Welding information management system-oriented welding data storage and analysis method

Abstract

The invention discloses a welding information management system-oriented welding data storage and analysis method, which relates to the technical field of information data management, and comprises the following steps of: acquiring welding information at each sampling moment in real time based on a preset sampling frequency; for various welding monitoring data, based on the change condition of the welding monitoring data and the frequency spectrum result of the welding monitoring data, the data abnormal time period in the welding process is recognized; performing correlation analysis on each target data exception time period of the target welding monitoring data and each associated data exception time period corresponding to the associated welding monitoring data to obtain an exception evaluation coefficient of each target data exception time period in the welding process; determining a welding evaluation coefficient of the target welding monitoring data in the welding process based on the abnormal evaluation coefficient of each target data abnormal time period; and based on the welding evaluation coefficient of each welding process, classifying and storing the target welding monitoring data in each welding process. The welding data storage and analysis effect can be improved.

Description

Technical Field

[0001] This invention relates to the field of information data management technology, and more specifically to a welding data storage and analysis method for welding information management systems. Background Technology

[0002] The welding process encompasses operational data from welding equipment, such as voltage, current, and temperature; it also involves fundamental physical properties of the welding materials, including thickness and composition. Storing and analyzing welding data through a welding information management system enables goals such as welding quality traceability, process optimization, and anomaly warning.

[0003] Currently, in the process of welding data storage and analysis, dynamic time warping algorithm is used to calculate the similarity of different batches of welding records in time series, and clustering algorithm is used to group and store multiple batches of welding records according to high similarity.

[0004] However, existing methods measure similarity solely based on temporal differences and welding values ​​in welding records, neglecting the inherent relationships between changes in a single dimension of data and changes in other related data. This leads to significant biases in the data grouping and storage results based on similarity metrics, impacting the effectiveness of welding data storage and analysis. Summary of the Invention

[0005] This invention provides a welding data storage and analysis method for welding information management systems, which can improve the effect of welding data storage and analysis.

[0006] A first aspect of this invention provides a welding data storage and analysis method for a welding information management system, comprising: Based on a preset sampling frequency, welding information at each sampling moment during each welding process is acquired in real time, including various welding monitoring data. Based on the changes in welding monitoring data and the spectral results of the welding monitoring data, abnormal data periods during the welding process can be identified. Correlation analysis is performed on the abnormal time periods of each target data in the target welding monitoring data and the abnormal time periods of each associated data in the associated welding monitoring data to obtain the abnormal evaluation coefficient of each abnormal time period of the target data during the welding process; the target welding monitoring data can be any type of welding monitoring data, and the associated welding monitoring data is the welding monitoring data affected by the fluctuation of the target welding monitoring data; Based on the anomaly evaluation coefficient of each target data abnormal period during the welding process, the welding evaluation coefficient of the target welding monitoring data during the welding process is determined; Based on the welding evaluation coefficients of each welding process, the target welding monitoring data of each welding process are classified and stored.

[0007] Furthermore, this invention also proposes identifying abnormal data periods during the welding process based on changes in welding monitoring data and the spectral results of the welding monitoring data, including: Based on the changes in welding monitoring data and the spectral results of welding monitoring data, the abnormal parameters of welding monitoring data at each sampling time are determined; Welding monitoring data with abnormal parameters exceeding a preset abnormal threshold are identified as abnormal monitoring data. The time period consisting of the sampling times corresponding to multiple consecutive abnormal monitoring data is determined as the data abnormal period in the welding process.

[0008] Furthermore, this invention also proposes determining the data anomaly parameters of the welding monitoring data at each sampling time based on the changes in welding monitoring data and the spectral results of the welding monitoring data, including: Based on the changes in welding monitoring data, the time-domain deviation coefficient of the welding monitoring data at each sampling time is determined; the time-domain deviation coefficient is used to characterize the data anomalies obtained by time-domain analysis of the welding monitoring data. Fourier transform was performed on the welding monitoring data to obtain the spectral results of the welding monitoring data; Based on the spectral results of welding monitoring data, determine the frequency resolution of the welding monitoring data; The time-domain deviation coefficients of each welding monitoring data are multiplied by the frequency resolution to obtain the data anomaly parameters of the welding monitoring data at each sampling time.

[0009] Furthermore, this invention also proposes determining the temporal deviation coefficient of the welding monitoring data at each sampling time based on the changes in the welding monitoring data, including: The welding monitoring data at each sampling time were averaged to obtain the welding mean data; The absolute value of the difference between the welding monitoring data and the welding mean data at each sampling time is divided by the welding mean data to obtain the data fluctuation at each sampling time. The temporal deviation coefficient of the welding monitoring data at each sampling time is obtained by subtracting the overall fluctuation range constant from the data fluctuation at each sampling time.

[0010] Furthermore, this invention also proposes to perform correlation analysis on each abnormal period of the target welding monitoring data and the corresponding abnormal periods of the associated welding monitoring data to obtain the anomaly evaluation coefficient for each abnormal period of the target data during the welding process, including: The correlation analysis between the abnormal period of the target data and the corresponding abnormal period of each related data is carried out to obtain the fluctuation similarity between the abnormal period of the target data and the corresponding abnormal period of each related data. Based on the similarity of fluctuations in the abnormal periods of the target data, the anomaly evaluation coefficient for the abnormal periods of the target data is determined.

[0011] Furthermore, this invention also proposes to perform correlation analysis between the target data anomaly period and the corresponding associated data anomaly periods to obtain the fluctuation similarity between the target data anomaly period and the corresponding associated data anomaly periods, including: The difference between the start time of the anomaly in the target data's anomaly period and the start time of the anomaly in the associated data's anomaly period is defined as the anomaly start time difference. The difference between the duration of the abnormality in the target data during an abnormal period and the duration of the abnormality in the associated data during an abnormal period is defined as the abnormal duration difference. Dynamic time warping analysis is performed on the abnormal monitoring data of the target data during abnormal periods and the abnormal monitoring data of the associated data during abnormal periods to obtain the time similarity index between the abnormal periods of the target data and the abnormal periods of the associated data. By utilizing differences in the start time of anomalies, differences in the duration of anomalies, and time period similarity indices, the fluctuation similarity between the anomaly periods of the target data and the anomaly periods of related data can be determined.

[0012] Furthermore, this invention also proposes determining the anomaly evaluation coefficient for each period of the target data anomaly based on the similarity of fluctuations during that period, including: Based on the data anomaly parameters of each anomaly monitoring data in each associated data anomaly period of the target data anomaly period, the similarity weight of each fluctuation similarity is determined respectively; By utilizing the similarity of each fluctuation and the corresponding similarity weight, the anomaly evaluation coefficient for the abnormal period of the target data is determined.

[0013] Furthermore, this invention also proposes determining the welding evaluation coefficient of the target welding monitoring data during the welding process based on the anomaly evaluation coefficient of each target data anomaly period during the welding process, including: Connect the various anomaly evaluation coefficients during the welding process in chronological order to form the target anomaly evaluation curve. The welding monitoring data of each target during the welding process are connected in chronological order to form the target welding monitoring curve; By comparing the target anomaly evaluation curve with the target welding monitoring curve, the welding evaluation coefficient of the target welding monitoring data during the welding process is obtained.

[0014] Furthermore, this invention also proposes to compare the target anomaly evaluation curve with the target welding monitoring curve to obtain the welding evaluation coefficient of the target welding monitoring data during the welding process, including: By comparing the target anomaly evaluation curve with the target welding monitoring curve, the fitting difference parameters between each anomaly node in the target anomaly evaluation curve and the target welding monitoring curve are obtained. Based on the dynamic time warping distance between the target anomaly evaluation curve and the target welding monitoring curve, the curve similarity between the target anomaly evaluation curve and the target welding monitoring curve is determined. Welding evaluation coefficients for target welding monitoring data during the welding process are determined by utilizing curve similarity and various fitting difference parameters.

[0015] Furthermore, this invention also proposes to classify and store the target welding monitoring data for each welding process based on the welding evaluation coefficient of each welding process, including: Based on the welding evaluation coefficients of each welding process, clustering operations are performed on each welding process to obtain clustering results; Based on the clustering results, the target welding monitoring data of each welding process are classified and stored according to a preset storage strategy. The preset storage strategy is to perform lossless compression storage on abnormal monitoring data in the target welding monitoring data, and to perform lossy compression storage on normal monitoring data other than abnormal monitoring data in the target welding monitoring data.

