Quality assessment method, system, device and medium for robotic automatic welding

Abstract

This application provides a method, system, equipment, and medium for quality assessment of robotic automated welding. It extracts the temporal statistical characteristics of current fluctuations, the standard deviation of arc voltage, and the two-dimensional morphological features of the molten pool image to obtain multi-dimensional process features. These multi-dimensional process features are compared with the current standard process parameter range to determine the cumulative deviation of each feature within a sliding time window, thereby determining the current welding quality risk level. Based on the original time-series data within the sliding time window, the evolution trend and coupling relationship of each abnormal process feature are determined, identifying the dominant abnormal factor combination and its temporal triggering relationship. Combining the quality risk level, the dominant abnormal factor combination and its temporal triggering relationship, and associating it with keyframe evidence from the corresponding molten pool image, a quality assessment report for robotic automated welding is output. Using the scheme of this application, dynamic risk assessment of robotic automated welding can be performed based on multi-source fusion information.

Description

Technical Field

[0001] This application relates to the field of robotic automatic welding technology, and in particular to a quality assessment method, system, equipment and medium for robotic automatic welding. Background Technology

[0002] Robotic automated welding technology has been widely used in key industrial fields such as automobile manufacturing, pressure vessels, and construction machinery. Welding quality is directly related to the structural safety and service life of products. Therefore, it is crucial to realize online assessment and real-time early warning of welding process quality.

[0003] Currently, quality monitoring in robotic welding processes mainly relies on threshold monitoring of single or a few process parameters. A common method involves collecting electrical signals such as welding current and voltage using sensors and comparing them to preset fixed thresholds or ranges. An alarm is triggered if the limit is exceeded. However, this method has significant limitations: First, welding defects are the result of nonlinear coupling of multiple parameters (electrical, visual, acoustic, etc.). A single-dimensional signal cannot comprehensively and accurately characterize the complex welding state, easily leading to missed or false alarms. Second, static thresholds cannot adapt to dynamic changes during the welding process, such as workpiece thermal deformation and bevel gap fluctuations. This causes alarm systems to either be overly sensitive, generating numerous false alarms, or react slowly, missing early anomalies. Therefore, how to conduct dynamic risk assessment of robotic automated welding based on multi-source fusion information and identify dominant defects has become a challenge for the industry. Summary of the Invention

[0004] Based on this, this application provides a quality assessment method, system, equipment, and medium for robotic automatic welding that performs dynamic risk assessment based on multi-source fusion information.

[0005] In a first aspect, this application provides a quality assessment method for automated robotic welding, comprising the following steps: Collect images of current, arc voltage, and molten pool during automated robotic welding; The time-domain statistical features of current fluctuations, the standard deviation of arc voltage, and the two-dimensional morphological features of the molten pool image are extracted to obtain multi-dimensional process features. The multidimensional process features are compared with the current standard process parameter range in the pre-built process feature-defect association knowledge base to determine the cumulative deviation of each process feature within the sliding time window, thereby determining the current welding quality risk level. When the quality risk level exceeds the preset quality risk threshold, the evolution trend and coupling relationship of each abnormal process feature are determined based on the original time series data within the sliding time window, thereby identifying the combination of dominant abnormal factors and their time series triggering relationship. By combining the quality risk level, the combination of dominant anomaly factors and their temporal triggering relationships, and associating them with key frame evidence from the corresponding molten pool image, a quality assessment report for robotic automated welding is output.

[0006] In some embodiments, extracting the time-domain statistical features of current fluctuations, the standard deviation of arc voltage, and the two-dimensional morphological features of the molten pool image to obtain multidimensional process features specifically includes: The time-domain statistical features of current fluctuations are extracted from the collected welding current time-series signal to obtain the root mean square value of welding current, the standard deviation of welding current, and the peak factor of welding current. The standard deviation of the arc voltage is calculated from the acquired arc voltage time-series signal; Two-dimensional morphological features of the molten pool are extracted from the image frames in the preprocessed molten pool image sequence to obtain the molten pool aspect ratio, molten pool area, and molten pool leading edge angle.

[0007] In some embodiments, the construction of the process feature-defect association knowledge base specifically includes: For specific welding tasks, establish a mapping relationship between welding process parameters, material information, and standard ranges of various process characteristics; Based on process test data or historical qualified production data, the standard parameter range of each process feature in the mapping relationship is determined, thereby completing the construction of the process feature-defect association knowledge base.

[0008] In some embodiments, comparing the multidimensional process features with the current standard process parameter range in a pre-built process feature-defect association knowledge base to determine the cumulative deviation of each process feature within a sliding time window specifically includes: Based on the current welding task information, the standard process parameter ranges corresponding to each dimension of process features are retrieved from the process feature-defect association knowledge base; Determine the instantaneous relative deviation between the characteristic values ​​of each process feature and the corresponding midpoint of the standard process parameter range; Within the sliding time window, the instantaneous relative deviation of each process feature is accumulated to obtain the cumulative deviation of each process feature.

[0009] In some embodiments, determining the quality risk level of the current weld specifically includes: Based on the cumulative deviation of each process characteristic, a comprehensive deviation index characterizing the overall process state deviation is calculated. According to the predefined deviation-risk mapping rules, the comprehensive deviation index is mapped to the corresponding quality risk level.

[0010] In some embodiments, when the quality risk level exceeds a preset quality risk threshold, the evolution trend and coupling relationship of each abnormal process feature are determined based on the original time-series data within the sliding time window: Based on the cumulative deviation of each process feature, abnormal process features exceeding the corresponding cumulative deviation threshold are identified. For each identified anomalous process feature, analyze the evolution trend of its original time series data within the sliding time window; Determine the correlation coefficients between the identified anomalous process features to determine the coupling relationships between them.

[0011] In some embodiments, identifying the dominant combination of anomaly factors and their temporal triggering relationships specifically includes: Each abnormal process feature, its corresponding evolution trend, and the coupling relationship between each abnormal process feature are matched with the pre-stored defect diagnosis rules in the process feature-defect association knowledge base to determine the defect pattern with the highest matching degree. The abnormal process feature pair corresponding to the defect pattern with the highest matching degree is determined as the dominant abnormal factor combination; Based on the original time series data within the sliding time window, the order in which each abnormal process feature in the dominant abnormal factor combination reaches its corresponding cumulative deviation threshold is analyzed to determine its timing triggering relationship.

