOv5n for human detection, and Lite-HRNet replaces some modules to enhance the key point detection accuracy for identifying the torso, head and arm nodes and their coordinates. During the working process, YOLOv5n is responsible for detecting the whole human body block diagram, and Lite-HRNet performs fine-grained identification on the key parts within the block diagram. By optimizing the structure of the neural network, the detection speed and accuracy can be improved, and the real-time performance can be enhanced.

[0022] When the present invention is implemented, reliability and real-time verification are also required. The logic diagram of the verification is as Figure 3As shown, when the recognition accuracy requirement and timeliness requirement are not met, further optimization is carried out. The front-end data display can be implemented in the way of jsp, css, and html to achieve the display on the web page.

[0023] In some embodiments, different dangerous areas of different nodes are pre-stored in the data layer. The intelligent safety recognition algorithm is further used to respectively determine whether the recognized torso node, arm node, and head node are in the dangerous area, and if so, an abnormal alarm is given. As Figure 4 shown, the right red box is the dangerous area of the head node and the torso node. If the head enters this area, danger may occur, and the system gives an alarm. By presetting and storing the dangerous area information corresponding to different nodes in the data layer of the present invention, after the intelligent safety recognition algorithm recognizes the node coordinates, it can determine whether the operator enters the dangerous area and trigger the corresponding abnormal alarm. It can automatically judge potential safety hazards and improve safety.

[0024] Furthermore, the intelligent safety recognition algorithm is further used to calculate the angle between the torso and the horizontal line according to the recognized torso node. If the angle is greater than the preset value, an abnormal alarm is given. By recognizing the torso node and calculating the angle between it and the horizontal line, if it exceeds the set range, an abnormal alarm is triggered. The present invention can analyze the posture angle of the person, such as whether the torso tilt angle is abnormal, which helps to detect irregular postures such as excessive bending, head-down operation, and fainting. Thus, the safety of the digital production line is improved, and the possibility of injury caused by abnormal postures is reduced.

[0025] In some embodiments, the abnormal alarm is divided into different levels, including system record prompt, sound and light alarm. Different alarm modules are driven to give alarms according to the abnormal alarm level. The alarm modules include voice broadcast, red light warning, display screen prompt, etc. This embodiment realizes the hierarchical strategy of the alarm mechanism. The system can trigger different levels of prompts according to the degree of abnormality, such as recording, voice broadcast, red light warning, display screen reminder, etc. During the working process, the intelligent safety recognition algorithm outputs the alarm level, and the appropriate output form is selected by the alarm module to feedback to the operator. Through diverse prompt means, the effectiveness of the warning is improved.

[0026] Furthermore, when an abnormal alarm occurs, the service layer records the product ID information of the current production and establishes a corresponding relationship between the abnormal alarm and the product ID information in the data layer. When an abnormal alarm occurs, the service layer should record the product ID corresponding to the current production line and establish an association between the product and the abnormality in the data layer. During the operation of the present invention, each alarm not only triggers a prompt but also automatically traces to the specific product, realizing the binding of product and abnormal data. Thus, the associated management of product and quality accountability is realized, providing a basis for subsequent traceability.

[0027] Furthermore, it also includes a detection module, which is used to obtain the product ID information of the data layer and detect the product corresponding to the product ID information. The present invention increases the function of the detection module: the module can obtain the product ID, and based on this detection, the status of the corresponding product. Under normal circumstances, the production line adopts random inspection, and when an alarm occurs, the product when the alarm is triggered is detected in real time, which can further improve the detection effect and avoid the occurrence of abnormal products due to non-standard operations. After identifying the abnormality, the detection module of the present invention performs quality inspection on the corresponding product. It solves the problem that the abnormal behavior of the digital production line cannot be linked to the quality inspection process. Through this function, the efficiency of exception handling is improved, and the accuracy of abnormal product removal is further guaranteed.

[0028] In some embodiments, the production line equipment is a lithium battery processing production line equipment. The present invention can be applied to the lithium battery industry with high risks and high process precision requirements, thereby improving the product qualification rate of lithium battery production and avoiding safety problems in the production line.