[0016] The present invention has the following beneficial effects: The welding data storage and analysis method for a welding information management system provided in this invention first acquires multiple types of welding monitoring data for each welding process in real time, enabling a comprehensive understanding of welding information. Next, based on changes in the welding monitoring data and spectral results, abnormal data periods during the welding process are identified, accurately pinpointing time periods that may affect welding quality. Then, correlation analysis is performed on the abnormal periods of the target data and related data to obtain anomaly evaluation coefficients. This fully considers the inherent connections between different welding monitoring data, moving beyond isolated views of single-dimensional data changes and more accurately reflecting the impact of data anomalies on the welding process. Subsequently, welding evaluation coefficients are determined based on each anomaly evaluation coefficient, providing an overall assessment of the welding process. Finally, the target welding monitoring data is categorized and stored according to the welding evaluation coefficients. This categorization method combines the correlation between data and the overall welding situation, providing a more scientific and reasonable grouping and storage of welding data compared to existing methods. This effectively avoids storage bias caused by ignoring the inherent connections between data, thereby improving the effectiveness of welding data storage and analysis. Attached Figure Description

[0017] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a welding data storage and analysis method for a welding information management system according to an embodiment of the present invention. Figure 2 This is a schematic flowchart of S200 provided in one embodiment of the present invention; Figure 3 This is a schematic flowchart of S300 provided in one embodiment of the present invention; Figure 4 This is a schematic diagram of the S400 process provided in one embodiment of the present invention. Detailed Implementation

[0019] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a welding data storage and analysis method for a welding information management system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0021] It should be noted that the acquisition, storage, use, and processing of data in the technical solution of this invention all comply with the relevant provisions of laws and regulations.

[0022] It should be noted that in the embodiments of the present invention, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of the present invention. However, they do not mean that the applicant has used or necessarily used the solution.

[0023] In traditional welding data storage and analysis methods, dynamic time warping and clustering algorithms are used to measure the temporal similarity of multiple batches of welding records and achieve grouped storage. This method calculates similarity based solely on the temporal differences and numerical fluctuations of data in a single dimension, without considering the correlation between different welding monitoring data in the time and frequency domains. For example, when welding voltage fluctuates abnormally, the associated welding current and temperature data may simultaneously exhibit frequency domain feature shifts. However, existing methods cannot capture such cross-dimensional correlation features, leading to discrepancies between the identification results of abnormal data periods and the actual process conditions. The resulting erroneous grouping directly affects the accuracy of welding quality traceability, and the mixed storage of abnormal and normal data increases the complexity of subsequent process optimization analysis.

[0024] For example, in the automotive chassis welding process, the welding equipment synchronously collects voltage, current, wire feed speed, and workpiece temperature data through sixteen sensors, with a sampling frequency set to 2000Hz. When the welding voltage oscillates at high frequency within the time interval [12.3s, 14.8s], the associated current data shows amplitude decay in the interval [12.5s, 14.6s], while the workpiece temperature exhibits non-periodic fluctuations in the interval [13.1s, 15.2s]. Existing methods perform independent anomaly detection on the three types of data separately, classifying the voltage anomaly period as [12.0s, 15.0s], the current anomaly period as [12.2s, 14.5s], and the temperature anomaly period as [13.0s, 15.5s]. Because a cross-data-dimensional correlation analysis mechanism is not established, the system classifies three anomaly periods with overlapping time domains but different frequency domain characteristics as independent events, which can easily lead to significant deviations in the data grouping and storage results, affecting the effectiveness of welding data storage and analysis.

[0025] Faced with the aforementioned problems, this invention first recognizes that existing technologies detect anomalies by independently analyzing the temporal differences of single-dimensional data, neglecting the correlation between different welding monitoring data in the time and frequency domains. For example, when voltage anomalies cause frequency domain shifts in current and temperature data, existing methods cannot capture such cross-dimensional features, leading to biased identification and incorrect storage classification of abnormal periods. To address this, this invention considers combining time-domain fluctuations with frequency-domain features, improving the accuracy of abnormal period identification through multi-dimensional data correlation analysis. Furthermore, this invention discovers that abnormal periods in different monitoring data may overlap in time but have inherent differences in correlation, necessitating the establishment of a cross-data-dimensional correlation analysis mechanism. By assessing the fluctuation similarity between abnormal periods, the anomaly characteristics of multi-source data linkage can be accurately reflected. Based on this, this invention proposes fusing time-domain deviation coefficients and spectral results in the anomaly detection stage, while introducing dynamic time warping and period similarity calculation in the correlation analysis. Finally, an anomaly evaluation coefficient is constructed to guide classification and storage, ensuring that the storage structure truly reflects the correlation of process states.

[0026] In this regard, such as Figure 1 As shown, the present invention provides a flowchart of a welding data storage and analysis method for a welding information management system. This welding data storage and analysis method for a welding information management system can be applied to a welding data storage and analysis device or a welding data storage and analysis equipment. The welding data storage and analysis method for a welding information management system may include the following steps S100 to S500: The S100 acquires welding information at each sampling moment during each welding process in real time based on a preset sampling frequency. The welding information includes various welding monitoring data.

[0027] In this step, the preset sampling frequency refers to the pre-set frequency of data acquisition for the welding process, that is, the number of times welding information is collected per unit time. For example, 10 times or 100 times per second. It determines the level of detail and real-time nature of the welding information obtained. The higher the sampling frequency, the more refined the data obtained, and the more timely the changes in the welding process can be reflected.

[0028] The welding process refers to a complete welding operation, from the start of welding to the end of welding. Different welding processes may vary due to factors such as welding process parameters.

[0029] Welding information includes a variety of welding monitoring data related to the welding process, used to comprehensively describe the status and characteristics of the welding process; welding monitoring data is a subdivision of the specific content of welding information, such as voltage, current, temperature, changes in the thickness of welding materials, welding speed, etc., which reflect the operation of the welding process from different aspects.

[0030] Specifically, the sampling frequency is preset, and the data acquisition system is started at the beginning of each welding operation. According to the set frequency, various parameters in the welding process are measured and recorded at each sampling time, such as voltage measurement by voltage sensor and current measurement by current sensor. The collected data is then transmitted to the storage device in real time.

[0031] Based on signal sampling theory, the continuously changing welding process signal can be discretized in time, and data points can be collected at a certain frequency. This ensures that sufficient information reflecting the characteristics of the welding process is obtained while avoiding excessive data volume that would increase the processing burden. The preset sampling frequency needs to comprehensively consider the rate of change in the welding process and the data accuracy requirements. Too low a frequency will lose key information, while too high a frequency will increase cost and computational complexity. By acquiring multi-dimensional welding information in real time, a comprehensive data foundation is provided for subsequent analysis, accurately reflecting the real-time status of the welding process.

[0032] The S200 identifies abnormal periods in the welding process based on changes in welding monitoring data and the spectral results of the welding monitoring data.

[0033] In this step, the changes in welding monitoring data refer to the trend of numerical changes in welding monitoring data between sampling times, such as whether the voltage gradually increases, decreases, or remains stable. By analyzing the changes, the dynamic characteristics of the welding process can be understood.

[0034] The spectral results of welding monitoring data are obtained by converting the welding monitoring data to the frequency domain using spectral analysis methods. Spectral analysis can decompose complex signals in the time domain into components of different frequencies. Through the spectral results, we can understand the distribution of different frequency components in the welding monitoring data, which helps to discover some abnormal features that are difficult to detect in the time domain, such as the presence of interference signals at specific frequencies.

[0035] Abnormal data periods are determined based on changes in welding monitoring data and spectral results, indicating time periods during the welding process where data anomalies occur. Examples include periods of sudden and significant voltage fluctuations or abnormal frequency peaks in the spectrum.

[0036] Specifically, for each type of welding monitoring data, the change in data between adjacent sampling times is first calculated, and its trend is analyzed, such as whether there are abrupt changes or gradual changes. Simultaneously, spectral analysis algorithms such as Fast Fourier Transform are used to convert the time-domain data into the frequency domain, obtaining the spectral results. Based on preset anomaly judgment rules, such as changes exceeding thresholds or the appearance of specific abnormal frequency components in the spectrum, abnormal data periods are identified.

[0037] The data changes reflect the dynamic characteristics of the welding process, and abnormal changes may indicate welding failures or instability. Spectrum analysis utilizes the characteristics of different frequency signals in the frequency domain to detect periodic interference or abnormal frequency components that are difficult to detect in the time domain. By combining time-domain changes and frequency-domain characteristics, abnormal periods in the welding process can be identified more accurately and comprehensively, improving the reliability of anomaly detection.

[0038] S300, perform correlation analysis on the abnormal time periods of each target data of the target welding monitoring data and the abnormal time periods of each associated data of the associated welding monitoring data to obtain the abnormal evaluation coefficient of each abnormal time period of the target data during the welding process; the target welding monitoring data can be any type of welding monitoring data, and the associated welding monitoring data is the welding monitoring data affected by the fluctuation of the target welding monitoring data.

