Welding parameter stability-based welding defect system and monitoring method thereof

Abstract

The invention discloses a welding parameter stability-based welding defect system and a monitoring method thereof. The monitoring method comprises the following steps: respectively acquiring welding parameter waveform characteristics of equipment in a normal state and a fault state; s2, comparing the health parameter library obtained in S1 with the fault parameter library to obtain a feature offset vector between the health parameters and the fault parameters, and obtaining a fault mode vector library according to the feature offset vector; welding parameters in the production process are monitored in real time, real-time welding features are extracted, and real-time welding parameters are generated; s3, according to the real-time welding parameters obtained in the S3, the similarity between the real-time welding parameters obtained in the S3 and the health parameter library obtained in the S1 is calculated, when the similarity is lower than a preset threshold value, a real-time offset vector is calculated, the real-time offset vector is matched with the fault mode vector library obtained in the S2, and the reason of the welding defect is diagnosed; and real-time offset vectors are calculated and subjected to direction matching with vectors in the library, so that the system can accurately discriminate defect reasons.

Description

Technical Field

[0001] This invention relates to the field of welding defects, and more particularly to a welding defect system and its monitoring method based on the stability of welding parameters. Background Technology

[0002] Welding quality is the cornerstone of ensuring the safety and reliability of engineering structures. Traditional welding defect detection methods, such as radiographic testing and ultrasonic testing, are mostly post-weld or offline inspection methods. While these methods can effectively identify macroscopic defects that have already formed, such as porosity, slag inclusions, and incomplete penetration, their inherent lag means they cannot intervene in the welding process. Once a defect is detected, it often means product rework or even scrapping, increasing production and time costs. Therefore, industry and academia have long been committed to developing a technology that can identify the tendency of defects to emerge in real time during the welding process and has early warning capabilities, thereby achieving a paradigm shift from "post-inspection" to "process prevention."

[0003] The stability of welding process parameters is considered a key indicator of weld quality. Extensive research and practice have demonstrated that in a stable and controlled welding process, electrical parameters (such as welding current and arc voltage) and physical parameters (such as wire feed speed and welding speed) remain highly stable within an ideal range. Defects are often directly causally related to abnormal fluctuations or deviations in these parameters. For example, a sudden drop in current may indicate a risk of short circuit or arc interruption, while a sustained deviation in voltage may be related to poor shielding gas performance or excessive arc extension. Therefore, real-time acquisition and stability analysis of welding parameter signals are crucial for effective process monitoring. Summary of the Invention

[0004] This invention overcomes the shortcomings of the prior art and provides a welding defect system and its monitoring method based on the stability of welding parameters.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is: a welding defect monitoring method based on welding parameter stability, comprising the following steps:

[0006] S1: Obtain the waveform characteristics of welding parameters under both normal and faulty equipment conditions, and construct a health parameter library and a fault parameter library based on the waveform characteristics of the welding parameters;

[0007] S2: Compare the health parameter library obtained in S1 with the fault parameter library to obtain the feature offset vector between the health parameters and the fault parameters, and obtain the fault mode vector library based on the feature offset vector.

[0008] S3: Real-time monitoring of welding parameters during the production process, acquisition of welding parameter waveforms during the welding process, extraction of real-time welding features, and generation of real-time welding parameters;

[0009] S4: Based on the real-time welding parameters obtained in S3, calculate the similarity between them and the health parameter library obtained in S1. When the similarity is lower than the preset threshold, calculate the real-time offset vector and match the real-time offset vector with the fault mode vector library obtained in S2 to diagnose the cause of welding defects.

[0010] In a preferred embodiment of the present invention, in S1, establishing a health parameter database specifically involves acquiring the original waveforms of welding current and arc voltage through a welding defect system and its monitoring method based on welding parameter stability when the welding equipment is in good condition.

[0011] Extracting multidimensional statistical and spectral features from the original waveform;

[0012] Based on the feature data set of multiple qualified welds, the mean vector and covariance matrix are calculated to form a parameter library as a health benchmark.

[0013] In a preferred embodiment of the present invention, in S2, establishing the fault mode vector library specifically involves artificially simulating various typical fault modes and performing welding.