[0012] Secondly, this application provides a quality assessment system for robotic automated welding, comprising: The data acquisition module is used to acquire current, arc voltage, and molten pool images during automatic welding by the robot. The processing module is used to extract the time-domain statistical features of current fluctuations, the standard deviation of arc voltage, and the two-dimensional morphological features of the molten pool image to obtain multi-dimensional process features. The processing module is also used to compare the multi-dimensional process features with the current standard process parameter range in the pre-built process feature-defect association knowledge base, determine the cumulative deviation of each process feature within the sliding time window, and then determine the quality risk level of the current welding. The processing module is also used to determine the evolution trend and coupling relationship of each abnormal process feature based on the original time series data within the sliding time window when the quality risk level exceeds the preset quality risk threshold, thereby identifying the combination of dominant abnormal factors and their time series triggering relationship. The execution module is used to combine the quality risk level, the combination of dominant anomaly factors and their temporal triggering relationship, and associate the key frame evidence of the corresponding molten pool image to output a quality assessment report for robot automatic welding.

[0013] Thirdly, this application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described quality assessment method for robotic automated welding.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described quality assessment method for automated robotic welding.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The quality assessment method, system, equipment, and medium for robotic automated welding provided in this application firstly acquire current, arc voltage, and molten pool images during robotic automated welding. This step enables multi-source synchronous and accurate acquisition and preprocessing of current, voltage, and visual information, thus providing a reliable raw data foundation with time alignment and anti-interference capabilities for subsequent analysis, fundamentally ensuring the comprehensiveness and accuracy of the information upon which the quality assessment relies. Secondly, the time-domain statistical characteristics of current fluctuations, the standard deviation of arc voltage, and the two-dimensional morphological characteristics of the molten pool image are extracted to obtain multi-dimensional process characteristics. This step can extract characteristics of arc heat from the synchronous data. Key quantitative indicators of input, stability, and molten pool dynamics are used to construct a multi-dimensional feature vector integrating electrical and visual information, providing a calculable comprehensive description of complex welding process states. Subsequently, the multi-dimensional process features are compared with the current standard process parameter range in a pre-built process feature-defect association knowledge base to determine the cumulative deviation of each process feature within a sliding time window, thereby determining the current welding quality risk level. This step enables dynamic comparison between real-time process states and the standard knowledge base, and smooths instantaneous fluctuations by calculating cumulative deviations through a sliding time window, thus quantifying the multi-dimensional feature deviation into a single... The comprehensive risk level, which evolves over time, significantly improves the dynamic adaptability of status assessment and the stability of early warning, reducing false alarms. Then, when the quality risk level exceeds the preset quality risk threshold, based on the original time-series data within the sliding time window, the evolution trend and coupling relationship of each abnormal process feature are determined, thereby identifying the combination of dominant abnormal factors and their temporal triggering relationships. This step can automatically trace the temporal evolution pattern and synergistic relationship of abnormal features after risk discovery and intelligently match them with defect rules in the knowledge base, thus achieving a leap from "discovering anomalies" to "locating core abnormal parameters and inferring the cause of defects," providing a basis for process adjustment. The system provides direct and accurate diagnostic basis for tracing the source of the problem. Finally, by combining the quality risk level, the combination of dominant abnormal factors and their temporal triggering relationship, and associating them with key frame evidence from the corresponding molten pool image, a quality assessment report for robotic automatic welding is output. This step integrates the analysis conclusions and visual evidence into a structured diagnostic report, thereby generating a complete file containing risk level, suspected defect causes, parameter details, and image evidence. This greatly improves the interpretability, traceability, and practicality of the assessment results for on-site process guidance. In summary, the solution proposed in this application can perform dynamic risk assessment for robotic automatic welding based on multi-source fusion information. Attached Figure Description

[0016] Figure 1 This is an exemplary flowchart of a quality assessment method for automated robotic welding according to some embodiments of this application; Figure 2 This is a schematic diagram illustrating an application scenario of a quality assessment data processing system according to some embodiments of this application; Figure 3 This is a flowchart illustrating the determination of cumulative deviation according to some embodiments of this application; Figure 4 This is a structural schematic diagram of a quality assessment system for automated robotic welding, according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device for implementing a quality assessment method for automated robotic welding, according to some embodiments of this application. Detailed Implementation

[0017] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0018] refer to Figure 1 The figure is an exemplary flowchart of a quality assessment method for automated robotic welding according to some embodiments of this application. The quality assessment method for automated robotic welding mainly includes the following steps: In step 101, the current, arc voltage, and molten pool images are collected during the robot's automatic welding process.

[0019] It should be noted that the electrical charge in this application refers to the amount of charge output by the welding power source and flowing through the welding wire and workpiece during the robotic automatic welding process. It is the main source of welding heat input, and its dynamic changes directly affect the formation of the molten pool and the metallurgical quality of the weld. The arc voltage refers to the potential difference between the two ends of the welding arc during the robotic automatic welding process. It is a key electrical parameter characterizing the arc length and energy transfer stability. The molten pool image refers to a two-dimensional image containing visual information of the molten pool and its adjacent area obtained by an image acquisition device during the robotic automatic welding process. It is a direct basis for observing and analyzing the dynamic morphology of the molten pool.

[0020] In practical implementation, the acquisition of current, arc voltage, and molten pool images during robotic automatic welding can be achieved as follows: First, a unified hardware trigger pulse is emitted through the welding start signal of the robot control system or a preset program position point, thereby synchronously starting the data acquisition tasks of the current sensor, voltage sensor, and high-speed industrial camera, ensuring that all data sources have millisecond-level time alignment accuracy; for the acquisition of welding current and arc voltage signals, a closed-loop Hall effect current sensor and a differential isolation voltage sensor can be used, respectively connected in series between the welding power supply output circuit and between the welding torch and the workpiece. The output analog voltage signal is amplified and anti-aliasing low-pass filtered by the signal conditioning circuit before being transmitted to a high-speed data acquisition card. Synchronous analog-to-digital conversion is performed at a sampling rate of not less than 10 kHz, and the generated discrete time-series data stream is timestamped in real time and stored in a... A first-in-first-out data buffer is used. For the acquisition of molten pool images, a high-speed CMOS industrial camera equipped with a narrow-band pass filter is used to strongly suppress strong light interference of a specific wavelength at the center of the arc. The camera is started by receiving the unified hardware trigger pulse mentioned above and is set to shoot at a frame rate of no less than 100 frames per second during the stable arc burning stage to ensure that each frame image completely contains the molten pool, the tail of the arc, and the area of ​​part of the solidified weld bead. The acquired raw image sequence is immediately subjected to median filtering to eliminate the interference of random spatter noise, and is also stamped with a microsecond-level timestamp that is strictly synchronized with the electrical signal, and then stored in an independent image buffer queue. Finally, this step outputs a time-synchronized and continuous welding current timing signal, arc voltage timing signal, and a pre-processed molten pool image sequence. Other methods can also be used in other embodiments, which are not limited here.