[0039] In this step, the associated welding monitoring data refers to the welding monitoring data that is affected by fluctuations in the target welding monitoring data. For example, when the voltage (target welding monitoring data) fluctuates, it may affect the current (associated welding monitoring data) because there is a certain relationship between voltage and current in the welding circuit.

[0040] Correlation analysis examines the relationship between abnormal periods in the target data and abnormal periods in related data, studying their temporal correspondence, similarity of change trends, etc., in order to determine the degree of influence of fluctuations in the target welding monitoring data on the related welding monitoring data.

[0041] The anomaly evaluation coefficient is an index obtained through correlation analysis to evaluate the degree of anomaly in the target data during an abnormal period. It comprehensively considers the anomaly of the target welding monitoring data itself and its impact on the associated welding monitoring data. The larger the anomaly evaluation coefficient, the more serious the data anomaly during that period.

[0042] Specifically, after selecting the target welding monitoring data, the abnormal time periods for each target data are identified. For each target abnormal time period, the associated welding monitoring data affected by it and the corresponding associated abnormal time periods are determined. A correlation analysis algorithm, such as the Pearson correlation coefficient method, is used to calculate the correlation between the target and the associated abnormal time period data. Combined with expert experience or preset rules, multiple factors are comprehensively considered to obtain the anomaly evaluation coefficient.

[0043] In the welding process, various monitoring data are interconnected; fluctuations in one data point can affect other related data. By analyzing the relationship between the target data and related abnormal periods, the scope and extent of the target data anomaly can be understood. Correlation analysis algorithms can quantify this correlation, and combined with expert experience, the severity of the anomaly can be assessed more reasonably. The resulting anomaly evaluation coefficient more accurately reflects the impact of the target data anomaly on the overall welding process.

[0044] S400 determines the welding evaluation coefficient of the target welding monitoring data during the welding process based on the abnormal evaluation coefficient of each target data abnormal period during the welding process.

[0045] In this step, the welding evaluation coefficient is a comprehensive evaluation index of the abnormality of the target welding monitoring data throughout the entire welding process. It is obtained by comprehensively calculating the abnormality evaluation coefficients of each abnormal period of the target data. For example, a weighted average method can be used to reflect the degree of abnormality of the target welding monitoring data throughout the entire welding process.

[0046] Specifically, the abnormal evaluation coefficients of each target data abnormal period during the welding process are collected. Weights are assigned according to factors such as the importance or duration of each period in the welding process. The weighted average method is used to comprehensively calculate each abnormal evaluation coefficient to obtain the welding evaluation coefficient of the target welding monitoring data.

[0047] The impact of abnormal data during different target periods on the welding process varies, with abnormalities in important periods or those lasting longer having a greater impact on welding quality. By assigning weights and conducting comprehensive calculations, the abnormalities in the target welding monitoring data throughout the entire welding process can be reflected more reasonably. Weighted averaging and other methods can more fully reflect abnormalities in important periods in the evaluation coefficients, enabling the welding evaluation coefficients to more accurately and objectively assess the degree of impact of the target welding monitoring data on the welding process.

[0048] S500 classifies and stores the target welding monitoring data for each welding process based on the welding evaluation coefficient of each welding process.

[0049] In this step, classification storage refers to classifying the target welding monitoring data of each welding process according to the size of the welding evaluation coefficient or other relevant rules, and then storing the data of the same category together. This facilitates subsequent targeted analysis and management of different types of data, and improves the efficiency of data storage and analysis.

[0050] Specifically, reasonable classification rules can be set based on the welding evaluation coefficients of each welding process. For example, the risk can be divided into high, medium, and low risk levels according to the coefficient value, or specific intervals can be set for classification. The target welding monitoring data of each welding process can then be stored in the corresponding database or storage area according to the classification rules.

[0051] Among them, the welding evaluation coefficients for different welding processes reflect the degree of anomalies in the target welding monitoring data, thereby reflecting the welding quality status. By classifying and storing data, data with similar quality status can be centrally managed, facilitating targeted analysis of different categories of data. For example, high-risk data can be used for focused inspection and process optimization, while low-risk data can be used to summarize experiences, thus improving data management and analysis efficiency and providing strong support for welding quality improvement.

[0052] This invention overcomes the similarity measurement bias caused by isolated analysis of a single data dimension in existing technologies by jointly evaluating abnormal periods in target data and related data through correlation analysis. It identifies abnormal periods by combining time-domain fluctuation analysis and frequency-domain spectrum analysis, and calculates anomaly evaluation coefficients by incorporating the overlap of abnormal periods and fluctuation consistency of related data. Ultimately, it achieves classified storage of welding data based on multi-dimensional correlation analysis, improving the accuracy of data storage grouping and the effectiveness of process optimization.

[0053] As an example, in the automotive chassis welding process, the welding equipment synchronously collects data on voltage, current, wire feed speed, and workpiece temperature using multiple sensors, with a sampling frequency set to 2000Hz. First, welding information at each sampling moment during the welding process is acquired in real time based on this preset sampling frequency.

[0054] Next, for various welding monitoring data, the time-domain deviation coefficient is calculated and Fourier transform is performed to obtain the spectrum results. Combining the time-domain deviation coefficient and the spectrum results, the data anomaly parameters at each sampling time are determined. Monitoring data with data anomaly parameters greater than a preset anomaly threshold are identified as abnormal monitoring data, and the time period corresponding to multiple consecutive abnormal monitoring data is identified as the data anomaly period.

[0055] Furthermore, voltage data was selected as the target welding monitoring data, while current and temperature data were selected as the associated welding monitoring data. Correlation analysis was performed between each abnormal period of the voltage data and the corresponding abnormal periods of the current and temperature data. Specifically, the differences in the start time and duration of the abnormality were calculated, and dynamic time warping analysis was conducted to obtain the period similarity index. Based on these indicators, the fluctuation similarity was determined, and thus the abnormality evaluation coefficient for each abnormal period of the voltage data was obtained.

[0056] Then, based on the anomaly evaluation coefficients for each voltage data anomaly period, a target anomaly evaluation curve is constructed. This target anomaly evaluation curve is compared with the target welding monitoring curve for the voltage data to obtain the welding evaluation coefficient. Finally, based on the welding evaluation coefficients for each welding process, clustering is performed, and the voltage data is classified and stored according to a preset storage strategy. Specifically, anomaly monitoring data is stored using lossless compression, while normal monitoring data is stored using lossy compression.

[0057] This embodiment first acquires multiple types of welding monitoring data in real time for each welding process, enabling a comprehensive understanding of the welding information. Next, based on changes in the welding monitoring data and spectral results, it identifies data anomaly periods during the welding process, accurately pinpointing time periods that may affect welding quality. Then, it performs correlation analysis between the target data anomaly periods and related data anomaly periods to obtain anomaly evaluation coefficients. This fully considers the inherent connections between different welding monitoring data, moving beyond isolated views of single-dimensional data changes and more accurately reflecting the impact of data anomalies on the welding process. Subsequently, it determines welding evaluation coefficients based on each anomaly evaluation coefficient, providing an overall assessment of the welding process. Finally, it categorizes and stores the target welding monitoring data according to the welding evaluation coefficients. This classification method combines the correlation between data and the overall welding situation, providing a more scientific and reasonable grouping and storage of welding data compared to existing methods. This effectively avoids storage bias caused by ignoring the inherent connections between data, thereby improving the effectiveness of welding data storage and analysis.

[0058] In some of the solutions described above in this invention, a method for identifying abnormal data periods based on changes in welding monitoring data and spectral results is proposed. However, in the specific implementation process, there is still a problem of unclear judgment criteria in how to accurately quantify the degree of data abnormality and effectively distinguish between abnormal periods and normal fluctuations, which may lead to the abnormal period identification results being affected by noise interference or subjective threshold settings.

[0059] In this regard, such as Figure 2 As shown, the present invention further proposes that S200 may specifically include the following S210 to S230: S210, Based on the changes in welding monitoring data and the spectral results of welding monitoring data, determine the abnormal parameters of welding monitoring data at each sampling time. S220, Welding monitoring data with abnormal parameters exceeding a preset abnormal threshold are identified as abnormal monitoring data; S230 defines the time period consisting of the sampling times corresponding to multiple consecutive abnormal monitoring data as the data abnormal period in the welding process.

