Vision-based weld defect detection method and system

Abstract

The invention discloses a vision-based weld defect detection method and system, and relates to the technical field of weld defect detection, and the method comprises the steps: adjusting welding parameters when scene environment data in a welding region is monitored and obtained to be inconsistent with expectation; after heat distribution data of the welding part is collected, a distribution abnormal value is generated, whether the distribution abnormal value exceeds the expectation or not is judged, if yes, welding defects of the welding part are predicted, and welding process parameters are optimized according to predicted data; after surface defect data of the welding part are recognized and obtained, internal defects of the welding part are detected, and simulation test data are obtained after the performance of the welding part is tested; and taking the surface defect data, the internal defect data and the simulation test data as feedback, and giving a corresponding process optimization scheme through the welding process optimization knowledge graph. The current welding defect is subjected to preliminary defect detection, the detection result can be quickly obtained, and whether the welding position has the welding defect at present or not is verified.

Description

Technical Field

[0001] The present invention relates to the technical field of weld defect detection, and in particular to a vision-based weld defect detection method and system. Background Art

[0002] Weld defect detection is a key step to ensure welding quality. Common weld defects include cracks, pores, solid inclusions, lack of fusion and lack of penetration. There are many detection methods, such as visual inspection (observing surface defects), penetrant testing (detecting defects using liquid penetrability), ultrasonic testing (detecting defects using the principle of ultrasonic propagation and reflection), radiographic testing (detecting defects using the principle of radiographic transmission and absorption), and magnetic particle testing (detecting defects using the principle of force on magnetic particles in a magnetic field).

[0003] Weld defect detection is widely used in various welded components and structures. It aims to timely discover and deal with defects in welds to ensure welding quality and structural safety. Selecting appropriate detection methods and strictly following relevant standards and requirements is the key to ensuring weld quality.

[0004] In the Chinese invention patent with the authorization announcement number CN103134809B, it includes pixel ratio judgment and image shape judgment. Pixel ratio judgment: determine the weld position, collect the weld image, lock a rectangular frame at the weld position of the image, calculate the grayscale value of each pixel point of the image in the rectangular frame, and divide the grayscale value into low-value grayscale of 0-128 and high-value grayscale of 128-255. When the proportion of low-value grayscale of each pixel point in the image exceeds 85%, the weld is defective; when the proportion is less than 85%, image shape judgment is performed; Image shape judgment: perform binary conversion on the collected weld image, perform edge extraction on the converted image, extract the shape of the weld in the image, and judge the shape of the weld through the eight-connected graph algorithm. When the shape of the weld is an irregular shape, the weld is defective.

[0005] Combined with the above application and the contents of the prior art: When welding parts to be welded, welding quality is particularly important. If there are still many welding defects on the surface of the welded parts after the welding process is completed, the subsequent use quality and effect of the welded parts may be greatly affected. Therefore, during and after welding, it is necessary to detect welding defects at the welded parts, so that when welding defects exist, you can choose to re-weld when the welding effect is poor.

[0006] After completing the welding of the welded parts, the welders usually conduct a visual inspection to determine whether there are welding defects. However, the efficiency of this inspection is too low and it is also prone to omissions and misjudgments. Therefore, in order to improve the efficiency and reliability of defect detection, image recognition or machine vision is usually introduced in the existing welding defect detection methods; however, considering that there are many factors that affect the welding quality during the welding process, for example, the environmental conditions in the welding area will also have a certain impact on the welding quality, especially when there are abnormalities in the air flow rate data, humidity change data and air quality data in the welding area, the risk of defects in the weld will further increase. At this time, if welding defect detection is still only performed by vision, it is difficult to play a feedback role, and the risk of welding defects cannot be effectively reduced in the continuous detection process.