[0021] It should be noted that the above steps enable multi-source synchronous and accurate acquisition and preprocessing of current, voltage and visual information, thereby providing a reliable raw data foundation with time alignment and anti-interference for subsequent analysis, fundamentally ensuring the comprehensiveness and accuracy of the information on which quality assessment depends.

[0022] In some embodiments, reference Figure 2 As shown in the figure, this figure is a schematic diagram of the application scenario of the quality assessment data processing system shown in some embodiments of this application. The figure includes three main components: acquisition device, server and data storage device. The acquisition device is responsible for collecting current, voltage and molten pool images during robot automatic welding, and sending the acquired current, voltage and molten pool images to the server through the communication network. The quality assessment data processing system runs in the server, and the server stores the processing results in the data storage device and visualizes them.

[0023] In step 102, the time-domain statistical features of current fluctuations, the standard deviation of arc voltage, and the two-dimensional morphological features of the molten pool image are extracted to obtain multi-dimensional process features.

[0024] In some embodiments, extracting the time-domain statistical features of current fluctuations, the standard deviation of arc voltage, and the two-dimensional morphological features of the molten pool image to obtain multidimensional process features can be achieved through the following steps: The time-domain statistical features of current fluctuations are extracted from the collected welding current time-series signal to obtain the root mean square value of welding current, the standard deviation of welding current, and the peak factor of welding current. The standard deviation of the arc voltage is calculated from the acquired arc voltage time-series signal; Two-dimensional morphological features of the molten pool are extracted from the image frames in the preprocessed molten pool image sequence to obtain the molten pool aspect ratio, molten pool area, and molten pool leading edge angle.

[0025] It should be noted that the time-domain statistical features in this application are mathematical statistics extracted from the time-series waveform of the welding current signal to quantify its fluctuation patterns and distribution characteristics in the time dimension, used to characterize the stability and abnormal patterns of the current; the two-dimensional morphological features are quantitative indicators extracted from the molten pool image to describe the geometric shape, size, and boundary contour characteristics of the molten pool region on a two-dimensional plane, and are important visual information reflecting the stability of the molten pool and the state of the welding metallurgical process; the process features include: root mean square value of welding current, standard deviation of welding current, peak factor of welding current, standard deviation of arc voltage, width-to-length ratio of molten pool, area of ​​molten pool, and leading edge angle of molten pool; the root mean square value of welding current is the root mean square value of the welding current signal within an analysis time window, and is a key feature characterizing the average heat input level of welding during that time period; the welding The current standard deviation is the standard deviation of the welding current signal within an analysis time window. It is a key feature characterizing the fluctuation range of the current around its average value and reflecting the stability of the arc. The welding current peak factor is the ratio of the peak value to the root mean square value of the welding current signal within an analysis time window. It is a key feature characterizing whether there are abnormal instantaneous impacts or spikes in the current signal. The weld pool width-to-length ratio describes the ratio of the width to the length of the minimum bounding rectangle of the weld pool profile. It is a key indicator characterizing the width or length of the weld pool geometry. The weld pool area describes the total number of pixels occupied by the weld pool profile in the image. It is a key indicator characterizing the size of the weld pool. The weld pool leading edge angle describes the angle between the tangent at the center point of the weld pool leading edge and the direction of weld travel. It is a key dynamic indicator characterizing the flow state of the weld pool and its wetting condition with the base metal.

[0026] In specific implementation, the time-domain statistical features of current fluctuations are extracted from the acquired welding current time-series signal to obtain the root mean square (RMS) value, standard deviation, and peak factor of the welding current. This can be achieved in the following way: A fixed analysis time window is set, for example, 200 milliseconds. A complete data segment corresponding to the length of the analysis time window is extracted from the first-in-first-out (FIFO) data buffer storing the welding current time-series signal. The discrete current sample values ​​contained in this data segment are calculated. First, the RMS value is calculated to characterize the average heat input level within the time window. Then, the standard deviation is calculated to quantify the fluctuation amplitude of the current around the mean. Finally, the peak factor, i.e., the ratio of the peak value to the RMS value, is calculated to reflect whether there are abnormal instantaneous impacts or spikes in the current signal. Ultimately, the calculated RMS value, standard deviation, and peak factor of the welding current are output as the three time-domain statistical features extracted from the welding current time-series signal. Other methods can also be used in other embodiments, and are not limited here.

[0027] In specific implementation, the standard deviation of the arc voltage can be calculated from the acquired arc voltage time series signal in the following way: using an analysis time window of the same length as the extracted current feature, voltage data segments within the same time period are synchronously extracted from the first-in-first-out data buffer storing the arc voltage time series signal; statistical calculations are performed on the discrete voltage sample values ​​contained in the voltage data segment, and its standard deviation is directly calculated. The obtained arc voltage standard deviation directly characterizes the degree of fluctuation of the arc length within the analysis time window, and is used as the key feature output extracted from the arc voltage time series signal. Other methods can also be used in other embodiments, which are not limited here.

[0028] In specific implementation, the two-dimensional morphological features of the molten pool are extracted from the image frames in the pre-processed molten pool image sequence to obtain the molten pool width-to-length ratio, molten pool area, and molten pool leading edge angle. This can be achieved in the following way: First, a molten pool image frame that has undergone median filtering preprocessing and is aligned with the center time of the current and voltage analysis time window is selected as the processing object. Second, the molten pool image frame is binarized and segmented using the maximum inter-class variance method or the adaptive threshold method to separate the bright molten pool area from the dark base material background. Subsequently, contour searching is performed on the binary image to extract the surface features. The connected component with the largest product is taken as the molten pool outline. Based on the molten pool outline, the ratio of the width to the length of its smallest bounding rectangle is calculated as the molten pool width-to-length ratio feature. The total number of pixels contained within the molten pool outline is calculated as the molten pool area feature. The angle between the tangent at the center point of the molten pool outline and the direction of weld travel is calculated as the molten pool leading edge angle feature. Finally, the molten pool width-to-length ratio, molten pool area, and molten pool leading edge angle are output as three two-dimensional morphological features extracted from the molten pool image. Other methods can also be used in other embodiments, which are not limited here.

[0029] It should be noted that the above steps can extract key quantitative indicators characterizing arc heat input, stability, and molten pool dynamics from synchronous data, thereby constructing a multi-dimensional feature vector that integrates electrical and visual information, providing a calculable comprehensive description for a complete and detailed characterization of complex welding process states.