[0060] In this embodiment, for each sampling time and the welding monitoring data within a certain time window before and after it (e.g., welding monitoring data from five sampling times before and after), common statistics such as mean, standard deviation, and range are calculated to reflect the changes in the welding monitoring data. Simultaneously, a Fourier transform is performed on the welding monitoring data to convert the time-domain signal into a frequency-domain signal, yielding the spectrum of the welding monitoring data. Key features, such as the amplitude of the main frequency components and the concentration of frequency distribution, are then extracted from the spectrum. This integrates the statistical measures and spectral characteristics of the welding monitoring data changes to construct a calculation model for data anomaly parameters. For example, a weighted average method can be used, assigning different weights to different statistics and spectral characteristics, and then calculating the data anomaly parameter A.

[0061] For preset anomaly thresholds, a corresponding preset threshold can be set based on historical welding data and welding process requirements, combined with expert experience. For example, by analyzing a large amount of data from normal welding processes, the typical distribution range of abnormal parameters can be determined, and then a value exceeding a certain proportion of that range can be selected as the preset anomaly threshold. Furthermore, in practical applications, adaptive methods can be used to adjust the preset anomaly threshold. For instance, based on the distribution of abnormal parameters over a recent period, the threshold can be dynamically adjusted to better adapt to changes in the welding process.

[0062] The sampling times of all identified anomaly monitoring data are sorted to form a time series. Then, the time interval between adjacent anomaly monitoring data sampling times is checked. If the time interval between adjacent anomaly monitoring data sampling times is less than a preset threshold (e.g., 0.1 seconds), these anomaly monitoring data are considered continuous. The first sampling time among the multiple consecutive anomaly monitoring data is then taken as the start time of the data anomaly period, and the last sampling time is taken as the end time of the data anomaly period, thus determining a data anomaly period.

[0063] As an example, based on the changes in welding monitoring data and the spectral results of the welding monitoring data, the data anomaly parameters of the welding monitoring data at each sampling time are first determined. Specifically, the data anomaly parameters of the welding monitoring data at each sampling time can be comprehensively evaluated by calculating the fluctuation amplitude of the welding monitoring data in the time domain and the frequency domain characteristics obtained by performing a Fourier transform on the welding monitoring data.

[0064] Welding monitoring data with abnormal parameters exceeding a preset abnormal threshold are then identified as abnormal monitoring data. For example, the preset abnormal threshold can be set to 3 times the standard deviation. When the abnormal parameters of the welding monitoring data at a certain sampling point exceed this preset abnormal threshold, it is marked as abnormal monitoring data.

[0065] Finally, the time period consisting of the sampling times corresponding to multiple consecutive abnormal monitoring data is determined as the data abnormal period in the welding process. Specifically, if the time interval between the sampling times of adjacent abnormal monitoring data is less than a preset threshold (e.g., 0.1 seconds), these abnormal monitoring data are considered to be continuous, thus constituting a data abnormal period for multiple consecutive abnormal monitoring data.

[0066] This embodiment enables accurate identification of abnormal data periods during the welding process. This effectively distinguishes between normal and abnormal welding data, providing a foundation for subsequent data analysis and storage. Furthermore, by comprehensively considering both time-domain and frequency-domain characteristics, the accuracy and robustness of anomaly detection are improved.

[0067] In some of the solutions described above in this invention, a method for identifying abnormal data periods based on changes in welding monitoring data and spectral results is proposed. However, in the process of determining abnormal data parameters, the abnormal data situation is evaluated solely through time domain analysis or frequency domain analysis without considering the coupling relationship between time domain fluctuations and frequency domain characteristics, resulting in insufficient accuracy in identifying abnormal periods.

[0068] In this regard, the present invention further proposes that S210 may specifically include: Based on the changes in welding monitoring data, the time-domain deviation coefficient of the welding monitoring data at each sampling time is determined; the time-domain deviation coefficient is used to characterize the data anomalies obtained by time-domain analysis of the welding monitoring data. Fourier transform was performed on the welding monitoring data to obtain the spectral results of the welding monitoring data; Based on the spectral results of welding monitoring data, determine the frequency resolution of the welding monitoring data; The time-domain deviation coefficients of each welding monitoring data are multiplied by the frequency resolution to obtain the data anomaly parameters of the welding monitoring data at each sampling time.

[0069] In this embodiment, the time-domain deviation coefficient is used to reflect the degree of data fluctuation in the time domain. For example, it can be obtained by calculating the ratio of the absolute value of the difference between the welding monitoring data and the welding mean data to the welding mean data. The Fourier transform converts the time-domain signal into a frequency-domain signal, and the spectrum result is used to extract the frequency resolution. The frequency resolution is determined by the interval between adjacent frequency points in the spectrum, which can be specifically determined by dividing the sampling rate by the number of sampling points. The data anomaly parameter is generated by multiplying the time-domain deviation coefficient and the frequency resolution, and the influence of time-domain fluctuation amplitude and frequency-domain characteristics on anomaly detection is combined.

[0070] Specifically, welding monitoring data is averaged to generate welding mean data. The absolute value of the difference between the welding monitoring data at each sampling time and the welding mean data is divided by the welding mean data to obtain the data volatility. Subtracting the overall volatility range constant from the data volatility yields the time-domain anomaly coefficient. After performing a Fourier transform on the welding monitoring data, the reciprocal of the interval between adjacent frequency points in the spectrum result is defined as the frequency resolution, which reflects the accuracy of the frequency domain analysis. Multiplying the time-domain anomaly coefficient by the frequency resolution yields the data anomaly parameters of the welding monitoring data.

[0071] As an example, we first determine the time-domain deviation coefficient of the welding monitoring data at each sampling time based on the changes in the welding monitoring data. The time-domain deviation coefficient is used to characterize the data anomalies obtained from time-domain analysis of the welding monitoring data. Specifically, the time-domain deviation coefficient can be obtained by calculating the deviation between the welding monitoring data and its mean. Since the time-domain deviation coefficient has specific dimensions, these dimensions must be considered in subsequent practical processing.

[0072] Furthermore, a Fourier transform is performed on the welding monitoring data to obtain the spectral results. The Fourier transform can convert the time-domain signal into a frequency-domain signal, thereby revealing the periodic characteristics in the welding monitoring data.

[0073] Based on the spectral results of welding monitoring data, the frequency resolution of the welding monitoring data is determined. Frequency resolution reflects the accuracy of the spectral analysis and can be calculated using the sampling rate and the number of sampling points. Similarly, frequency resolution also has its corresponding dimensions. Since the time-domain partial anisotropy coefficient and frequency resolution have different units, resulting in a dimensionality issue, both can be normalized during the acquisition of the time-domain partial anisotropy coefficient and frequency resolution to eliminate the influence of dimensions.

[0074] Finally, the time-domain anomaly coefficients of each welding monitoring data point, after standardization / normalization, are multiplied by the standardized / normalized frequency resolution to obtain the data anomaly parameters for each sampling time. Thus, the data anomaly parameters comprehensively consider both time-domain and frequency-domain characteristics, providing a more complete reflection of anomalies in the welding monitoring data.

[0075] For example, in practical applications, welding current data can be collected first, and its time-domain anomaly coefficient can be calculated. Then, a fast Fourier transform is performed on the current data to obtain a spectrum. The frequency resolution, such as 0.1 Hz, is determined based on the spectrum. Next, the time-domain anomaly coefficient and the frequency resolution are processed accordingly according to a preset dimensionality processing method. Finally, the processed time-domain anomaly coefficient is multiplied by the frequency resolution to obtain the data anomaly parameters for each sampling point.

[0076] This embodiment comprehensively considers the time-domain and frequency-domain characteristics of welding monitoring data, improving the accuracy of anomaly detection. By introducing frequency resolution, periodic anomalies in the welding process can be better reflected. This method can effectively identify both instantaneous and persistent anomalies in the welding process, providing a reliable basis for subsequent welding quality assessment and process optimization.

[0077] In some of the above-mentioned solutions of the present invention, when determining the time-domain deviation coefficient of welding monitoring data through time-domain analysis, if the judgment is based solely on the difference between the current sampling time and the mean data, it may not be able to effectively distinguish between normal process fluctuations and real abnormal fluctuations, leading to the risk of misjudgment in the identification of abnormal periods.

[0078] To address this, the present invention further proposes determining the temporal deviation coefficient of welding monitoring data at each sampling time based on changes in welding monitoring data, including: The welding monitoring data at each sampling time were averaged to obtain the welding mean data; The absolute value of the difference between the welding monitoring data and the welding mean data at each sampling time is divided by the welding mean data to obtain the data fluctuation at each sampling time. The temporal deviation coefficient of the welding monitoring data at each sampling time is obtained by subtracting the overall fluctuation range constant from the data fluctuation at each sampling time.