[0007] To this end, the present invention provides a vision-based weld defect detection method and system. Summary of the invention

[0008] 1. Technical issues to be resolved In view of the deficiencies of the prior art, the present invention provides a vision-based weld defect detection method and system, which generates distribution anomaly values ​​after collecting the thermal distribution data of the weldment, determines whether the distribution anomaly values ​​exceed expectations, and if so, predicts the welding defects of the weldment, and optimizes the welding process parameters based on the predicted data; after identifying and obtaining the surface defect data of the weldment, detects the internal defects of the weldment, and obtains simulation test data after testing the performance of the weldment; with the surface defect data, internal defect data and simulation test data as feedback, the corresponding process optimization scheme is given by the welding process optimization knowledge map. Preliminary defect detection is performed on the current welding defects, and the detection results can be quickly obtained, thereby solving the technical problems recorded in the background technology.

[0009] (II) Technical solution To achieve the above objectives, the present invention is implemented through the following technical solutions: A vision-based weld defect detection method includes monitoring and acquiring scene environment data in a welding area, and if the abnormality of the welding scene data exceeds expectations, adjusting the welding parameters as scene abnormality compensation; Collect the thermal distribution data of the welded parts, generate distribution abnormality values ​​from the thermal distribution data, and issue an alarm command to the outside when the distribution abnormality value exceeds the expectation; Predict welding defects of welded parts, optimize welding process parameters based on prediction data, obtain surface defect data of welded parts during identification, and generate surface defect identification data set; If the defect value generated by the surface defect recognition number exceeds the defect threshold, the internal defects of the weldment are detected, and the performance of the weldment is tested to obtain simulation test data; Using welding defect related data as feedback data and identifying the significant features and defect patterns that affect welding defects, the welding process optimization knowledge graph gives the corresponding process optimization plan.

[0010] Furthermore, the sensor network monitors the air velocity data, humidity change data and air quality data at different positions and heights in the welding area in real time, and aggregates them to generate a welding scene data set; Generate environmental outliers from welding scene data set , if the environment is abnormal If the preset abnormal threshold is exceeded, a parameter adjustment instruction will be issued to the outside; Furthermore, after receiving the parameter adjustment instruction, the welding material characteristics are determined, and the current and voltage during welding are automatically adjusted by the trained welding parameter control model using the collected real-time scene environment data as input; Furthermore, after entering the welding observation cycle, the thermal image data of the weldment is continuously collected, and the thermal distribution data of the weld is obtained after the thermal image data is identified; Taking the heat distribution data of the welding part as input, the trained distribution anomaly recognition model is used to identify the abnormal temperature area of ​​the welding part and the related temperature distribution data. After summarizing, the temperature anomaly data set is obtained, and the distribution anomaly value is generated from the abnormal data in the temperature anomaly data set. ; Furthermore, after receiving the alarm instruction, the convolutional neural network is trained by historical data to obtain a trained welding defect prediction model; the real-time heat distribution data is used as input, and the trained welding defect prediction model is used to predict the welding defects of the welded parts. If there is a possibility of welding defects, a defect warning instruction is issued to the outside; Furthermore, after receiving the defect warning instruction, the pre-trained ant colony algorithm is used to optimize the current welding process parameters in real time with real-time environmental condition data as input; After the current welding stage is completed, the surface image of the welded part is collected, and the surface defects of the welded part are identified using the trained welding defect detection model with the welded surface image as input, and the surface defect identification data set is obtained by summarizing; Further, the defect recognition data in the surface defect recognition data set is used as input, and the surface welding defects are evaluated by the trained welding defect evaluation model to generate defect values; If the acquired defect value exceeds the defect threshold, a secondary inspection instruction is issued to the outside; Furthermore, after receiving the secondary inspection instruction, laser scanning is used to identify and measure the defects in the weld, and comprehensive weld defect point cloud data is collected. X-ray imaging is used to penetrate the weldment to obtain internal structure image data, which is then aggregated to generate a defect inspection data set; Furthermore, the digital twin model of the welded part is trained and generated based on the weld defect point cloud data, internal structure image data and specification attribute data of the welded part. After the working scene of the welded part is determined, the digital twin model of the welded part is used to simulate and test the reliability of the welded part. Furthermore, the welding defect data detected each time, all welding process parameters, environmental condition data, and simulation test data during the welding process are recorded, and a welding defect management data set is generated after aggregation; Use principal component analysis to identify the process parameters and environmental conditions that mainly affect defect generation, and use the identification results as influencing features. Use influencing features where the correlation between defect data and influencing features exceeds expectations as significant features. Furthermore, the welding defect management data is used as input and cluster analysis algorithm is used in combination with significant features to identify different categories of defect patterns; Taking welding process optimization as the target word, a welding process optimization knowledge graph is pre-built; after extracting the significant features and defect patterns, the corresponding process optimization features are obtained, and the corresponding process optimization plan is determined based on the process optimization features, and the process optimization plan is executed to optimize the current welding process; The vision-based weld defect detection system includes an environmental analysis unit that monitors and obtains scene environmental data in the welding area. If the abnormality of the welding scene data exceeds expectations, the welding parameters are adjusted as scene abnormality compensation; The heat distribution detection unit collects the heat distribution data of the welded parts, generates distribution abnormality values ​​from the heat distribution data, and issues an alarm instruction to the outside when the distribution abnormality value exceeds the expectation; The welding parameter optimization unit predicts the welding defects of the welded parts, optimizes the welding process parameters according to the predicted data, and obtains the surface defect data of the welded parts during identification to generate a surface defect identification data set; The performance test data unit detects internal defects of the welded part if the defect value generated by the surface defect recognition data exceeds the defect threshold, and tests the performance of the welded part to obtain simulation test data; The welding process optimization unit uses welding defect-related data as feedback data, identifies the significant features and defect patterns that affect welding defects, and then gives corresponding process optimization solutions based on the welding process optimization knowledge graph.

