CN-121982398-A - Visual fool-proof monitoring method, device and system for industrial production line and storage medium

CN121982398ACN 121982398 ACN121982398 ACN 121982398ACN-121982398-A

Abstract

The invention relates to a visual foolproof monitoring method, a visual foolproof monitoring device, visual foolproof equipment and a visual foolproof storage medium for an industrial production line. The method comprises the steps of obtaining real-time monitoring images of an industrial production line, carrying out image preprocessing and element extraction on the real-time monitoring images to obtain foolproof monitoring elements, carrying out type identification on the foolproof monitoring elements to generate an error event identification set, carrying out flow comparison and alarm triggering on the error event identification set to obtain hierarchical alarm information, and carrying out defect recording on the basis of the hierarchical alarm information, the error event identification set and the foolproof monitoring elements to obtain a production quality data chain. The invention can realize automatic identification of multiple hidden risks such as incorrect assembly sequence, missing parts and the like, and remarkably reduces the missing detection rate caused by relying on manpower.

Inventors

CAO MENG
CHEN YONGZHOU
QIN SHUYAN

Assignees

广州德程智能科技股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260122

Claims (10)

1. The visual foolproof monitoring method for the industrial production line is characterized by comprising the following steps of: Acquiring a real-time monitoring image of an industrial production line, and carrying out image preprocessing and element extraction on the real-time monitoring image to obtain a fool-proof monitoring element; Performing type recognition on the fool-proof monitoring elements to generate an error event identification set; Performing flow comparison and alarm triggering on the error event identification set to obtain hierarchical alarm information; And carrying out defect recording based on the hierarchical alarm information, the error event identification set and the fool-proof monitoring element to obtain a production quality data chain.
2. The method for visual fool-proof monitoring of industrial production line according to claim 1, wherein the step of obtaining the real-time monitoring image of the industrial production line, performing image preprocessing and element extraction on the real-time monitoring image to obtain fool-proof monitoring elements comprises: Positioning and extracting monitoring area images corresponding to production stations on the industrial production line from the real-time monitoring images; Performing visual identification on each monitoring area image through a preset target detection model to obtain visual characteristic information; Performing element detection on each piece of visual feature information based on a preset element feature rule, and extracting the feature parameters as area elements when the corresponding feature parameters are identified in the visual feature information; Generating an abnormal region mark when the characteristic parameter is not detected; and summarizing all the area elements and the abnormal area marks to obtain the fool-proof monitoring elements.
3. The method for monitoring the visual fool-proofing of the industrial production line according to claim 2, wherein the step of visually recognizing each monitoring area image through a preset target detection model to obtain visual characteristic information comprises the following steps: Inputting the monitoring area image into a feature extraction network of the target detection model, and extracting a primary feature image from the monitoring area image through the feature extraction network; Inputting the primary feature map into a feature fusion network of the target detection model, and carrying out fusion enhancement on the primary feature map through an information interaction path of the feature fusion network to generate a fusion feature map; inputting the fusion feature map into a decoupling detection head of the target detection model; performing object probability calculation on the fusion feature map through the classification branches of the decoupling detection head to obtain target object class probability; carrying out object coordinate recognition on the fusion feature map through a regression branch of the decoupling detection head to obtain a target object position coordinate; and combining the target object category probability and the target object position coordinates to obtain the visual characteristic information.
4. The industrial production line visual fool-proof monitoring method of claim 1, wherein the performing type recognition on the fool-proof monitoring element to generate an error event identification set comprises: traversing each judgment rule in a preset error judgment rule set, and comparing the fool-proof monitoring element with each judgment rule, wherein each judgment rule is associated with an error mark; When the fool-proof monitoring element meets the current judging rule, associating the fool-proof monitoring element with the corresponding error mark; If the judging rule is not satisfied, judging that the fool-proof monitoring element has no error; And summarizing all the error identifications associated after the judgment rules are traversed, and forming the error event identification set.
5. The industrial production line vision foolproof monitoring method of claim 1, wherein the performing flow comparison and alarm triggering on the error event identification set to obtain hierarchical alarm information comprises: performing node matching on each event identifier in the error event identifier set and a process node in a preset process node sequence; When the matched process node is a target node, marking the event identifier as a first error identifier, otherwise marking the event identifier as a second error identifier; Counting the first identification number marked as the first error identification and the second identification number marked as the second error identification; And comparing the first identification number and the second identification number with a preset grading threshold value respectively, triggering corresponding grading alarm according to the obtained comparison result, and generating grading alarm information.
6. The method for monitoring the visual fool-proofing of an industrial production line according to claim 5, wherein the steps of comparing the first number of marks and the second number of marks with a preset classification threshold, triggering a corresponding classification alarm according to the obtained comparison result, and generating the classification alarm information comprise: comparing the identification number with a first threshold value of the preset grading threshold value; If the number of the marks is higher than the first threshold, determining to trigger a first-level alarm to generate first-level alarm information; If the number of the marks is not higher than the first threshold, comparing the second number of the marks with a second threshold of the preset grading threshold; if the second identification number is higher than the second threshold value, judging to trigger a second-level alarm to generate second-level alarm information; if the second identification number is not higher than the second threshold value, judging that the alarm is not triggered currently, and recording the error event identification set as a flow deviation log; and taking the first-stage alarm information, the second-stage alarm information or the flow deviation log as the hierarchical alarm information.
7. The industrial production line vision fool-proofing monitoring method according to claim 1, wherein said performing defect recording based on said hierarchical alarm information, said error event identification set and said fool-proofing monitoring element to obtain a production quality data chain comprises: Performing defect association on the fool-proof monitoring element and the error event identification set to generate an initial defect record set; Marking the priority of the initial defect record set according to the grading alarm information to obtain a priority record set; Classifying all the priority record sets according to the same identification according to a preset production work order identification to form a batch defect data set; And establishing an area index relation for each batch of defect data sets based on all the real-time monitoring images, and carrying out association connection on all the batch of defect data sets according to the area index relation to obtain the production quality data chain.
8. An industrial production line vision foolproof monitoring device, characterized in that it is applied to the industrial production line vision foolproof monitoring method as set forth in any one of the above claims 1-7, comprising: the acquisition module is used for acquiring a real-time monitoring image of the industrial production line, and carrying out image preprocessing and element extraction on the real-time monitoring image to obtain a foolproof monitoring element; The analysis module is used for carrying out type identification on the fool-proof monitoring elements and generating an error event identification set; The association module is used for carrying out flow comparison and alarm triggering on the error event identification set to obtain hierarchical alarm information; And the processing module is used for carrying out defect recording based on the hierarchical alarm information, the error event identification set and the fool-proof monitoring element to obtain a production quality data chain.
9. An industrial production line vision fool-proof monitoring system, which is characterized by comprising: a memory for storing a program; a processor for executing the program to implement the method of any one of claims 1-7.
10. A storage medium having stored thereon computer instructions for causing a computer to perform the method of any one of claims 1 to 7.