[0079] In this embodiment, the mean value processing can employ a sliding window algorithm, selecting a preset number of adjacent data points centered on the current sampling time to calculate the average value as the welding mean data. Data fluctuation is calculated by the ratio of the absolute value of the difference to the welding mean data, used to characterize the relative deviation of the current sampled data relative to the baseline. The overall fluctuation range constant is obtained statistically from historical welding process data and is used to eliminate the interference of normal process fluctuations on anomaly detection.

[0080] Specifically, during the operation of the welding equipment, the welding current data is used to calculate a dynamic mean curve through a sliding window. For the current value at a certain sampling moment, the absolute difference between it and the corresponding window mean is calculated, and this difference is divided by the window mean to obtain the current fluctuation percentage at that moment. For example, when the window mean is 200A and the current sampling value is 220A, the fluctuation percentage is (20 / 200)×100%=10%. If the overall fluctuation range constant is 8%, then the time-domain deviation coefficient at that moment is 10%-8%=2%. This time-domain deviation coefficient can effectively filter out normal fluctuations allowed by the process, retaining only abnormal fluctuation data that exceeds the normal range, thereby improving the accuracy of abnormal period identification. This scheme achieves dual verification of time-domain and frequency-domain anomaly characteristics through collaborative processing with spectrum analysis results.

[0081] Specifically, the time-domain deviation coefficient can be determined using the following formula 1: Formula 1 In formula 1, Used to characterize time-domain partial differential coefficients The welding monitoring data used to characterize the i-th sampling time. Used to characterize the welding mean data at each sampling time. Used to characterize the overall fluctuation range constant.

[0082] in, This is the ratio of the absolute value of the difference to the average welding data, reflecting the data fluctuation at the i-th sampling time. This is the difference between the data volatility and the overall volatility range constant. The larger this value is, the more severe the data anomalies generated during the welding process at the i-th sampling time.

[0083] As an example, the welding monitoring data at each sampling time is averaged to obtain the welding mean data. For instance, for the current data in a certain welding process, the current values ​​from a total of 100 measurements can be averaged to obtain the average current value as the welding mean data.

[0084] Next, the absolute value of the difference between the welding monitoring data at each sampling time and the welding mean data is divided by the welding mean data to obtain the data fluctuation at each sampling time. Specifically, the absolute value of the difference between the current value at each sampling time and the average current value can be calculated, and then divided by the average current value to obtain the current fluctuation at that time.

[0085] Next, the overall fluctuation range constant is subtracted from the data fluctuation at each sampling time to obtain the time-domain deviation coefficient of the welding monitoring data at each sampling time. The overall fluctuation range constant can be determined based on welding process requirements and empirical values; for example, it can be set to 0.1. Subtracting this constant highlights abnormal fluctuations, resulting in a more accurate time-domain deviation coefficient.

[0086] This embodiment enables accurate calculation of the time-domain deviation coefficient of welding monitoring data, effectively identifying abnormal fluctuations during the welding process. This improves the accuracy of welding data analysis, providing a reliable basis for subsequent welding quality assessment and process optimization. Furthermore, this method considers the overall fluctuation range, avoiding misjudging normal fluctuations as abnormalities and improving the accuracy and reliability of anomaly detection.

[0087] In some of the above-mentioned solutions of the present invention, when obtaining the anomaly evaluation coefficient by performing correlation analysis between the target data anomaly period and the associated data anomaly period, the similarity measurement of a single associated data anomaly period is relied upon only, without comprehensively considering the correlation between the target data anomaly period and the corresponding associated data anomaly periods, which affects the reliability of the anomaly evaluation coefficient.

[0088] In this regard, such as Figure 3 As shown, the present invention further proposes that S300 may specifically include the following S310 to S320: S310, perform correlation analysis between the abnormal period of the target data and the corresponding abnormal period of each associated data to obtain the fluctuation similarity between the abnormal period of the target data and the corresponding abnormal period of each associated data; S320, Based on the similarity of fluctuations in the abnormal period of the target data, determine the abnormal evaluation coefficient of the abnormal period of the target data.

[0089] In this embodiment, a suitable algorithm or model is used to perform correlation analysis between the abnormal periods of the target data and the corresponding abnormal periods of each associated data to obtain the fluctuation similarity between them. For example, a common method is to use time series analysis techniques, such as the dynamic time warping algorithm. The dynamic time warping algorithm can non-linearly align two time series and calculate their similarity. In practice, the time series of the abnormal periods of the target data and the time series of the abnormal periods of the associated data are input into the dynamic time warping algorithm. The algorithm will find the best matching path between the two sequences and calculate the similarity value based on the length of the matching path and the difference between the sequences. This similarity value can be used as the fluctuation similarity.

[0090] When determining anomaly evaluation coefficients based on fluctuation similarity, different weights can be assigned to different abnormal periods of related data. The weight allocation can be determined based on factors such as the correlation and importance between the related welding monitoring data and the target welding monitoring data. For example, if the correlation between welding voltage data and welding current data is high, then when calculating the anomaly evaluation coefficient for abnormal periods of the target welding current data, a higher weight can be given to the fluctuation similarity of the abnormal periods of the welding voltage related data.

[0091] Finally, based on the assigned weights (if no weights are assigned, the weights of each fluctuation similarity are assumed to be the same), the similarities of each fluctuation during the abnormal period of the target data are weighted and summed or other appropriate mathematical operations (such as averaging) are used to obtain the abnormal evaluation coefficient of the abnormal period of the target data.

[0092] This embodiment analyzes the correlation between the abnormal time periods of the target welding monitoring data and the abnormal time periods of the associated welding monitoring data, and determines the anomaly evaluation coefficient based on fluctuation similarity. This approach comprehensively considers the impact of multiple factors on welding process anomalies, thereby improving the reliability of the anomaly evaluation coefficient.

[0093] In some of the solutions described above in this invention, when obtaining the anomaly evaluation coefficient by performing correlation analysis between the abnormal time period of the target data and the abnormal time period of the associated data, the analysis is performed only by the dynamic time warping algorithm. This results in the correlation analysis results failing to accurately reflect the intrinsic relationship between the data, thus affecting the reliability of the anomaly evaluation coefficient.

[0094] In this regard, the present invention further proposes that S310 may specifically include: The difference between the start time of the anomaly in the target data's anomaly period and the start time of the anomaly in the associated data's anomaly period is defined as the anomaly start time difference. The difference between the duration of the abnormality in the target data during an abnormal period and the duration of the abnormality in the associated data during an abnormal period is defined as the abnormal duration difference. Dynamic time warping analysis is performed on the abnormal monitoring data of the target data during abnormal periods and the abnormal monitoring data of the associated data during abnormal periods to obtain the time similarity index between the abnormal periods of the target data and the abnormal periods of the associated data. By utilizing differences in the start time of anomalies, differences in the duration of anomalies, and time period similarity indices, the fluctuation similarity between the anomaly periods of the target data and the anomaly periods of related data can be determined.

[0095] In this embodiment, the difference in the start time of an anomaly is achieved by calculating the time difference between the start times of two anomaly periods. For example, if the target anomaly period starts at second 5 and the associated anomaly period starts at second 6, the difference is 1 second. The difference in the duration of an anomaly is achieved by calculating the difference in the duration of two anomaly periods. For example, if the target anomaly period lasts for 3 seconds and the associated anomaly period lasts for 4 seconds, the difference is 1 second. The period similarity index is obtained by aligning and matching the data sequences within two anomaly periods using a dynamic time warping algorithm. After calculating the DTW distance, the normalized DTW distance is subtracted from 1 to obtain the period similarity index. Fluctuation similarity can be obtained by substituting the difference in the start time of anomaly, the difference in the duration of anomaly, and the period similarity index into a weighted calculation formula.

[0096] Specifically, the difference in the start time of an anomaly measures the synchronicity of the anomaly event triggering; a large difference indicates a low correlation. The difference in the duration of an anomaly determines whether the time coverage of the anomaly fluctuations is consistent; a smaller difference indicates a stronger correlation. The time period similarity index eliminates the influence of time axis offset on data shape through a dynamic time warping algorithm. It should be noted that directly obtaining the difference distance between data sequences within two anomaly time periods using the dynamic time warping algorithm requires processing the DTW distance as described above to obtain the time period similarity index. For example, if there is a time delay between the current anomaly fluctuation curve and the temperature anomaly fluctuation curve, dynamic time warping can align the two curves and calculate the shape matching degree. By combining parameters from these three dimensions, fluctuation similarity can simultaneously reflect time synchronization, continuity consistency, and the degree of matching of data fluctuation shapes.