[0011] (III) Beneficial effects The present invention provides a vision-based weld defect detection method and system, which has the following beneficial effects:

[0012] 1. Judge the abnormality of the current welding environment based on the environmental abnormality value, adjust the welding environment during the welding process, reduce the impact of external conditions on welding quality, and reduce welding defects.

[0013] 2. Adjust and optimize the welding control parameters in real time to reduce the impact of abnormal welding environment conditions on the welding effect, which can play a compensatory role and improve the welding quality. Based on the acquired thermal distribution data, it can be judged whether the current temperature data distribution is abnormal and whether the current welding state is abnormal.

[0014] 3. Based on the distribution abnormality value, the current welding abnormality degree can be judged, and the current welding control parameters can be adjusted in time to reduce the current welding risk.

[0015] 4. Provide early warning of possible welding risks during the welding process, and adjust the welding process in time to avoid further welding defects; perform preliminary defect detection on the current welding defects through image acquisition and recognition, and quickly obtain the test results to verify whether there are welding defects at the welding point.

[0016] 5. Detect welding defects from different angles to improve the reliability of welding defect detection; verify the reliability of the current welding effect based on the test analysis data, verify the current welding defect detection results, and optimize the current welding process and welding technology based on the simulation test data and the current defect detection data as feedback.

[0017] 6. Determine the key factors that have the greatest impact on welding defects. If there are many welding defects at present, use this as feedback to dynamically optimize the current welding process and various control parameters to improve the subsequent welding quality; use welding defect data as feedback to optimize the welding process and reduce the risk of subsequent welding defects. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the process of the weld defect detection method of the present invention; Figure 2 It is a structural schematic diagram of the weld defect detection system of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] See also Figure 1 The present invention provides a vision-based weld defect detection method, comprising: Step 1: Monitor and obtain scene environment data in the welding area. If the abnormality of the welding scene data exceeds expectations, adjust the welding parameters as compensation for the scene abnormality; The step 1 includes the following contents: Step 101: after determining the welding area of ​​the welded parts, a sensor network is arranged in the welding area, and the sensor network monitors the air velocity data, humidity change data and air quality data at different positions and heights in the welding area in real time, and the scene environment data obtained above are aggregated to generate a welding scene data set; Step 102: After setting the corresponding qualified interval for the scene environmental condition parameters in the welding area, the environmental anomaly value is generated from the welding scene data set. , according to the environmental anomaly Analyze the abnormality of the current welding scene in the following way: Perform linear normalization on the concentration of particulate matter, humidity and air velocity in the air, and map the corresponding data values ​​to the interval Then follow the following formula:

[0021] Where: , and is the weight coefficient, and its value falls within , the sum of the three is 1, and the value can be obtained by referring to the hierarchical analysis method; , , The particle concentration, humidity and air velocity at time t, , , are the mean values ​​of particle concentration, humidity and air velocity, respectively. , and They are qualified values ​​for particle concentration, humidity and air velocity respectively; , and is the standard deviation of particle concentration, humidity and air velocity, It is the environmental condition monitoring interval; Based on historical data and management expectations of the degree of environmental abnormality, abnormal thresholds are set in advance; if the environmental abnormal value If the abnormal threshold is exceeded, it means that the current environmental conditions may not be suitable for the current welding process. At this time, a parameter adjustment instruction is sent to the outside. When in use, when welding the welded parts, the environmental conditions in the welding area are collected and monitored in real time, and environmental anomaly values ​​are constructed from the welding scene data. , according to the environmental anomaly Judge the abnormality of the current welding environment. If the current environment is abnormal, the welding environment can be adjusted during the welding process to reduce the impact of external conditions on welding quality and reduce welding defects.

[0022] Step 103: train a convolutional neural network with the labeled sample data to obtain a trained welding parameter control model; after receiving the parameter adjustment instruction, determine the characteristics of the welding material, use the collected real-time scene environment data as input, and use the trained welding parameter control model to automatically adjust the current and voltage during welding to accurately control the heat input of the entire welding process; When using, combine the contents in steps 101 to 103: When the welding environment cannot be adjusted in time, the real-time environmental conditions are used as feedback data to adjust and optimize the welding control parameters in real time, thereby reducing the impact of abnormal welding environmental conditions on the welding effect, playing a compensatory role and improving the welding quality.

[0023] After completing the welding of the welded parts, the welders usually conduct a visual inspection to determine whether there are any welding defects. However, the efficiency of this inspection is too low and it is also prone to omissions and misjudgments. Therefore, in order to improve the efficiency and reliability of defect detection, image recognition or machine vision is usually introduced in the existing welding defect detection methods; however, considering that there are many factors that affect the welding quality during the welding process, for example, the environmental conditions in the welding area will also have a certain impact on the welding quality. In particular, when there are abnormalities in the air flow rate data, humidity change data and air quality data in the welding area, the risk of defects in the weld will further increase. At this time, if welding defect detection is still only performed by vision, it is difficult to play a feedback role, and the risk of welding defects cannot be effectively reduced in the continuous detection process.

[0024] Step 2: Collect the thermal distribution data of the welded parts, generate distribution abnormality values ​​from the thermal distribution data, and issue an alarm instruction to the outside when the distribution abnormality value exceeds expectations; The step 2 includes the following contents: Step 201, multiple image acquisition devices are arranged in the welding area, and the image acquisition devices continuously acquire thermal image data of the welded parts after entering the welding observation cycle, and obtain thermal distribution data of the weld after identifying the thermal image data; The neural convolution network is trained with the labeled sample data to obtain the trained distribution anomaly recognition model; the thermal distribution data of the welding part is used as input, and the trained distribution anomaly recognition model is used to identify the temperature anomaly area of ​​the welding part and the related temperature distribution data, and the temperature anomaly data set is obtained after aggregation; When in use, during the welding process, by monitoring the welding temperature data, it is possible to determine whether the current temperature data distribution is abnormal based on the acquired heat distribution data, and further determine whether the current welding state is abnormal; Step 202: Generate distribution anomaly values ​​from the anomaly data in the temperature anomaly data set , according to the distribution of outliers Analyze the abnormality of heat distribution, including the welding temperature After linear normalization, the corresponding data values ​​are mapped to the interval Then follow the following formula:

[0025] Weight coefficient: , and ; , is the number of sub-regions at the weld of the weldment; For the i The welding temperature in each sub-area, is the mean welding temperature, It is the qualified reference value of welding temperature; Pre-set distribution thresholds based on historical data and expectations for temperature distribution management; If the distribution abnormal value exceeds the preset abnormal threshold, it means that in the current temperature welding process, the temperature distribution of the weldment and its welding point is abnormal, and the risk of welding defects at the welding point is high. At this time, an alarm instruction is issued to the outside; When using, combine the contents in steps 201 and 202: When the current temperature distribution state is abnormal, the distribution abnormality value is constructed by identifying the acquired temperature abnormality data. The current welding abnormality degree can be judged based on the distribution abnormality value. If the abnormality degree is high, it means that the current welding risk is high and there is a greater risk of welding defects. By issuing an alarm command, the current welding control parameters can be adjusted in time to reduce the current welding risk.