[0030] In step 103, the multidimensional process features are compared with the current standard process parameter range in the pre-built process feature-defect association knowledge base to determine the cumulative deviation of each process feature within the sliding time window, thereby determining the current welding quality risk level.

[0031] In some embodiments, the construction of the process feature-defect association knowledge base can be achieved by the following steps: For specific welding tasks, establish a mapping relationship between welding process parameters, material information, and standard ranges of various process characteristics; Based on process test data or historical qualified production data, the standard parameter range of each process feature in the mapping relationship is determined, thereby completing the construction of the process feature-defect association knowledge base.

[0032] It should be noted that the process feature-defect association knowledge base in this application is a structured data and rule base. Its core function is to store the standard process feature parameter range under specific welding process conditions, as well as the association rules between different defect modes and abnormal process feature combinations, providing comparison benchmarks and diagnostic knowledge for online quality assessment and defect tracing.

[0033] In specific implementation, for a specific welding task, establishing a mapping relationship between welding process parameters, material information, and standard ranges of various process characteristics can be achieved in the following way: First, define a process code to uniquely identify a welding task. This code is generated by combining key process parameters such as base material type, base material thickness, welding material type, shielding gas composition, joint type, and welding method. Then, in the database of the process characteristic-defect association knowledge base, create an independent record for each process code. Each record has reserved structured fields to store the standard parameter range of each dimension of the multi-dimensional process characteristics corresponding to that process code. The standard parameter range includes at least the standard lower limit, standard upper limit, and standard median of the characteristic. In this way, a clear mapping relationship is established from specific welding task conditions to a set of quantifiable evaluation standards. Other methods can also be used in other embodiments, which are not limited here.

[0034] In specific implementation, based on process test data or historical qualified production data, the standard parameter range of each dimension of the process feature in the mapping relationship is determined, thereby completing the construction of the process feature-defect association knowledge base. This can be achieved in the following way: For any welding task represented by any process code in the knowledge base, multiple sets of historical welding process data are collected under the condition that the process parameters are set correctly and the welding quality is judged to be qualified by non-destructive testing; for each set of historical data, the standardized multidimensional process feature vector corresponding to the set of data is calculated according to the feature extraction method; subsequently, statistical analysis is performed on the multidimensional process feature vectors of all the collected qualified samples, and for each dimension of the process feature... Features such as the root mean square value of welding current, standard deviation of welding current, peak factor of welding current, standard deviation of arc voltage, width-to-length ratio of weld pool, area of ​​weld pool, and leading edge angle of weld pool are calculated for their mean and standard deviation in qualified samples. Finally, the range of the mean plus or minus a certain number of standard deviations (e.g., three times the standard deviation) is taken as the standard parameter range of the corresponding process feature under the corresponding process code. The lower limit, upper limit, and median of the standard as the mean are filled into the corresponding field of the process code record in the knowledge base, thereby completing the construction and data initialization of the process feature-defect association knowledge base. Other methods can also be used in other embodiments, which are not limited here.

[0035] In some embodiments, reference Figure 3 As shown in the figure, this is a flowchart illustrating the process of determining the cumulative deviation in some embodiments of this application. In this embodiment, the multi-dimensional process features are compared with the current standard process parameter range in the pre-built process feature-defect association knowledge base. The determination of the cumulative deviation of each process feature within the sliding time window can be achieved by the following steps: In step 1031, based on the current welding task information, the standard process parameter range corresponding to each process feature is retrieved from the process feature-defect association knowledge base; In step 1032, the instantaneous relative deviation between the characteristic value of each process feature and the corresponding standard process parameter range value is determined; In step 1033, within the sliding time window, the instantaneous relative deviation of each process feature is cumulatively calculated to obtain the cumulative deviation of each process feature.

[0036] It should be noted that the instantaneous relative deviation in this application is a dimensionless value used to quantify the degree of deviation of the current real-time process feature value relative to its center position in the standard parameter range in the knowledge base at a single moment; it is used to quantify the average performance of the degree of deviation of a certain process feature within a set sliding time window, reflecting the persistence and severity trend of the feature's abnormality.

[0037] In specific implementation, the standard process parameter ranges corresponding to each dimension of the process feature can be retrieved from the process feature-defect association knowledge base based on the current welding task information in the following manner: When performing welding quality assessment, based on the currently executing welding program, the corresponding base material type, base material thickness, welding material type, shielding gas composition, joint type, and welding method information are obtained, and a current process code consistent with the rules when the knowledge base was constructed is generated accordingly; then, the current process code is used as the query keyword to search in the process feature-defect association knowledge base, and all field information stored in the record bound to the code is found and read; finally, the standard lower limit, standard upper limit, and standard median of each dimension of the process feature that strictly corresponds to the current welding task are obtained, namely, the root mean square value of welding current, the standard deviation of welding current, the peak factor of welding current, the standard deviation of arc voltage, the width-to-length ratio of the weld pool, the area of ​​the weld pool, and the leading edge angle of the weld pool. Other methods can also be used in other embodiments, which are not limited here.

[0038] In specific implementation, the instantaneous relative deviation between the feature value of each process feature and the median of the corresponding standard process parameter range can be determined as follows: The feature value of each process feature is subtracted from the median of the corresponding dimension retrieved from the process feature-defect association knowledge base to obtain the absolute difference. Then, this absolute difference is divided by the width of the standard parameter range for the same dimension, where the standard parameter range width is the upper limit of the feature standard minus the lower limit. Through this calculation, a dimensionless numerical value is obtained for each dimension of the multi-dimensional process feature vector. This value represents the instantaneous relative deviation of the process feature within the current analysis time window, quantifying the degree to which the real-time feature value deviates from its standard center. Other methods can also be used in other embodiments, and are not limited here.

[0039] In specific implementation, within the sliding time window, the instantaneous relative deviation of each process feature is cumulatively calculated to obtain the cumulative deviation of each process feature. This can be achieved in the following way: a sliding time window covering multiple consecutive analysis time windows is set, for example, covering the most recent 10 analysis cycles, corresponding to a duration of 2 seconds; for each process feature, at the end of each analysis cycle, its calculated instantaneous relative deviation value is stored in a fixed-length first-in-first-out queue bound to that feature; whenever a new analysis cycle is completed and the queue is updated, the arithmetic mean of all instantaneous relative deviation values ​​in the queue is immediately calculated; this calculated arithmetic mean is defined as the cumulative deviation of that process feature at the current moment, which reflects the average deviation level of that process feature within a recent time window, rather than the instantaneous fluctuation of a single point. Other methods can also be used in other embodiments, which are not limited here.