[0097] Specifically, the wave similarity can be determined using the following formula 2: Formula 2 In Formula 2, X is used to characterize wave similarity. The difference in the start time of the anomaly between the target data anomaly period of the j-th welding monitoring data and the associated data anomaly period of the (j+1)-th welding monitoring data. The difference in the duration of anomalies between the target data anomaly period of the j-th welding monitoring data and the associated data anomaly period of the (j+1)-th welding monitoring data is used to characterize the difference in anomaly duration between the target data anomaly period and the associated data anomaly period of the (j+1)-th welding monitoring data. The time-similarity index, used to characterize the abnormal time period of the target data of the j-th welding monitoring data and the abnormal time period of the associated data of the (j+1)-th welding monitoring data, is normalized using the norm function.

[0098] in, This indicates the synchronicity of changes. The larger the value, the more similar the changes of the j-th welding monitoring data and the (j+1)-th welding monitoring data are. During welding, when the j-th welding monitoring data changes, the (j+1)-th welding monitoring data will change accordingly. Used to characterize the overall correlation between the j-th welding monitoring data and the (j+1)-th welding monitoring data. During the welding process, if the welding monitoring data items show similar change trends at the same or similar times, then the changes of the two welding monitoring data items are highly similar.

[0099] As an example, the difference between the start time of the anomaly in the target data's anomaly period and the start time of the anomaly in the associated data's anomaly period is defined as the anomaly start time difference. For instance, if the start time of the anomaly in the target data's anomaly period is t1 and the start time of the anomaly in the associated data's anomaly period is t2, then the anomaly start time difference is |t1-t2|.

[0100] The difference between the duration of the abnormality in the target data's abnormal period and the duration of the abnormality in the associated data's abnormal period is defined as the abnormal duration difference. For example, if the duration of the abnormality in the target data's abnormal period is d1 and the duration of the abnormality in the associated data's abnormal period is d2, then the abnormal duration difference is |d1-d2|.

[0101] Dynamic time warping analysis is performed on the anomaly monitoring data of the target data during anomaly periods and the anomaly monitoring data of the associated data during anomaly periods to obtain a time similarity index between the target data anomaly periods and the associated data anomaly periods. Specifically, the dynamic time warping algorithm can be used to calculate the similarity between the two time series, resulting in a similarity index value between 0 and 1.

[0102] Finally, by utilizing the differences in the start time of anomalies, the differences in the duration of anomalies, and the time period similarity index, the fluctuation similarity between the abnormal time periods of the target data and the abnormal time periods of the associated data is determined using the above formula 2.

[0103] This embodiment comprehensively considers the start time, duration, and data change trends of abnormal periods to perform multi-dimensional similarity analysis between abnormal periods in the target data and those in related data. This improves the accuracy of correlation analysis for abnormal periods, providing a more reliable basis for subsequent calculation of anomaly evaluation coefficients. Furthermore, this scheme can capture the inherent correlation between different welding monitoring data, helping to more comprehensively assess anomalies in the welding process, thereby providing more accurate data support for welding quality control and process optimization.

[0104] In some of the above-mentioned solutions of the present invention, when determining the anomaly evaluation coefficient based on the fluctuation similarity between the target data anomaly period and the associated data anomaly period, the difference in the degree of influence of different associated data anomaly periods on the target data anomaly period is not considered, which leads to the anomaly evaluation coefficient failing to accurately reflect the difference in contribution of associated data anomalies to target data anomalies, thus affecting the accuracy of subsequent welding evaluation coefficients.

[0105] In this regard, the present invention further proposes that S320 may specifically include: Based on the data anomaly parameters of each anomaly monitoring data in each associated data anomaly period of the target data anomaly period, the similarity weight of each fluctuation similarity is determined respectively; By utilizing the similarity of each fluctuation and the corresponding similarity weight, the anomaly evaluation coefficient for the abnormal period of the target data is determined.

[0106] In this embodiment, the similarity weight is determined by calculating the sum of the data anomaly parameters of all anomaly monitoring data within each associated data anomaly period. The larger the sum of the data anomaly parameters, the higher the corresponding fluctuation similarity weight. The product of each fluctuation similarity and its corresponding similarity weight is integrated using a weighted average algorithm to finally generate the anomaly evaluation coefficient. For example, when the associated data anomaly period contains three anomaly monitoring data with data anomaly parameters of 0.85, 0.92, and 0.78, the similarity weight for this period can be the result of normalizing the sum of the three (2.55). This similarity weight is multiplied by the corresponding fluctuation similarity and then used in the weighted calculation.

[0107] Specifically, after obtaining the fluctuation similarity between the target data anomaly period and each associated data anomaly period, the process first iterates through each associated data anomaly period, extracting the data anomaly parameters of all anomaly monitoring data within that period. The data anomaly parameters within the same associated data anomaly period are then summed and normalized to obtain the similarity weight for that period. Subsequently, each fluctuation similarity is multiplied by its corresponding similarity weight to obtain the anomaly evaluation coefficient for the target data anomaly period. In this process, the sum of data anomaly parameters serves as a weighting factor, effectively reflecting the differences in the dynamic impact of different associated welding monitoring data anomalies on the target welding monitoring data anomalies. For example, when the sum of data anomaly parameters for a certain associated welding monitoring data anomaly period is 1.5 times that of other periods, the contribution of its corresponding fluctuation similarity in the anomaly evaluation coefficient calculation increases by 1.5 times, thus ensuring a positive correlation between the anomaly evaluation coefficient and the actual impact of associated welding monitoring data anomalies.

[0108] Specifically, the anomaly evaluation coefficient can be determined using the following formula 3: Formula 3 In Formula 3, Q is used to characterize the anomaly evaluation coefficient. Used to characterize the duration of anomalies in the m-th associated data period. Data anomaly parameters used to characterize the anomaly monitoring data at time i within the m-th associated data anomaly period. Used to characterize the start time of an abnormal period in the associated data. M is used to represent the end time of an abnormal period in the associated data, and M is used to represent the total number of abnormal periods in the associated data. The fluctuation similarity of the m-th associated data during anomaly periods is characterized by normalization using the tanh function.

[0109] As an example, when determining the anomaly evaluation coefficient for a target data anomaly period based on the similarity of fluctuations within that period, the similarity weight of each fluctuation is first determined based on the data anomaly parameters of each anomaly monitoring data in each associated data anomaly period. Specifically, a weighted average method can be used to calculate the similarity weight. For instance, for a certain associated data anomaly period, the average value of the data anomaly parameters of the anomaly monitoring data within that period is calculated, and this average value is divided by the sum of the average values ​​of the data anomaly parameters of all associated data anomaly periods to obtain the similarity weight of the fluctuation similarity corresponding to that associated data anomaly period.

[0110] Furthermore, by utilizing the similarity of each fluctuation and its corresponding similarity weight, the anomaly evaluation coefficient for the abnormal period of the target data is determined. Specifically, a weighted summation method can be used, whereby each fluctuation similarity is multiplied by its corresponding similarity weight and then summed to obtain the anomaly evaluation coefficient for the abnormal period of the target data. Thus, by introducing similarity weights, the influence of different abnormal periods of related data on the abnormal period of the target data can be more accurately reflected, thereby improving the calculation accuracy of the anomaly evaluation coefficient.

[0111] This embodiment enables a more accurate assessment of the degree of anomaly during periods of abnormality in the target data. By introducing similarity weights, the differences in importance among different periods of abnormality in related data are considered, avoiding assessment biases that may result from simple averaging. This method better reflects the intrinsic relationship between anomalies in the target data and those in related data, improving the accuracy and reliability of welding data analysis. This, in turn, provides more reliable data support for subsequent welding quality assessments and process optimization.

[0112] In some of the solutions described above in this invention, anomaly evaluation coefficients for abnormal periods of target data are obtained through correlation analysis. However, time series based solely on anomaly evaluation coefficients may not fully reflect the overall abnormal dynamic changes of target welding monitoring data during the welding process, thus limiting the accuracy of welding evaluation coefficients.

[0113] In this regard, such as Figure 4As shown, the present invention further proposes that S400 may specifically include the following S410 to S430: S410 connects the evaluation coefficients of each abnormality during the welding process in chronological order to form the target abnormality evaluation curve; S420 connects the welding monitoring data of each target during the welding process in chronological order to form a target welding monitoring curve; S430, compare the target anomaly evaluation curve with the target welding monitoring curve to obtain the welding evaluation coefficient of the target welding monitoring data during the welding process.

[0114] In this embodiment, the target anomaly evaluation curve is generated by linearly connecting each anomaly evaluation coefficient in chronological order, reflecting the changing trend of the anomaly evaluation coefficient over time; the target welding monitoring curve is generated by connecting the original values ​​of the target welding monitoring data in chronological order, preserving the time dimension characteristics of the data.