[0026] Step 3: predict the welding defects of the welded parts, optimize the welding process parameters according to the predicted data, and obtain the surface defect data of the welded parts during identification to generate a surface defect identification data set; The step three includes the following contents: Step 301: after receiving the alarm instruction, record the heat distribution data under different welding process parameters and environmental conditions, as well as the corresponding welding defect data, summarize them as historical data, train a convolutional neural network with the historical data, and obtain a trained welding defect prediction model; Using real-time heat distribution data as input, the trained welding defect prediction model is used to predict welding defects of welded parts. If welding defects are likely to exist, a defect warning instruction is issued to the outside. When in use, the welding defect prediction model is trained with the support of historical data. When real-time data is used as input, it can warn of possible welding risks in the welding process and adjust the welding process in time to avoid further occurrence of welding defects.

[0027] Step 302: after receiving the defect warning instruction, the pre-trained ant colony algorithm is used to optimize various welding process parameters in the current welding process, such as voltage and current, in real time using the real-time scene environment data as input until the current welding process is completed; The deep learning model is trained by the labeled sample data to obtain a trained welding defect detection model; After the current welding stage is completed, a high-resolution camera is used to collect the surface image of the welded part. After preprocessing the acquired welding surface image data, the welded surface image is used as input, and the trained welding defect detection model is used to identify the surface defects of the weld, such as cracks, pores, and lack of fusion. The acquired surface defect data is summarized to obtain a surface defect recognition data set. When used, combine the contents in steps 301 and 302: After there may be welding risks and the welding process is adaptively adjusted, the current welding process is preliminarily completed. Through image acquisition and recognition, preliminary defect detection is performed on the current welding defects. The detection results can be quickly obtained to verify whether there are welding defects at the welding point.

[0028] Step 4: If the defect value generated by the surface defect recognition number exceeds the defect threshold, the internal defects of the weldment are detected, and the performance of the weldment is tested to obtain simulation test data; The step 4 includes the following contents: Step 401: train a convolutional neural network with the labeled sample data to obtain a trained welding defect evaluation model; use the defect recognition data in the surface defect recognition data set as input, and use the trained welding defect evaluation model to evaluate the surface welding defects and generate defect values; Defect thresholds are pre-set based on historical data and management expectations of the welding process. If the acquired defect value exceeds the defect threshold, it means that there are defects on the welding surface of the welded part, and the severity of the defect exceeds expectations, which may affect the subsequent use of the welded part. At this time, a secondary inspection instruction is issued to the outside. When in use, after the defect identification is initially completed, the defect value is constructed from the defect identification data, and the current defect degree is evaluated based on the defect value. If the current defect degree is too large, it means that there is a welding error, and the current welded part cannot be used anymore, or needs to be re-welded; Step 402: After receiving the secondary inspection instruction, laser scanning is used to identify and measure defects in the weld, such as lack of fusion, slag inclusions and pores, etc., to collect comprehensive weld defect point cloud data, and X-ray imaging is used to penetrate the weldment to obtain internal structure image data; the welding defect point cloud data and the internal structure image data are aggregated to generate a defect inspection data set; When in use, when there are many welding defects on the outside of the welded part, the welding structure inside the welded part is further judged, and the welding defects that may exist inside are completed to complete the current defect detection process; by detecting welding defects at different angles, the reliability of welding defect detection can be improved; Step 403: train and generate a digital twin model of the welded part based on the weld defect point cloud data, the internal structure image data, and the specification attribute data of the welded part. After determining the working scene of the welded part, use the digital twin model of the welded part to simulate and test the reliability of the welded part. After obtaining the simulation test data, verify the reliability of the model based on the simulation test data. If the performance of the welded part can meet expectations, the welding and welding inspection process is completed. When used, combine the contents in steps 401 to 403: On the basis of completing the welding defect detection, a digital twin model of the welded parts is constructed based on various detection data, and the reliability of the welded parts is tested and analyzed. The reliability of the current welding effect can be verified based on the test and analysis data, and the current welding defect detection results can be verified. If the current welding effect is poor, the current welding process and welding technology can be optimized based on the simulation test data and the current defect detection data as feedback.