[0040] In some embodiments, determining the quality risk level of the current weld can be achieved by the following steps: Based on the cumulative deviation of each process characteristic, a comprehensive deviation index characterizing the overall process state deviation is calculated. According to the predefined deviation-risk mapping rules, the comprehensive deviation index is mapped to the corresponding quality risk level.

[0041] It should be noted that the comprehensive deviation index in this application is a single scalar value obtained by weighted summation of the cumulative deviations of various process characteristics. Its core function is to comprehensively characterize the degree of deviation of the overall state of the current welding process from the standard process state. The deviation-risk mapping rule is a rule table that defines the correspondence between the numerical range of the comprehensive deviation index and the discrete quality risk level. Its function is to transform the quantified degree of process deviation into an intuitive, graded quality risk signal. The quality risk level is a discrete level label based on the comprehensive deviation index and determined by the deviation-risk mapping rule. Its function is to intuitively indicate the risk level of the current welding process's quality state.

[0042] In specific implementation, the comprehensive deviation index, which characterizes the overall deviation of the process state, based on the cumulative deviation of each process feature, can be calculated in the following way: A vector is obtained composed of the cumulative deviations corresponding to the seven process features: root mean square value of welding current, standard deviation of welding current, peak factor of welding current, standard deviation of arc voltage, width-to-length ratio of weld pool, area of ​​weld pool, and leading edge angle of weld pool. From the records bound to the current process code in the process feature-defect association knowledge base, the pre-configured weight coefficients for each process feature are read. These weight coefficients reflect the sensitivity of the feature to the welding quality. Then, the cumulative deviation of each process feature is multiplied by its corresponding weight coefficient to obtain a weighted deviation value. Finally, the weighted deviation values ​​of all seven process features are summed to obtain a single scalar value, which is defined as the comprehensive deviation index. This comprehensive deviation index comprehensively characterizes the degree of deviation of the overall state of the current welding process from the standard process range. Other methods can also be used in other embodiments, and are not limited here.

[0043] In specific implementation, mapping the comprehensive deviation index to the corresponding quality risk level according to the predefined deviation-risk mapping rule can be achieved in the following way: a deviation-risk mapping table is pre-stored, which defines the correspondence between the continuous numerical range of the comprehensive deviation index and the discrete quality risk levels. As an example, the mapping rule can be set as follows: when the comprehensive deviation index is less than 0.1, it is mapped to the "normal" level; when the comprehensive deviation index is greater than or equal to 0.1 and less than 0.3, it is mapped to the "low risk" level; when the comprehensive deviation index is greater than or equal to 0.3 and less than 0.6, it is mapped to the "medium risk" level; when the comprehensive deviation index is greater than or equal to 0.6, it is mapped to the "high risk" level. After obtaining the comprehensive deviation index, it is determined which of the above preset numerical ranges it falls into by looking up the table, thereby determining and outputting a quality risk level of "normal", "low risk", "medium risk" or "high risk" corresponding to the current welding. Other methods can also be used in other embodiments, which are not limited here.

[0044] It should be noted that the pre-definition of the deviation-risk mapping rule can be implemented in the following way: During the construction phase of the process feature-defect association knowledge base, the definition can be completed based on statistical analysis of a large amount of historical welding process data. Specifically, historical welding process samples covering multiple known quality states such as "qualified," "minor defects present," and "serious defects present" are collected, and the corresponding comprehensive deviation index is calculated for each sample. Subsequently, the distribution pattern of the comprehensive deviation index of samples of various quality states is analyzed, and different risk intervals are divided by setting classification thresholds. For example, the statistical upper limit of the comprehensive deviation index of "qualified" samples, such as the 95th percentile, is used as the threshold for "normal" and "low risk," the upper limit of the typical range of the comprehensive deviation index of "minor defects" samples is used as the threshold for "low risk" and "medium risk," and the lower limit of the typical range of the comprehensive deviation index of "serious defects" samples is used as the threshold for "medium risk" and "high risk." Finally, these determined numerical intervals (such as less than 0.1, [0.1, 0.3), [0.3, ..., ...) are defined. The values ​​of 0.6 (greater than or equal to 0.6) and their corresponding "normal", "low risk", "medium risk" and "high risk" level labels are stored in the corresponding fields of the process feature-defect association knowledge base as a fixed mapping rule, thereby completing the predefinition. Other methods can also be used in other embodiments, which are not limited here.

[0045] It should be noted that the above steps enable dynamic comparison between real-time process status and standard knowledge base, and smooth instantaneous fluctuations by calculating cumulative deviations through sliding time windows. This quantifies multidimensional feature deviations into a comprehensive risk level that evolves over time, significantly improving the dynamic adaptability of status assessment and the stability of early warning, and reducing false alarms.

[0046] In step 104, when the quality risk level exceeds the preset quality risk threshold, the evolution trend and coupling relationship of each abnormal process feature are determined based on the original time series data within the sliding time window, thereby identifying the combination of dominant abnormal factors and their time series triggering relationship.

[0047] It should be noted that the quality risk threshold in this application is a preset quality risk level threshold used to trigger the defect tracing analysis process. Its function is to serve as a decision boundary for determining whether in-depth diagnosis and cause tracing are required. The quality risk threshold can be preset in the following way: during the system parameter configuration stage, the minimum quality risk level for triggering defect tracing analysis is explicitly set to "medium risk". This means that when the quality risk level determined according to the deviation-risk mapping rule is "medium risk" or higher "high risk", it is determined that the quality risk threshold has been exceeded, and the subsequent defect tracing process will be automatically started. This threshold is a fixed logical judgment condition that works directly based on the defined quality risk level. Other methods can also be used in other embodiments, which are not limited here.

[0048] In some embodiments, when the quality risk level exceeds a preset quality risk threshold, determining the evolution trend and coupling relationship of each abnormal process feature based on the original time-series data within the sliding time window can be achieved through the following steps: Based on the cumulative deviation of each process feature, abnormal process features exceeding the corresponding cumulative deviation threshold are identified. For each identified anomalous process feature, analyze the evolution trend of its original time series data within the sliding time window; Determine the correlation coefficients between the identified anomalous process features to determine the coupling relationships between them.

[0049] It should be noted that the evolution trend in this application describes the overall direction of the numerical change of abnormal process features within a time window, such as a continuous increase. The coupling relationship describes the statistical correlation between multiple abnormal process features in terms of fluctuation changes. The combined effect of the two is to reveal the dynamic pattern of the occurrence, development and interaction of anomalies.