[0014] For each fault mode, welding parameter waveforms are collected and features are extracted to obtain the fault parameters;

[0015] Calculate the difference vector between each fault parameter and the health parameter, and use it as the feature offset vector for that fault; store all feature offset vectors and their corresponding fault descriptions.

[0016] In a preferred embodiment of the present invention, in S3, the sampling rate of the welding parameter waveform acquisition is not less than 10kHz.

[0017] In a preferred embodiment of the present invention, in S3, extracting real-time features includes calculating at least one statistical feature among the effective value, standard deviation, skewness, and kurtosis of the welding parameter waveform, as well as the spectral energy distribution feature obtained by fast Fourier transform.

[0018] In a preferred embodiment of the present invention, in S4, the similarity is calculated using Mahalanobis distance or cosine similarity algorithm to generate an overall score for the stability of the process.

[0019] In a preferred embodiment of the present invention, in S4, the reason for diagnosing welding defects is to calculate the directional similarity between the real-time offset vector and each feature offset vector in the fault mode vector library, and match them by the cosine value of the vector angle; to screen the fault mode with the best matching degree in a welding defect system and its monitoring method based on welding parameter stability, and to generate an early warning report based on the similarity.

[0020] In a preferred embodiment of the present invention, typical failure modes include poor wire feeding, wear of the contact tip, abnormal protective gas, and power grid fluctuations.

[0021] A welding defect system based on welding parameter stability includes a frequency data acquisition unit configured to acquire the raw waveforms of welding current and arc voltage in real time at a sampling rate of not less than 10 kHz.

[0022] The data processing and storage unit is connected to a frequency data acquisition unit of a welding defect system and its monitoring method based on welding parameter stability, and is configured to store a health parameter library and a fault mode vector library, and to run feature extraction, vector calculation and similarity matching algorithms.

[0023] The real-time diagnosis and early warning unit is integrated into or connected to the data processing and storage unit and is configured to perform real-time similarity calculation, offset vector matching, and generate an early warning report containing diagnostic results and processing suggestions.

[0024] The human-computer interaction unit is connected to the real-time diagnosis and early warning unit and is configured to display system status, real-time monitoring data, early warning information and diagnostic reports.

[0025] In a preferred embodiment of the present invention, the frequency data acquisition unit includes a welding current sensor, an arc voltage sensor, and an automated welding robot or special machine equipped with an EtherCAT or Profinet communication interface.

[0026] This invention addresses the shortcomings of the prior art and has the following beneficial effects:

[0027] (1) This invention provides a welding defect system and its monitoring method based on the stability of welding parameters. By constructing a fault mode vector library, it stores the characteristic offset vectors of various typical faults such as poor wire feeding and abnormal shielding gas. These vectors accurately record the direction and magnitude of the deviation of multidimensional features from the healthy baseline. Different fault root causes produce unique disturbances to electrical parameters. By calculating the real-time offset vector and matching the direction with the vectors in the library, the system can accurately identify the cause of the defect. For example, it can distinguish between equipment problems such as welding torch wear, process parameter problems such as insufficient gas flow, or material problems such as workpiece contamination. Compared with the existing technology, conventional monitoring only provides generalized alarms and lacks root cause diagnosis, relying on manual investigation. This invention provides maintenance personnel with precise action guidelines, realizes intelligent predictive maintenance, shortens downtime and optimizes resource utilization.

[0028] (2) This invention provides a welding defect system and its monitoring method based on the stability of welding parameters. The original waveforms of welding current and arc voltage are acquired in real time through a high-frequency data acquisition system, and multi-dimensional statistical features such as effective value, standard deviation, skewness, kurtosis and spectral energy distribution features are extracted. These features can capture microsecond-level transient changes during welding, such as droplet transfer and arc disturbance. Through this deep feature extraction, the health status of the process is digitized and a quantifiable health baseline is established. The similarity between the current welding parameters and the health parameter library is calculated in real time. Once the similarity is lower than the preset threshold, an early warning can be given that the product has a defect risk. Compared with the existing technology, traditional methods often rely on low sampling rate or simple parameter monitoring, which cannot identify subtle stability fluctuations, resulting in a delayed warning. A further effect is to achieve predictive quality control, reduce scrap rate, and improve production efficiency and quality consistency. Attached Figure Description

[0029] To more clearly illustrate the technical solutions 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 recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a three-dimensional structural diagram of a preferred embodiment of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein. Therefore, the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0033] like Figure 1 As shown, a welding defect monitoring method based on welding parameter stability includes the following steps:

[0034] S1: Obtain the waveform characteristics of welding parameters under both normal and faulty equipment conditions, and construct a health parameter library and a fault parameter library based on the waveform characteristics of the welding parameters;

[0035] S2: Compare the health parameter library obtained in S1 with the fault parameter library to obtain the feature offset vector between the health parameters and the fault parameters, and obtain the fault mode vector library based on the feature offset vector.