[0029] Step 5: Using the welding defect related data as feedback data, after identifying the significant features and defect patterns that affect the welding defects, the corresponding process optimization scheme is given by the welding process optimization knowledge map; The step five includes the following contents: Step 501, after completing several defect detections in succession, record the welding defect data of each detection, including surface defect recognition data and internal structure image data, and classify the data according to defect types, such as cracks, pores, and slag inclusions; Record all welding process parameters during welding, such as welding current, voltage, speed, gas flow, etc., record scene environment data, including temperature, humidity, air flow rate and concentration of particulate matter in the air, as well as simulation test data, and generate a welding defect management data set after aggregation; Step 502: Use principal component analysis to identify the process parameters and environmental conditions that mainly affect defect generation, and use the identification results as influencing features, calculate the Pearson correlation coefficient between the defect data and the influencing features, and use the influencing features whose Pearson correlation coefficient values ​​exceed a preset correlation threshold as significant features; When in use, after completing the data collection, the key factors that have the greatest impact on welding defects are determined through principal component analysis and correlation analysis. When there are many welding defects and the welding quality is poor, this can be used as feedback to dynamically optimize the current welding process and various control parameters to improve the subsequent welding quality; Using welding defect management data as input, cluster analysis algorithms are used in combination with significant features to identify different types of defect patterns. For example, it is found that crack defects are mainly concentrated under high current and low temperature conditions, while pore defects mostly appear in low current and high humidity environments. Step 503: Taking welding process optimization as the target word, after deep retrieval and entity relationship construction, a welding process optimization knowledge graph is pre-constructed; After extracting the significant features and defect patterns, the corresponding process optimization features are obtained. According to the correspondence between the process optimization scheme and the process optimization features, the corresponding process optimization scheme is given by the welding process optimization knowledge graph, and the process optimization scheme is executed to optimize the current welding process. When using, combine the contents in steps 501 to 503: When the current welding process needs to be optimized, the welding process optimization knowledge map gives the corresponding process optimization plan based on the identification of the main influencing factors on welding defects. The current welding process is optimized by executing the process optimization plan, and the welding defect data is used as feedback. By optimizing the welding process, the risk of subsequent welding defects can be reduced.

[0030] The Analytic Hierarchy Process (AHP) is a decision-making method that decomposes decision-related elements into levels such as goals, criteria, and plans, and conducts qualitative and quantitative analysis on this basis. It is particularly suitable for dealing with target systems with hierarchical and staggered evaluation indicators, and when the target value is difficult to describe quantitatively, the AHP is an effective decision-making tool.

[0031] The core of the hierarchical analysis method is to decompose the decision problem into multiple levels to form a hierarchical structure, which usually includes the target level, the criterion level, the sub-criterion level and the solution level. By solving the eigenvector of the judgment matrix, the priority weight of each element of each level to an element of the previous level is obtained, and the final weight of each alternative solution to the total goal is recursively merged by the weighted sum method, so as to find the optimal solution.

[0032] The construction method of welding process optimization knowledge graph can refer to the following: Determine the data source: Structured data: such as welding process database, welding equipment parameter table, etc. Unstructured data: such as academic papers, technical documents, expert experience, etc. related to welding process.

[0033] Data collection: Use web crawler technology to crawl data from relevant websites and databases. Preprocess text data, such as denoising, word segmentation, part-of-speech tagging, etc.

[0034] Knowledge representation: Use logical storage solutions such as RDF (Resource Description Framework) to represent entities, attributes, and relationships in the knowledge graph. Determine the core data structure of the knowledge graph, such as entity-attribute-value triples.