[0050] In specific implementation, the corresponding cumulative deviation threshold can be implemented in the following way: when constructing the process feature-defect association knowledge base, for each process feature, a comprehensive setting is made based on its historical qualified data statistical distribution and process expert experience; for example, the threshold can be set as an empirical proportion of the cumulative deviation statistical value of the feature in qualified samples, such as the mean plus two standard deviations, or an empirical value can be directly set, such as 0.2 or 0.25; finally, the individual cumulative deviation threshold determined for each feature is stored as an inherent attribute of the feature, along with the standard parameter range, weight coefficient, etc., in the record of the corresponding process code in the process feature-defect association knowledge base, for online evaluation and comparison. Other methods can also be used in other embodiments, which are not limited here.

[0051] In specific implementation, when the quality risk level exceeds the preset quality risk threshold, the identification of abnormal process features exceeding the corresponding cumulative deviation threshold based on the cumulative deviation of each process feature can be achieved in the following way: when the quality risk level is determined to exceed the preset quality risk threshold, the source tracing analysis is triggered; the cumulative deviation corresponding to the seven process features of the currently calculated root mean square value of welding current, standard deviation of welding current, peak factor of welding current, standard deviation of arc voltage, width-to-length ratio of weld pool, area of ​​weld pool, and leading edge angle of weld pool are read; the cumulative deviation of each process feature is compared with the cumulative deviation threshold preset in the process feature-defect association knowledge base and bound to the current process code; all process features with cumulative deviation greater than their corresponding cumulative deviation threshold are filtered out and marked as abnormal process features that need to be analyzed in depth. Other methods can also be used in other embodiments, which are not limited here.

[0052] In specific implementation, for each identified abnormal process feature, the analysis of the evolution trend of its original time series data within the sliding time window can be achieved in the following way: For each identified abnormal process feature, backtracking is performed to extract all original time series data points corresponding to the feature within the entire sliding time window length, for example, the most recent 2 seconds. The original time series data points are the unstandardized original feature values ​​calculated for the feature within each analysis time window, such as the actual root mean square value sequence of welding current. Subsequently, linear trend analysis is performed on the original time series data sequence, for example, using the least squares method for linear fitting, to calculate the slope of the sequence as a function of time. Based on the sign and magnitude of the slope, the evolution trend of the abnormal process feature within the sliding time window is determined, and it is qualitatively described as "continuously rising", "continuously falling", or "basically stable". This evolution trend describes whether the corresponding abnormality is aggravated, mitigated, or maintained. Other methods can also be used in other embodiments, which are not limited here.

[0053] In specific implementation, determining the correlation coefficients between the identified abnormal process features to determine the coupling relationship between them can be achieved in the following way: within the same sliding time window as the previous steps, acquire the original time series data sequences corresponding to all identified abnormal process features to form a set of multivariate time series; calculate the Pearson correlation coefficient between any two abnormal process feature sequences in this set of time series to obtain a correlation coefficient matrix; by analyzing this matrix, identify feature pairs whose absolute correlation coefficient values ​​are greater than a preset strong correlation threshold, such as 0.7; these strongly correlated feature pairs are considered to have a statistically significant coupling relationship, that is, their abnormal fluctuations are synergistic in time and may be driven by a common underlying process abnormality cause. This coupling relationship is an important basis for subsequent defect pattern matching and tracing. Other methods can also be used in other embodiments, which are not limited here.

[0054] In some embodiments, identifying the dominant combination of anomaly factors and their temporal triggering relationships can be achieved through the following steps: Each abnormal process feature, its corresponding evolution trend, and the coupling relationship between each abnormal process feature are matched with the pre-stored defect diagnosis rules in the process feature-defect association knowledge base to determine the defect pattern with the highest matching degree. The abnormal process feature pair corresponding to the defect pattern with the highest matching degree is determined as the dominant abnormal factor combination; Based on the original time series data within the sliding time window, the order in which each abnormal process feature in the dominant abnormal factor combination reaches its corresponding cumulative deviation threshold is analyzed to determine its timing triggering relationship.

[0055] It should be noted that the dominant abnormal factor combination in this application is a set of core abnormal features and their values ​​that play a major or decisive role in the current quality risk. It is determined based on the association rules between the identified multiple abnormal process features and specific defect patterns. Its function is to accurately locate the key process parameters that lead to suspected defects. The time-series triggering relationship describes the order in which the abnormal states (exceeding the cumulative deviation threshold) of each core abnormal feature in the dominant abnormal factor combination occur on the time axis. Its function is to reveal the possible causal chain or dynamic development logic within the quality abnormal event.

[0056] In specific implementation, the defect diagnosis rules in the process feature-defect association knowledge base can be set in the following way: during the construction or maintenance phase of the knowledge base, an independent diagnosis rule is created for each typical welding defect pattern that needs to be identified, such as porosity, incomplete penetration, and undercut; each rule clearly defines the typical abnormal process feature combination that leads to the defect, the expected evolution trend of each feature, and the expected coupling relationship between features; for example, the diagnosis rule for the "porosity" defect can be set as follows: the abnormal process feature combination must include an increase in the "standard deviation of arc voltage" and an abnormal increase in the "molten pool area", both of which have an "increasing" evolution trend and a "strong positive coupling relationship" between them; these rules are based on the process data analysis of a large number of defect samples, the principles of welding metallurgy, and the experience of experts in the field, and are stored in the process feature-defect association knowledge base in a structured form, associated with the process code or existing as general rules. Other methods can also be used in other embodiments, which are not limited here.

[0057] In specific implementation, each abnormal process feature, its corresponding evolution trend, and the coupling relationship between each abnormal process feature are matched with the pre-stored defect diagnosis rules in the process feature-defect association knowledge base to determine the defect pattern with the highest matching degree. This can be achieved in the following way: First, each abnormal process feature identified in the current step, the evolution trend description of each abnormal process feature (e.g., "continuously rising"), and the list of identified strongly coupled feature pairs are combined into a complete "current abnormal scenario description"; then, all defect diagnosis rules related to the current process in the knowledge base are traversed, and the "current abnormal scenario description" is matched with each rule. The abnormal feature combinations, trends, and coupling relationships defined in the rules are compared item by item. The matching degree can be quantified by calculating the overlap. For example, the proportion of occurrence of abnormal features defined in the rules in the current abnormal process features, the consistency ratio of trend description, and the consistency ratio of coupling relationship are assigned different weights and then weighted and summed to obtain a matching degree score. Finally, the defect diagnosis rule with the highest matching degree score is selected, and its corresponding defect type is determined as the most likely defect mode, that is, the defect mode with the highest matching degree. Other methods can also be used in other embodiments, which are not limited here.