[0115] Specifically, according to the time sequence of the welding process, the various anomaly evaluation coefficients are connected sequentially to form the target anomaly evaluation curve. For example, assuming the welding process starts at time t0 and ends at time tn, there are target data anomaly periods at time points t1, t2, and t3, with corresponding anomaly evaluation coefficients C1, C2, and C3, etc. Then, these anomaly evaluation coefficients are plotted on a coordinate system in chronological order, with the horizontal axis representing time and the vertical axis representing the anomaly evaluation coefficient. Finally, a smooth curve (or broken line) is used to connect the points to form the target anomaly evaluation curve.

[0116] Similarly, the collected target welding monitoring data are sequentially connected according to time to form a target welding monitoring curve. For example, with time as the horizontal axis and welding current value as the vertical axis, the welding current data measured at different time points are plotted and connected in the coordinate system to obtain the target welding monitoring curve. This curve intuitively shows the changes of the target parameters over time during the welding process.

[0117] Various methods can be used to compare the target anomaly evaluation curve and the target welding monitoring curve. Common methods include correlation analysis, difference analysis, and similarity measurement. Correlation analysis involves calculating the correlation coefficient between the two curves, such as the Pearson correlation coefficient. Difference analysis calculates the difference between the corresponding values ​​of the target anomaly evaluation curve and the target welding monitoring curve at the same time point, and then performs statistical analysis on these differences, such as calculating the mean and standard deviation of the differences. Similarity measurement uses similarity measurement methods such as dynamic time warping algorithms.

[0118] Finally, the welding evaluation coefficient is determined based on the comparison results. For example, if correlation analysis is used, a positive and large correlation coefficient indicates a positive correlation between the changes in the target welding monitoring data and the abnormal situation, which may mean that the abnormal situation has a significant impact on the target parameters during the welding process. In this case, a higher welding evaluation coefficient can be assigned based on the magnitude of the correlation coefficient. Conversely, if the correlation coefficient is close to 0 or negative, it indicates a weak relationship between the two, and a lower welding evaluation coefficient can be assigned. If difference analysis is used, the welding evaluation coefficient can be determined by combining the average and standard deviation of the differences. The larger the difference and the more drastic the fluctuation, the lower the welding evaluation coefficient may be.

[0119] This embodiment effectively solves the problem of calculation deviation in welding evaluation coefficients caused by isolated analysis of single-dimensional data. By simultaneously capturing the morphological matching degree and temporal dynamic regularity characteristics of the abnormal evaluation curve and the original monitoring curve, the welding evaluation coefficient can objectively reflect the comprehensive fluctuation characteristics of the target welding monitoring data under the influence of abnormal correlations. The welding evaluation coefficient established in this way can accurately characterize the degree of abnormal correlation influence in the welding process, providing a reliable basis for subsequent data classification and storage based on multi-dimensional correlation characteristics.

[0120] In some of the above-mentioned solutions of the present invention, the abnormal evaluation coefficients during the welding process are connected in chronological order to form a target abnormal evaluation curve, and the target welding monitoring data are connected in chronological order to form a target welding monitoring curve. However, by simply comparing the differences in the shape of the two curves, it is impossible to accurately measure the degree of dynamic matching between the abnormal evaluation coefficients and the target welding monitoring data in the time dimension, which leads to a deviation in the determination of the welding evaluation coefficients.

[0121] In this regard, the present invention further proposes that S430 may specifically include: By comparing the target anomaly evaluation curve with the target welding monitoring curve, the fitting difference parameters between each anomaly node in the target anomaly evaluation curve and the target welding monitoring curve are obtained. Based on the dynamic time warping distance between the target anomaly evaluation curve and the target welding monitoring curve, the curve similarity between the target anomaly evaluation curve and the target welding monitoring curve is determined. Welding evaluation coefficients for target welding monitoring data during the welding process are determined by utilizing curve similarity and various fitting difference parameters.

[0122] In this embodiment, the abnormal node is the point corresponding to the abnormal evaluation coefficient. The fitting difference parameter is obtained by calculating the Euclidean distance between the target welding monitoring curve data point at the time corresponding to the abnormal node in the target abnormal evaluation curve and the abnormal evaluation coefficient. The dynamic time warping distance is calculated by using the dynamic time warping algorithm to calculate the minimum path cumulative distance between the two curves. This distance reflects the morphological difference between the two curves on the time axis. Subtracting the normalized dynamic time warping distance between the target abnormal evaluation curve and the target welding monitoring curve from 1 yields the curve similarity between the target abnormal evaluation curve and the target welding monitoring curve.

[0123] Specifically, in the target anomaly evaluation curve, each anomaly node represents a time corresponding to an anomaly evaluation coefficient. By calculating the Euclidean distance between the target welding monitoring data point and the anomaly evaluation coefficient of the anomaly node at that time, the fitting difference parameter is obtained. The dynamic time warping algorithm nonlinearly aligns the time axes of the two curves and calculates the cumulative distance under the optimal matching path. This distance can eliminate the interference caused by time axis scaling or local offset.

[0124] The welding evaluation coefficient can be determined using the following formula 4: Formula 4 In Formula 4, F is used to characterize the welding evaluation coefficient. Used to characterize the target welding monitoring curve LN and the target anomaly evaluation curve Curve similarity between them Used to characterize the target anomaly evaluation curve The fitting difference parameter between the z-th abnormal node and the target welding monitoring curve LN is used to characterize the number of abnormal nodes, and is processed by inverse proportional normalization using the exp function.

[0125] The greater the similarity between the curves, the higher the temporal alignment between the target anomaly evaluation curve and the target welding monitoring curve, and thus the larger the welding evaluation coefficient. The larger the fitting difference parameter between each abnormal node and the target welding monitoring curve LN, the greater the exponential decay term. The smaller the value, the smaller the welding evaluation coefficient.

[0126] As an example, during the welding process, the target anomaly evaluation curve is formed by connecting the anomaly evaluation coefficients of the welding current in chronological order, while the target welding monitoring curve is formed by connecting the real-time acquired welding current data in chronological order. First, the two curves are aligned along their time axes, and the Euclidean distance between the welding current data and the evaluation coefficient at each anomaly node in the anomaly evaluation curve is calculated as a fitting difference parameter. Next, a dynamic time warping algorithm is used to non-linearly align the two curves, obtaining the path distance between them as the dynamic time warping distance. Subtracting the dynamic time warping distance between the target anomaly evaluation curve and the target welding monitoring curve from 1 yields the curve similarity between them. Finally, based on the curve similarity and the fitting difference parameters, the welding evaluation coefficient is determined using the formula described above (Formula 4).

[0127] This embodiment effectively solves the problem of similarity measurement bias caused by ignoring the inherent relationships between related data in existing technologies. By integrating a dual evaluation mechanism of temporal fitting differences and dynamic regularization distance, the abnormal correlation characteristics of welding monitoring data can be accurately quantified, thereby improving the accuracy of welding evaluation coefficient calculation. This enables the classified storage of welding data to truly reflect the collaborative anomaly patterns among multidimensional data, ultimately improving the data storage and analysis efficiency of the welding information management system.

[0128] In some of the solutions described above in this invention, clustering of the welding process based on welding evaluation coefficients is proposed to achieve classified storage. However, this process does not consider the differentiated needs of different categories of welding data during storage, resulting in unreasonable allocation of storage resources. Key details of abnormal monitoring data may be lost due to improper compression methods, while normal monitoring data occupies too much storage space.

[0129] In this regard, the present invention further proposes that S500 may include: Based on the welding evaluation coefficients of each welding process, clustering operations are performed on each welding process to obtain clustering results; Based on the clustering results, the target welding monitoring data of each welding process are classified and stored according to a preset storage strategy. The preset storage strategy is to perform lossless compression storage on abnormal monitoring data in the target welding monitoring data, and to perform lossy compression storage on normal monitoring data other than abnormal monitoring data in the target welding monitoring data.

[0130] In this embodiment, the clustering operation divides welding processes with similar welding evaluation coefficients into the same group. Welding processes within the same group have similar data fluctuation patterns and abnormal characteristics. In the preset storage strategy, lossless compression uses algorithms that preserve the integrity of the original data, such as LZW or FLAC, while lossy compression uses algorithms based on discrete cosine transform or wavelet transform. During classification storage, welding data from the same group are allocated to the same storage partition according to the clustering results, and within the partition, abnormal monitoring data and normal monitoring data are further separated and stored.

[0131] Specifically, after calculating the welding evaluation coefficients, the welding process is grouped using K-means or hierarchical clustering algorithms. The cluster center of each group is determined by the mean of the welding evaluation coefficients within that group. For target welding monitoring data within the same cluster, abnormal monitoring data is processed using a lossless compression algorithm and stored on a high-speed storage medium to ensure complete restoration of the original waveform during subsequent quality traceability. Normal monitoring data is processed using a lossy compression algorithm and stored on a large-capacity storage medium to reduce storage space usage. For example, sampling points corresponding to abnormal periods in current monitoring data are compressed using ZIP format, while normal periods are compressed using JPEG2000 format. This ensures the integrity of abnormal data while optimizing overall storage resource utilization and avoiding data distortion or resource waste caused by a single compression strategy.