Description

Visual fool-proof monitoring method, device and system for industrial production line and storage medium Technical Field The invention relates to the technical field of industrial production line monitoring, in particular to a visual foolproof monitoring method, device and system for an industrial production line and a storage medium. Background Along with the zero defect management and control and real-time intelligent monitoring of the production process, higher requirements are put forward. The existing monitoring method mainly depends on automatic detection or periodic manual inspection of fixed rules, and usually only aims at identifying single and dominant defect types, so that hidden errors such as complex assembly logic, dynamic operation normalization, multi-dimensional operation sequence and the like are difficult to judge in real time and comprehensively. The limitation of the method is that the monitoring dimension is single, the rule is stiff, the self-adaptive production line change cannot be realized, the experience and the attention of inspection personnel are seriously relied on, the hysteresis and the high omission rate exist in the identification of systematic risks such as assembly sequence errors, part neglected loading, tool misuse and the like, and the full-flow and traceable intelligent quality assurance is difficult to realize. Disclosure of Invention The invention mainly aims to provide a cooperative control method and system for the aggregation prediction and peak shaving task instructions of an air source heat pump, which can realize automatic identification of multiple hidden risks such as incorrect assembly sequence, missing parts and the like, and remarkably reduce the missing detection rate caused by relying on manpower. In order to achieve the above object, the present invention provides a visual foolproof monitoring method for an industrial production line, comprising: Acquiring a real-time monitoring image of an industrial production line, and carrying out image preprocessing and element extraction on the real-time monitoring image to obtain a fool-proof monitoring element; Performing type recognition on the fool-proof monitoring elements to generate an error event identification set; Performing flow comparison and alarm triggering on the error event identification set to obtain hierarchical alarm information; And carrying out defect recording based on the hierarchical alarm information, the error event identification set and the fool-proof monitoring element to obtain a production quality data chain. Further, the acquiring the real-time monitoring image of the industrial production line, performing image preprocessing and element extraction on the real-time monitoring image to obtain a foolproof monitoring element, includes: Positioning and extracting monitoring area images corresponding to production stations on the industrial production line from the real-time monitoring images; Performing visual identification on each monitoring area image through a preset target detection model to obtain visual characteristic information; Performing element detection on each piece of visual feature information based on a preset element feature rule, and extracting the feature parameters as area elements when the corresponding feature parameters are identified in the visual feature information; Generating an abnormal region mark when the characteristic parameter is not detected; and summarizing all the area elements and the abnormal area marks to obtain the fool-proof monitoring elements. Further, the visual recognition is performed on each monitoring area image through a preset target detection model to obtain visual characteristic information, which includes: Inputting the monitoring area image into a feature extraction network of the target detection model, and extracting a primary feature image from the monitoring area image through the feature extraction network; Inputting the primary feature map into a feature fusion network of the target detection model, and carrying out fusion enhancement on the primary feature map through an information interaction path of the feature fusion network to generate a fusion feature map; inputting the fusion feature map into a decoupling detection head of the target detection model; performing object probability calculation on the fusion feature map through the classification branches of the decoupling detection head to obtain target object class probability; carrying out object coordinate recognition on the fusion feature map through a regression branch of the decoupling detection head to obtain a target object position coordinate; and combining the target object category probability and the target object position coordinates to obtain the visual characteristic information. Further, the performing type recognition on the fool-proof monitoring element to generate an error event identification set includes: traversing each judgment rule in a preset