[0035] Knowledge storage: Use graph databases (such as Neo4j) or relational databases (such as MySQL) to store knowledge graph data. Consider serialization formats of data, such as JSON-LD, N-TRIPLE, etc., to facilitate data exchange and sharing.

[0036] Knowledge extraction: Extract named entities (such as welding methods, welding materials, welding parameters, etc.) from text data. Extract the relationship between entities, such as "welding method A is suitable for material B". Use NLP technology (such as word2vec algorithm) to vectorize the text to improve the accuracy of knowledge extraction.

[0037] Knowledge fusion: Integrate knowledge from different sources, structures, and formats. Handle knowledge conflicts and redundancies to ensure the consistency and integrity of the knowledge graph.

[0038] Knowledge computing: Perform computing tasks such as entity relationship network analysis and distributed representation learning on the knowledge graph. Use machine learning algorithms to mine and predict knowledge, such as predicting welding quality and optimizing welding parameters.

[0039] Knowledge reasoning: Logical reasoning based on entities and relationships in the knowledge graph. Ontology technology is used to express concepts, attributes and relationships in welding processes, and reasoning and verification are performed based on these ontologies.

[0040] Construct a knowledge graph: Based on the knowledge extracted, integrated and calculated in the above steps, construct a knowledge graph for welding process optimization. Determine the hierarchical structure and relationship network of the knowledge graph, such as a concept hierarchy diagram and an entity relationship diagram.

[0041] Optimize the knowledge graph: Iterate and optimize the knowledge graph, and continuously update and improve the knowledge graph based on new data sources and user feedback. Perform performance evaluation on the knowledge graph, such as calculating query efficiency, accuracy and other indicators to evaluate the practicality and reliability of the knowledge graph.

[0042] See also Figure 2 The present invention provides a vision-based weld defect detection system, comprising: The environmental analysis unit monitors and obtains scene environmental data in the welding area. If the abnormality of the welding scene data exceeds expectations, the welding parameters are adjusted as compensation for the scene abnormality. The heat distribution detection unit collects the heat distribution data of the welded parts, generates distribution abnormality values ​​from the heat distribution data, and issues an alarm instruction to the outside when the distribution abnormality value exceeds the expectation; The welding parameter optimization unit predicts the welding defects of the welded parts, optimizes the welding process parameters according to the predicted data, and obtains the surface defect data of the welded parts during identification to generate a surface defect identification data set; The performance test data unit detects internal defects of the welded part if the defect value generated by the surface defect recognition data exceeds the defect threshold, and tests the performance of the welded part to obtain simulation test data; The welding process optimization unit uses welding defect-related data as feedback data, identifies the significant features and defect patterns that affect welding defects, and then gives corresponding process optimization solutions based on the welding process optimization knowledge graph.

[0043] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0044] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0045] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only some logical function divisions. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0046] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0047] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A vision-based weld defect detection method, characterized in that: include, When the scene environment data acquired by monitoring in the welding area is inconsistent with the expected, the welding parameters are adjusted; After collecting the thermal distribution data of the welded parts, the distribution abnormality value is generated to determine whether the distribution abnormality value exceeds the expectation. If it exceeds, the welding defects of the welded parts are predicted and the welding process parameters are optimized according to the predicted data; After identifying and obtaining the surface defect data of the welded part, the internal defects of the welded part are detected, and the performance of the welded part is tested to obtain simulation test data; With surface defect data, internal defect data and simulation test data as feedback, the corresponding process optimization plan is given by the welding process optimization knowledge graph.

2. The vision-based weld defect detection method according to claim 1, characterized in that: The sensor network monitors the air velocity data, humidity change data and air quality data at different positions and heights in the welding area in real time, and aggregates them to generate a welding scene data set; An environmental anomaly value is generated by the welding scene data set. If the environmental anomaly value exceeds a preset anomaly threshold, a parameter adjustment instruction is issued to the outside.

3. The vision-based weld defect detection method according to claim 2, characterized in that: After receiving the parameter adjustment instruction, the characteristics of the welding material are determined, and the collected real-time scene environment data is used as input. The trained welding parameter control model automatically adjusts the welding parameters during welding.