[0029] Furthermore, it also includes a web interaction layer, which obtains the user's login information and displays the video screen. The web interaction layer is also used to obtain the user's mark information update operation to re-mark the focus points and button points. The present invention allows users to log in, view videos and re-mark focus points and button points through the web interaction layer. The interaction layer supports the administrator to set up the system after the system is deployed. The user can adjust the marking points according to changes in the camera or production line equipment, etc., to improve adaptability. It solves the problem of fixed pre-marking and difficult to adjust flexibly.

[0030] It should be noted that, although the above embodiments have been described in this article, the patent protection scope of the present invention is not limited thereby. Therefore, based on the innovative concept of the present invention, changes and modifications made to the embodiments described herein, or equivalent structures or equivalent process changes made using the contents of the present invention specification and drawings, directly or indirectly applying the above technical solutions to other related technical fields, are all included in the patent protection scope of the present invention.

Claims

1. An intelligent safety management system for a digital production line, characterized in that: It includes a business layer, an algorithm layer, and a data layer. The data layer acquires and stores the real-time video data on the digital production line, as well as the attention points and button points pre-labeled in the real-time video data, and stores the attention points and button points corresponding to different action sequences in the production line equipment action information. The business layer is connected to the production line equipment through a standardized interface and acquires the real-time production line equipment action information, classifies it according to the production line equipment action information, and divides it into a head action category or a hand action category. The algorithm layer includes an intelligent safety recognition algorithm. The intelligent safety recognition algorithm is used to obtain a human body image according to the real-time video data, and recognize the torso node, arm node, and head node according to the human body image. When classified as the head action category, calculate the head direction according to the head node, and then judge whether the head direction is within the preset range of the attention points corresponding to the action sequence in the production line equipment action information. If it is not within the preset range, an abnormal alarm is given. When classified as the hand action category, calculate according to the torso node and arm node and the button points corresponding to the action sequence in the production line equipment action information, and calculate the arm angle information. If the arm angle information exceeds the preset range, an abnormal alarm is given.

2. The intelligent safety management system for a digital production line according to claim 1, wherein: The intelligent safety recognition algorithm is also used to identify the current action sequence according to the production line equipment action information, and judge whether the action sequence meets the preset standard action sequence. If it does not meet the standard action sequence, an abnormal alarm is given.

3. The intelligent safety management system for a digital production line according to claim 1, characterized in that: The intelligent safety recognition algorithm includes a YOLOv5n detection algorithm and a Lite-HRNet detection algorithm. The backbone of the YOLOv5n detection algorithm is embedded with a SimAM module. The embedded YOLOv5n detection algorithm is used to realize the recognition of the human body image, and then replace the Conv2D in LiteHRNet with GSConv and replace the FPN upsampling layer of the Lite-HRNet detection algorithm through CARAFE. The Lite-HRNet detection algorithm is used to realize the recognition of the torso node, arm node, and head node, and return the names and coordinates of each node.

4. The intelligent safety management system for a digital production line according to claim 1, characterized in that: The data layer pre-stores different dangerous areas for different nodes. The intelligent safety recognition algorithm is also used to judge whether the recognized torso node, arm node, and head node are in the dangerous area respectively. If they are in the dangerous area, an abnormal alarm is given.

5. The intelligent safety management system for a digital production line according to claim 1, characterized in that: The intelligent safety recognition algorithm is also used to calculate the angle between the torso and the horizontal according to the recognized torso node. If the angle is greater than the preset value, an abnormal alarm is given.

6. The intelligent safety management system for a digital production line according to claim 1, characterized in that: The abnormal alarm is divided into different levels, including system record prompt, sound and light alarm. Different alarm modules are driven according to the abnormal alarm level for alarm. The alarm modules include voice broadcast, red light warning, display screen prompt, etc.

7. The intelligent safety management system for a digital production line according to claim 1, characterized in that: When an abnormal alarm occurs, the business layer records the product ID information of the current production and establishes a correspondence between the abnormal alarm and the product ID information in the data layer.

8. The intelligent safety management system of a digital production line according to claim 7, characterized in that: It also includes a detection module for acquiring the product ID information of the data layer and detecting the product corresponding to the product ID information.

9. The intelligent safety management system for a digital production line according to claim 1, characterized in that: The production line equipment is lithium battery processing production line equipment.

10. The intelligent safety management system for a digital production line according to claim 1, wherein: It further includes a web interaction layer which obtains the user's login information and displays the video picture, and the web interaction layer is also used to obtain the user's marking information update operation to re-mark the focus points and button points.

Images