[0132] As an example, hierarchical clustering is performed on multiple welding processes based on welding evaluation coefficients. After inputting the welding evaluation coefficients into the clustering model, a sample similarity matrix is ​​constructed by calculating Euclidean distance. The nearest clusters are merged using the class average method until all samples are clustered into a preset number of categories. The category labels output by hierarchical clustering are used as the classification criteria to classify and store the welding monitoring data for each welding process. Specifically, abnormal monitoring data in the target welding monitoring data are encoded and stored using the LZW lossless compression algorithm to preserve the original data accuracy; normal monitoring data are stored using the JPEG2000 lossy compression algorithm, and high-frequency redundant information is removed through wavelet transform. The compressed data is then stored in an independent partition.

[0133] This embodiment effectively solves the classification bias problem caused by neglecting multidimensional data correlation during welding data storage. Data grouping optimization is achieved through cluster analysis incorporating welding evaluation coefficients, while a differentiated compression strategy balances storage space and data integrity. Anomaly monitoring data retains complete fluctuation characteristics through lossless compression, facilitating subsequent quality traceability and fault diagnosis; normal monitoring data reduces storage resource consumption through lossy compression, improving the data storage efficiency and processing capacity of the welding information management system.

[0134] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0135] It should also be noted that the exemplary embodiments mentioned in this invention describe methods or systems based on a series of steps or apparatus. However, this invention is not limited to the order of the steps described above; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0136] The above description is merely a specific embodiment of the present invention. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the protection scope of the present invention.

Claims

1. A welding data storage and analysis method for a welding information management system, characterized in that, The method includes: Based on a preset sampling frequency, welding information at each sampling moment during each welding process is acquired in real time, and the welding information includes various welding monitoring data. Based on the changes in the welding monitoring data and the spectral results of the welding monitoring data, abnormal data periods in the welding process are identified. Correlation analysis is performed on each abnormal period of the target welding monitoring data and each abnormal period of the associated welding monitoring data to obtain the abnormal evaluation coefficient of each abnormal period of the target data during the welding process; the target welding monitoring data can be any type of welding monitoring data, and the associated welding monitoring data is the welding monitoring data affected by the fluctuation of the target welding monitoring data; Based on the anomaly evaluation coefficient of each of the target data anomaly periods during the welding process, the welding evaluation coefficient of the target welding monitoring data during the welding process is determined; Based on the welding evaluation coefficients of each welding process, the target welding monitoring data of each welding process are classified and stored.

2. The welding data storage and analysis method for a welding information management system according to claim 1, characterized in that, The method of identifying abnormal data periods during the welding process based on changes in the welding monitoring data and the spectral results of the welding monitoring data includes: Based on the changes in the welding monitoring data and the spectral results of the welding monitoring data, the abnormal data parameters of the welding monitoring data at each sampling time are determined; Welding monitoring data whose abnormal parameters are greater than a preset abnormal threshold are identified as abnormal monitoring data. The time period consisting of the sampling times corresponding to multiple consecutive abnormal monitoring data is determined as the data abnormal period in the welding process.

3. The welding data storage and analysis method for a welding information management system according to claim 2, characterized in that, The determination of data anomaly parameters for the welding monitoring data at each sampling time, based on the changes in the welding monitoring data and the spectral results of the welding monitoring data, includes: Based on the changes in the welding monitoring data, the time-domain deviation coefficient of the welding monitoring data at each sampling time is determined; the time-domain deviation coefficient is used to characterize the data anomalies obtained by performing time-domain analysis on the welding monitoring data; Perform a Fourier transform on the welding monitoring data to obtain the spectrum results of the welding monitoring data; Based on the spectral results of the welding monitoring data, the frequency resolution of the welding monitoring data is determined; The time-domain deviation coefficients of each welding monitoring data are multiplied by the frequency resolution to obtain the data anomaly parameters of the welding monitoring data at each sampling time.

4. The welding data storage and analysis method for a welding information management system according to claim 3, characterized in that, The determination of the temporal deviation coefficient of the welding monitoring data at each sampling time based on the changes in the welding monitoring data includes: The welding monitoring data at each sampling time point are averaged to obtain the welding mean data; The absolute value of the difference between the welding monitoring data and the welding mean data at each sampling time is divided by the welding mean data to obtain the data fluctuation at each sampling time. The temporal deviation coefficient of the welding monitoring data at each sampling time is obtained by subtracting the overall fluctuation range constant from the data fluctuation at each sampling time.

5. The welding data storage and analysis method for a welding information management system according to claim 1, characterized in that, The step of performing correlation analysis between each abnormal period of the target welding monitoring data and each abnormal period of the associated welding monitoring data to obtain the abnormality evaluation coefficient of each of the abnormal periods of the target data during the welding process includes: The correlation analysis is performed between the target data anomaly period and the corresponding associated data anomaly period to obtain the fluctuation similarity between the target data anomaly period and the corresponding associated data anomaly period; Based on the similarity of fluctuations in each of the abnormal periods of the target data, the anomaly evaluation coefficient for the abnormal period of the target data is determined.

6. The welding data storage and analysis method for a welding information management system according to claim 5, characterized in that, The step of performing correlation analysis between the target data anomaly period and the corresponding associated data anomaly periods to obtain the fluctuation similarity between the target data anomaly period and the corresponding associated data anomaly periods includes: The difference between the abnormal start time of the target data abnormal period and the abnormal start time of the associated data abnormal period is determined as the abnormal start time difference. The difference between the duration of the abnormality in the target data abnormal period and the duration of the abnormality in the associated data abnormal period is determined as the abnormal duration difference. Dynamic time warping analysis is performed on the abnormal monitoring data of the target data abnormal period and the abnormal monitoring data of the associated data abnormal period to obtain the time similarity index between the target data abnormal period and the associated data abnormal period. By utilizing the differences in the start time of the anomaly, the differences in the duration of the anomaly, and the time period similarity index, the fluctuation similarity between the anomaly period of the target data and the anomaly period of the associated data is determined.

7. The welding data storage and analysis method for a welding information management system according to claim 5, characterized in that, The step of determining the anomaly evaluation coefficient for each of the abnormal periods of the target data based on the similarity of fluctuations includes: Based on the data anomaly parameters of each anomaly monitoring data in each of the associated data anomaly periods in the target data anomaly period, the similarity weight of each fluctuation similarity is determined respectively; By utilizing the aforementioned fluctuation similarities and their corresponding similarity weights, the anomaly evaluation coefficient for the abnormal period of the target data is determined.

8. The welding data storage and analysis method for a welding information management system according to claim 1, characterized in that, The determination of the welding evaluation coefficient of the target welding monitoring data during the welding process based on the anomaly evaluation coefficient of each of the target data anomaly periods during the welding process includes: The abnormality evaluation coefficients during the welding process are connected in chronological order to form a target abnormality evaluation curve. The target welding monitoring data of each target welding process are connected in chronological order to form a target welding monitoring curve. The target anomaly evaluation curve is compared with the target welding monitoring curve to obtain the welding evaluation coefficient of the target welding monitoring data during the welding process.

9. The welding data storage and analysis method for a welding information management system according to claim 8, characterized in that, The step of comparing the target anomaly evaluation curve with the target welding monitoring curve to obtain the welding evaluation coefficient of the target welding monitoring data during the welding process includes: The target anomaly evaluation curve is compared with the target welding monitoring curve to obtain the fitting difference parameters between each anomaly node in the target anomaly evaluation curve and the target welding monitoring curve. Based on the dynamic time warping distance between the target anomaly evaluation curve and the target welding monitoring curve, the curve similarity between the target anomaly evaluation curve and the target welding monitoring curve is determined; The welding evaluation coefficient of the target welding monitoring data during the welding process is determined by using the curve similarity and the fitting difference parameters.

10. The welding data storage and analysis method for a welding information management system according to claim 1, characterized in that, The method of classifying and storing the target welding monitoring data for each welding process based on the welding evaluation coefficient of each welding process includes: Based on the welding evaluation coefficients of each welding process, a clustering operation is performed on each welding process to obtain the clustering results; Based on the clustering results, the target welding monitoring data in each welding process are classified and stored according to a preset storage strategy. The preset storage strategy is to perform lossless compression storage on abnormal monitoring data in the target welding monitoring data, and to perform lossy compression storage on normal monitoring data other than abnormal monitoring data in the target welding monitoring data.

Images