[0036] S3: Real-time monitoring of welding parameters during the production process, acquisition of welding parameter waveforms during the welding process, extraction of real-time welding features, and generation of real-time welding parameters;

[0037] S4: Based on the real-time welding parameters obtained in S3, calculate the similarity between them and the health parameter library obtained in S1. When the similarity is lower than the preset threshold, calculate the real-time offset vector and match the real-time offset vector with the fault mode vector library obtained in S2 to diagnose the cause of welding defects.

[0038] It should be noted that this invention digitizes the health status of the process by acquiring and deeply analyzing the dynamic waveforms of current and voltage during the welding process at high frequency, thereby predicting potential quality risks of the product and ultimately achieving accurate identification of the root cause of defects.

[0039] By establishing a healthy baseline in the welding process—that is, when the system is in an ideal state in terms of equipment and process—a massive amount of raw waveforms of welding electrical parameters are collected, and a series of multi-dimensional statistical characteristics, namely volatility, symmetry, and spectral energy distribution, are extracted from them. A digital golden parameter representing absolute health is constructed, and the characteristics that a stable welding process should have are defined in turn. Any fluctuation that deviates from this baseline indicates that the stability of the process is being compromised, thus providing the earliest and most forward-looking signal of potential product defects. This is equivalent to establishing a quantifiable "electrocardiogram" standard for the welding process.

[0040] Furthermore, intelligent diagnosis based on a fault mode vector library is introduced. By recognizing that different root causes of faults, such as equipment, processes, and materials, will cause drastically different disturbances to the physical characteristics of the welding arc, which are reflected in the waveform characteristics of current and voltage. For example, wear of the wire feeding mechanism and equipment problems will lead to uneven wire feeding, producing periodic small spikes or drops in the current waveform, characterized by an increase in the current standard deviation and specific high-frequency energy. Insufficient shielding gas flow and process parameter problems will disrupt the stability of the arc, affecting voltage fluctuations and the spectral characteristics of the sound signal. If there is oil or other material on the workpiece surface, the oil will instantly vaporize at the high temperature of the arc, causing the arc to suddenly explode and contract, which will manifest as instantaneous and violent voltage and current mutations in electrical parameters.

[0041] During the initialization phase, the system artificially induces various typical faults through experiments and records the unique feature offset vector of each fault, that is, the direction and magnitude of the deviation of the multidimensional features caused by the fault from the healthy baseline. This makes the fault mode vector library a knowledge graph for the system to perform intelligent diagnosis.

[0042] In actual production monitoring, when the system detects a decrease in process stability by calculating the similarity between real-time parameters and health parameters, that is, when it predicts an increased risk of product defects, it will initiate a root cause diagnosis program to calculate the current real-time offset vector and perform directional matching with various benchmark patterns in the fault mode vector library.

[0043] If the direction of the real-time offset vector is highly consistent with the vector direction of the wire feeding disorder mode, the system can determine with high confidence that this is a performance degradation problem of the wire feeder in the equipment itself; if it matches the abnormal protective gas mode, it points to a problem with the process parameter settings or execution; if it matches the instantaneous change mode unique to workpiece contamination, it can be attributed to a material cleanliness problem.

[0044] Through this vector direction matching, the system successfully shifts the focus of monitoring from what phenomenon occurred to why this phenomenon occurred, thereby providing maintenance personnel with extremely accurate action guidelines. This achieves a leap from vague alarms to precise diagnosis, truly realizing intelligent predictive maintenance and quality control.

[0045] S1: Obtain the waveform characteristics of welding parameters under both normal and faulty equipment conditions, and construct a health parameter library and a fault parameter library based on the waveform characteristics of the welding parameters;

[0046] In a preferred embodiment of the present invention, in S1, establishing a health parameter database specifically involves acquiring the original waveforms of welding current and arc voltage through a welding defect system and its monitoring method based on welding parameter stability when the welding equipment is in good condition.