[0058] In specific implementation, determining the abnormal process feature pair corresponding to the defect pattern with the highest matching degree as the dominant abnormal factor combination can be achieved in the following way: After determining the defect pattern with the highest matching degree, directly read the "typical abnormal process feature combination" explicitly listed in the defect diagnosis rule; this combination is a subset of key features that lead to the defect pattern, such as "arc voltage standard deviation" and "molten pool area"; then, extract the corresponding process features that have been marked as abnormal in the current real-time data, that is, their feature names and current feature values, to form a specific, instantiated feature set; this feature set is formally determined as the dominant abnormal factor combination of this quality anomaly event, which indicates the core parameters that lead to the current high-risk level and suspected defects. Other methods can also be used in other embodiments, which are not limited here.

[0059] In specific implementation, based on the original time-series data within the sliding time window, the order in which each abnormal process feature in the dominant abnormal factor combination reaches its corresponding cumulative deviation threshold is analyzed to determine its temporal triggering relationship. This can be achieved in the following way: For each process feature in the dominant abnormal factor combination, backtrack and query the sequence of instantaneous relative deviation values ​​of each process feature changing with time throughout the entire sliding time window, such as within the last 2 seconds; precisely find the time point in this sequence when the instantaneous relative deviation value first exceeds its individual cumulative deviation threshold, i.e., the threshold used to identify the abnormal feature; compare the time points when all features in the dominant abnormal factor combination first exceed the limit, and sort them according to time order to obtain a time sequence list; for example, "arc voltage standard deviation" first exceeds the limit at time T1, and "molten pool area" subsequently exceeds the limit at time T2 (T2 > T1); based on this time sequence, generate a natural language description or structured representation of the temporal triggering relationship, such as "feature A precedes feature B [several milliseconds]". "An anomaly has occurred"—this relationship helps in understanding the dynamic development process and possible causal chains of anomaly events. Other methods can also be used in other embodiments, which are not limited here.

[0060] It should be noted that the above steps can automatically trace the temporal evolution pattern and collaborative relationship of abnormal features after a risk is discovered, and intelligently match them with the defect rules in the knowledge base, thereby achieving a leap from "discovering an anomaly" to "locating the core abnormal parameters and inferring the cause of the defect", providing a direct and accurate source tracing and diagnostic basis for process adjustment.

[0061] In step 105, the quality risk level, the combination of dominant anomaly factors and their temporal triggering relationship are combined with the key frame evidence of the corresponding molten pool image to output a quality assessment report for robot automatic welding.

[0062] In some embodiments, the quality assessment report for automated robotic welding can be generated by combining the quality risk level, the combination of dominant anomaly factors and their temporal triggering relationships, and associating them with keyframe evidence from the corresponding molten pool image, using the following steps: The system collects the current welding process's quality risk level, dominant anomaly factor combinations, and their temporal triggering relationships, and extracts associated molten pool image keyframes from the image cache. The collected information is organized and populated according to a predefined structured report template; Generate and output the final quality assessment report for robotic automated welding.

[0063] It should be noted that the keyframes of the molten pool image in this application are image frames selected from a continuous sequence of molten pool images that correspond to key time points of quality anomaly events, such as the start of the anomaly or the current state. Their purpose is to provide direct, visual evidence that is synchronized with the parameter anomaly, for the purpose of assisting verification and intuitive judgment.

[0064] In specific implementation, the collection of the current welding process's quality risk level, dominant anomaly factor combination, and their temporal triggering relationships, and the extraction of associated molten pool image keyframes from the image cache, can be achieved in the following way: Read the current quality risk level output by the quality risk level determination step, the dominant anomaly factor combination output by the defect tracing step (including specific feature names, such as "arc voltage standard deviation" and its current value, standard range, and deviation), and the temporal triggering relationship description (such as "feature A is anomaly 300 milliseconds earlier than feature B"). Simultaneously, to provide visual evidence, based on the timestamp of the earliest anomaly feature first exceeding the limit recorded in the temporal triggering relationship, and the current timestamp, retrieve and extract the molten pool image frame closest to these two key time points from the image cache queue storing the pre-processed molten pool image sequence, respectively, as the anomaly initiation keyframe and the current state keyframe. At this point, all text information and image evidence used to generate the report are collected. Other methods can also be used in other embodiments, which are not limited here.

[0065] In specific implementation, the collected information can be organized and populated according to a predefined structured report template in the following way: organize the data in key-value pairs and lists, such as using JSON format; during organization, fill the collected text information and image paths, or Base64 encoded image data, into the corresponding fields of the template: the timestamp field is filled with the current time, the weld_seam_id field is filled with the current welding point identifier, the risk_level field is filled with the quality risk level, the suspected_defect field is filled with the name of the defect pattern with the highest matching degree, the key_abnormal_factors field is filled with a list, each item in the list corresponds to a dominant abnormal factor, and records its name, real-time value, standard range median, and instantaneous relative deviation in detail, the trigger_sequence field is filled with descriptive text of the timing trigger relationship, and the image_evidence field is filled with an object containing index information or embedded data of onset_frame (abnormal start keyframe) and current_frame (current state keyframe); in this way, the loosely collected information is integrated into a data object with a strict internal structure. Other methods can also be used in other embodiments, which are not limited here.

[0066] In specific implementation, the generation and output of the final quality assessment report for robotic automated welding can be achieved in the following way: After completing the data structuring and organization, the structured data object is serialized into an independent file or a database record. At the same time, it is pushed in real time to the host computer software of the manufacturing execution system or the workshop monitoring center through a predefined communication interface, such as MQTT topics, OPC UA methods, or RESTful API. After receiving the report, the host computer software can parse its contents and display them in a graphical interface, such as highlighting the risk level on the workshop dashboard, listing the comparison curves between abnormal parameters and standard values, and displaying key frames of the molten pool image at the beginning of the abnormality and the current state side by side for comparative analysis. This report serves as the final, archiveable, and traceable quality assessment report for robotic automated welding in this welding process. Other methods can also be used in other embodiments, which are not limited here.

[0067] It should be noted that the above steps can integrate the analytical conclusions and visual evidence into a structured diagnostic report, thereby generating a complete file containing risk level, suspected defect cause, parameter details and image evidence, which greatly improves the interpretability, traceability and practicality of the assessment results for on-site process guidance.