4. The vision-based weld defect detection method according to claim 3, characterized in that: After entering the welding observation cycle, the thermal image data of the weldment is continuously collected, and the heat distribution data of the weld is obtained after the thermal image data is identified; Taking the heat distribution data of the welding point as input, the trained distribution anomaly recognition model is used to identify the temperature abnormality area and related temperature distribution data of the weldment. After summarizing, the temperature abnormality data set is obtained, and the distribution anomaly value is generated from the abnormal data in the temperature abnormality data set.

5. The vision-based weld defect detection method according to claim 4, characterized in that: The convolutional neural network is trained with historical data to obtain a trained welding defect prediction model; After receiving the alarm command, the real-time heat distribution data is used as input and the trained welding defect prediction model is used to predict the welding defects of the welded parts. If welding defects are likely to exist, a defect warning command is issued to the outside.

6. The vision-based weld defect detection method according to claim 5, characterized in that: After receiving the defect warning instruction, the system uses the real-time scene environment data as input and uses the pre-trained ant colony algorithm to optimize the current welding process parameters in real time; The surface image of the welding part of the welded part is collected as input, and the surface defects of the welding part are identified using the trained welding defect detection model, and the surface defect identification data set is obtained by summarizing.

7. The vision-based weld defect detection method according to claim 6, characterized in that: The defect recognition data in the surface defect recognition data set is used as input, and the trained welding defect evaluation model evaluates the surface welding defects and generates a defect value; if the obtained defect value exceeds the defect threshold, a secondary detection instruction is issued to the outside.

8. The vision-based weld defect detection method according to claim 7, characterized in that: After receiving the secondary inspection instruction, laser scanning is used to identify and measure defects in the weld, comprehensive weld defect point cloud data is collected, and X-ray imaging is used to penetrate the weldment to obtain internal structure image data, which is then summarized to generate a defect detection data set.

9. The vision-based weld defect detection method according to claim 8, characterized in that: The digital twin model of the welded part is trained and generated based on the weld defect point cloud data, internal structure image data and specification attribute data of the welded part. After determining the working scenario of the welded part, the digital twin model of the welded part is used to simulate and test the reliability of the welded part.

10. The vision-based weld defect detection method according to claim 9, characterized in that: The welding defect data, welding process parameters, scene environment data and simulation test data of each inspection are recorded and summarized to generate a welding defect management data set; principal component analysis is used to identify the process parameters and environmental conditions that affect defect generation, and the identification results are used as influencing features, and the influencing features whose correlation between defect data and influencing features exceeds the expected influencing features are used as significant features.

11. The vision-based weld defect detection method according to claim 10, characterized in that: Taking welding defect management data as input, cluster analysis algorithm is used in combination with significant features to identify different types of defect patterns; taking welding process optimization as the target word, a welding process optimization knowledge graph is pre-built; After feature extraction of significant features and defect patterns, the corresponding process optimization features are obtained, and the corresponding process optimization plan is determined based on the process optimization features. The process optimization plan is executed to optimize the current welding process.

12. Vision-based weld defect detection system, characterized by: include, The environmental analysis unit monitors and obtains scene environmental data in the welding area. If the abnormality of the welding scene data exceeds expectations, the welding parameters are adjusted as compensation for the scene abnormality. The heat distribution detection unit collects the heat distribution data of the welded parts, generates distribution abnormality values ​​from the heat distribution data, and issues an alarm instruction to the outside when the distribution abnormality value exceeds the expectation; The welding parameter optimization unit predicts the welding defects of the welded parts, optimizes the welding process parameters according to the predicted data, and obtains the surface defect data of the welded parts during identification to generate a surface defect identification data set; The performance test data unit detects internal defects of the welded part if the defect value generated by the surface defect recognition data exceeds the defect threshold, and tests the performance of the welded part to obtain simulation test data; The welding process optimization unit uses welding defect-related data as feedback data, identifies the significant features and defect patterns that affect welding defects, and then gives corresponding process optimization solutions based on the welding process optimization knowledge graph.

Images