[0047] Extracting multidimensional statistical and spectral features from the original waveform;

[0048] Based on the feature data set of multiple qualified welds, the mean vector and covariance matrix are calculated to form a parameter library as a health benchmark.

[0049] It should be noted that in this step, a health parameter library and a fault parameter library are constructed for comparison. In constructing the health parameter library, the welding equipment is in the best condition for descaling, that is, the equipment has been fully maintained, key components such as the contact nozzle and wire feed wheel are brand new, and the process parameters have been optimized and verified to stably produce non-destructive testing level 1 welds.

[0050] At this point, the original waveform signals of welding current and arc voltage are continuously acquired through a high-frequency data acquisition system with a sampling rate of no less than 10kHz. This is because the welding arc is a rapidly changing physical process, and its microsecond-level transient characteristics, such as droplet transfer and arc disturbance, contain key information reflecting the stability of the process. Therefore, a high sampling rate is required. Conventional sampling of tens or hundreds of times per second cannot capture these details. Only high-frequency sampling can obtain raw data sufficient for in-depth waveform analysis.

[0051] After acquiring massive amounts of raw waveforms, multidimensional indicators that characterize the essential features of the waveforms are extracted. These include two main categories of features: time-domain statistical features and frequency-domain energy features. Time-domain features are calculated directly from the waveform sequence, such as: RMS (equivalent to the mean, reflecting the overall energy level), standard deviation (reflecting the volatility of the waveform, a direct measure of stability), skewness (reflecting the asymmetry of the waveform distribution), and kurtosis (reflecting the steepness of the waveform distribution). Specifically, RMS is equivalent to the mean, reflecting the overall energy level; standard deviation reflects the volatility of the waveform, a direct measure of stability; skewness reflects the asymmetry of the waveform distribution; and kurtosis reflects the steepness of the waveform distribution.

[0052] The waveform shape is described from different perspectives by the above features. The frequency domain features are obtained by performing a fast Fourier transform on the waveform data, which converts the signal from the time axis to the frequency axis, thereby analyzing its spectral energy distribution characteristics, such as the dominant frequency (the frequency with the most concentrated energy) and the energy ratio of different frequency bands. The dominant frequency is the frequency with the most concentrated energy. Changes in the stability of the electric arc will directly lead to changes in its acoustic characteristics and energy release frequency, which will then be reflected in the spectrum.

[0053] Based on feature extraction from hundreds or thousands of qualified welding processes, the system obtains a massive feature dataset. Through statistical analysis of this dataset, its mean vector and covariance matrix are calculated, ultimately forming a condensed golden parameter library representing an absolute healthy state. This golden parameter library not only contains the average level of each feature, but also the interrelationships between features described by the covariance matrix. Together, they mathematically define a healthy region in a multidimensional feature space. Any welding process that falls into or approaches this region can be considered to be in good condition.

[0054] At the same time, a fault parameter library needs to be constructed. This operation requires artificially and individually inducing various known typical fault modes in a controlled experimental environment. For example, simulating poor wire feeding by slightly obstructing the wire feeding mechanism, simulating contact tip wear by using a severely worn contact tip, simulating protective gas abnormality by adjusting the gas flow rate, and simulating power grid fluctuations by introducing interference.

[0055] For each of the above failure modes, multiple welding experiments are conducted, and the same high sampling rate process and feature extraction method as used in building the health database are employed to obtain the set of failure parameters under that failure state. This step is equivalent to establishing a detailed case database for the system, recording the unique symptom manifestations of each disease.

[0056] S2: Compare the health parameter library obtained in S1 with the fault parameter library to obtain the feature offset vector between the health parameters and the fault parameters, and obtain the fault mode vector library based on the feature offset vector.

[0057] In a preferred embodiment of the present invention, in S2, establishing the fault mode vector library specifically involves artificially simulating various typical fault modes and performing welding.

[0058] For each fault mode, welding parameter waveforms are collected and features are extracted to obtain the fault parameters;

[0059] Calculate the difference vector between each fault parameter and the health parameter, and use it as the feature offset vector for that fault; store all feature offset vectors and their corresponding fault descriptions.