[0068] In another aspect, in some embodiments, this application provides a quality assessment system for robotic automated welding, with reference to... Figure 4 The figure is a schematic diagram of the structure of a quality assessment system for automated robotic welding according to some embodiments of this application. The quality assessment system for automated robotic welding includes: a data acquisition module 401, a processing module 402, and an execution module 403, which are described below: The acquisition module 401 in this application is mainly used to acquire the current, arc voltage and molten pool image during automatic welding by the robot. Processing module 402, in this application, is mainly used to extract the time-domain statistical features of current fluctuations, the standard deviation of arc voltage, and the two-dimensional morphological features of the molten pool image to obtain multi-dimensional process features. The processing module 402 described in this application is also used to compare the multi-dimensional process features with the current standard process parameter range in the pre-built process feature-defect association knowledge base, determine the cumulative deviation of each process feature within the sliding time window, and then determine the quality risk level of the current welding. The processing module 402 described in this application is also used to determine the evolution trend and coupling relationship of each abnormal process feature based on the original time series data within the sliding time window when the quality risk level exceeds the preset quality risk threshold, thereby identifying the combination of dominant abnormal factors and their time series triggering relationship. The execution module 403 in this application is mainly used to combine the quality risk level, the combination of dominant abnormal factors and their temporal triggering relationship, and associate the key frame evidence of the corresponding molten pool image to output a quality assessment report for robot automatic welding.

[0069] The modules in the aforementioned quality assessment system for automated robotic welding can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0070] In another embodiment, this application provides a computer device, which may be a server, and its internal structure diagram may be as follows. Figure 5As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores quality assessment data for robotic automated welding. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a quality assessment method for robotic automated welding.

[0071] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0072] In one embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps described in the embodiment of the quality assessment method for automated robotic welding.

[0073] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps described in the embodiment of the quality assessment method for automated robotic welding.

[0074] In one embodiment, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps described in the embodiment of the quality assessment method for automated robotic welding.

[0075] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0076] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0077] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for quality assessment of robotically automated welding, characterized by, Includes the following steps: Collect images of current, arc voltage, and molten pool during automated robotic welding; The time-domain statistical features of current fluctuations, the standard deviation of arc voltage, and the two-dimensional morphological features of the molten pool image are extracted to obtain multi-dimensional process features. The multidimensional process features are compared with the current standard process parameter range in the pre-built process feature-defect association knowledge base to determine the cumulative deviation of each process feature within the sliding time window, thereby determining the current welding quality risk level. When the quality risk level exceeds the preset quality risk threshold, the evolution trend and coupling relationship of each abnormal process feature are determined based on the original time series data within the sliding time window, thereby identifying the combination of dominant abnormal factors and their time series triggering relationship. By combining the quality risk level, the combination of dominant anomaly factors and their temporal triggering relationships, and associating them with key frame evidence from the corresponding molten pool image, a quality assessment report for robotic automated welding is output.

2. The method of claim 1, wherein, By extracting the time-domain statistical features of current fluctuations, the standard deviation of arc voltage, and the two-dimensional morphological features of the molten pool image, multidimensional process features are obtained, including: The time-domain statistical features of current fluctuations are extracted from the collected welding current time-series signal to obtain the root mean square value of welding current, the standard deviation of welding current, and the peak factor of welding current. The standard deviation of the arc voltage is calculated from the acquired arc voltage time-series signal; Two-dimensional morphological features of the molten pool are extracted from the image frames in the preprocessed molten pool image sequence to obtain the molten pool aspect ratio, molten pool area, and molten pool leading edge angle.

3. The method of claim 1, wherein, The construction of the process feature-defect association knowledge base specifically includes: For specific welding tasks, establish a mapping relationship between welding process parameters, material information, and standard ranges of various process characteristics; Based on process test data or historical qualified production data, the standard parameter range of each process feature in the mapping relationship is determined, thereby completing the construction of the process feature-defect association knowledge base.

4. The method of claim 1, wherein, The multidimensional process features are compared with the current standard process parameter range in the pre-built process feature-defect association knowledge base to determine the cumulative deviation of each process feature within the sliding time window. Specifically, this includes: Based on the current welding task information, the standard process parameter ranges corresponding to each dimension of process features are retrieved from the process feature-defect association knowledge base; Determine the instantaneous relative deviation between the characteristic values ​​of each process feature and the corresponding midpoint of the standard process parameter range; Within the sliding time window, the instantaneous relative deviation of each process feature is accumulated to obtain the cumulative deviation of each process feature.

5. The method of claim 1, wherein, Determining the current quality risk level of the welding specifically includes: Based on the cumulative deviation of each process characteristic, a comprehensive deviation index characterizing the overall process state deviation is calculated. According to the predefined deviation-risk mapping rules, the comprehensive deviation index is mapped to the corresponding quality risk level.

6. The method of claim 1, wherein, When the quality risk level exceeds the preset quality risk threshold, based on the original time-series data within the sliding time window, the evolution trend and coupling relationship of each abnormal process feature are determined, specifically including: Based on the cumulative deviation of each process feature, abnormal process features exceeding the corresponding cumulative deviation threshold are identified. For each identified abnormal process feature, analyze the evolution trend of its original time series data within the sliding time window; Determine the correlation coefficient between each identified abnormal process feature to determine the coupling relationship between the abnormal process features.

7. The method of claim 1, wherein, Identifying the dominant abnormal factor combination and its time sequence trigger relationship specifically includes: Matching each abnormal process feature, the corresponding evolution trend, and the coupling relationship between the abnormal process features with the pre-stored defect diagnosis rules in the process feature-defect association knowledge base to determine the defect mode with the highest matching degree; Determining the pair of abnormal process features corresponding to the defect mode with the highest matching degree as the dominant abnormal factor combination; Based on the original time series data within the sliding time window, analyzing the order of each abnormal process feature in the dominant abnormal factor combination reaching its corresponding cumulative deviation threshold to determine its time sequence trigger relationship.

8. A quality assessment system for robotic automatic welding, characterized by, Comprise: The acquisition module is used for collecting current, arc voltage and molten pool image during robot automatic welding; The processing module is used for extracting time domain statistical features of current fluctuation, standard deviation of arc voltage and two-dimensional morphological features of molten pool image respectively to obtain multi-dimensional process features; The processing module is also used for comparing the multi-dimensional process features with the current standard process parameter range in the pre-built process feature-defect association knowledge base, determining the cumulative deviation degree of each dimensional process feature within the sliding time window, and further determining the quality risk level of the current welding; The processing module is also used for determining the evolution trend and coupling relationship of each abnormal process feature based on the original time series data within the sliding time window when the quality risk level exceeds the preset quality risk threshold, and further identifying the dominant abnormal factor combination and its time sequence trigger relationship; The execution module is used for combining the quality risk level, the dominant abnormal factor combination and its time sequence trigger relationship, and associating the key frame evidence of the corresponding molten pool image to output the quality evaluation report of robot automatic welding. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to realize the steps of the quality evaluation method for robot automatic welding in any one of claims 1 to 7.

10. A computer readable storage medium storing a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the quality evaluation method for robot automatic welding as claimed in any one of claims 1 to 7.

Images