[0060] It should be noted that in this step, each fault parameter obtained in S1, i.e. the feature vector of a certain fault mode, is subtracted from the health parameter, i.e. the mean vector of the health parameter library. The difference vector is calculated, and then the feature offset vector is obtained. This vector not only contains the magnitude of each feature's deviation from the health benchmark, but more importantly, it accurately records whether the deviation of each feature is positively increasing or negatively decreasing.

[0061] Different fault modes, due to their different physical mechanisms, will cause drastically different changes in welding electrical signal parameters. For example, poor wire feeding will cause periodic and instantaneous violent fluctuations in current, manifested as a significant increase in the current standard deviation and the energy of certain high-frequency components. On the other hand, wear of the contact tip will disrupt arc stability, which may manifest as an increase in the voltage standard deviation, while the fundamental frequency of arc combustion will shift to a lower frequency. These differences are reflected in the characteristic offset vector, which is a completely different vector direction. Therefore, each characteristic offset vector uniquely points to a specific fault root cause.

[0062] By associating and storing the feature offset vectors of all known fault modes with their corresponding fault descriptions, such as poor wire feeding, a fault mode vector library is formed. This vector library is essentially a sophisticated diagnostic rule library. When the system encounters an anomaly later, it can match the current anomaly mode with the rules in the library, just like a doctor comparing medical records, and thus make an informed diagnosis rather than guesswork.

[0063] S3: Real-time monitoring of welding parameters during the production process, acquisition of welding parameter waveforms during the welding process, extraction of real-time welding features, and generation of real-time welding parameters;

[0064] In a preferred embodiment of the present invention, in S3, the sampling rate of the welding parameter waveform acquisition is not less than 10kHz.

[0065] In a preferred embodiment of the present invention, in S3, extracting real-time features includes calculating at least one statistical feature among the effective value, standard deviation, skewness, and kurtosis of the welding parameter waveform, as well as the spectral energy distribution feature obtained by fast Fourier transform.

[0066] It should be noted that in this step, the high-frequency data acquisition unit continuously captures the original waveforms of welding current and arc voltage at a rate of no less than 10kHz. Ensuring that the sampling rate remains consistent with that used when establishing the knowledge base is crucial; otherwise, the extracted features will not be comparable, leading to subsequent diagnostic failures. Next, the data processing unit performs the same feature extraction calculations on the real-time incoming waveform data blocks, for example, with an analysis window of 100 milliseconds. This involves calculating the same effective values, standard deviations, and frequency spectrum features of the time-domain statistical features. The feature values ​​calculated in real time collectively constitute the real-time welding parameters representing the instantaneous state of the current welding process, thereby providing the system with a continuous quantitative data stream of production process parameter characteristics.

[0067] S4: Based on the real-time welding parameters obtained in S3, calculate the similarity between them and the health parameter library obtained in S1. When the similarity is lower than the preset threshold, calculate the real-time offset vector and match the real-time offset vector with the fault mode vector library obtained in S2 to diagnose the cause of welding defects.

[0068] In a preferred embodiment of the present invention, in S4, the similarity is calculated using Mahalanobis distance or cosine similarity algorithm to generate an overall score for the stability of the process.

[0069] In a preferred embodiment of the present invention, in S4, the reason for diagnosing welding defects is to calculate the directional similarity between the real-time offset vector and each feature offset vector in the fault mode vector library, and match them by the cosine value of the vector angle; to screen the fault mode with the best matching degree in a welding defect system and its monitoring method based on welding parameter stability, and to generate an early warning report based on the similarity.

[0070] In a preferred embodiment of the present invention, typical failure modes include poor wire feeding, wear of the contact tip, abnormal protective gas, and power grid fluctuations.

[0071] It should be noted that in this step, the real-time parameters generated in S3 are compared with the health parameter database established in S1 to calculate the similarity between the two. Mahalanobis distance or cosine similarity algorithms are used here because these algorithms can fully consider the correlation between multidimensional features, represented by the covariance matrix of the health database, and more accurately reflect the true proximity in multidimensional space than simple Euclidean distance. The calculated similarity score, a value between 0% and 100%, provides an overall score for process stability, which is equivalent to a rapid assessment of the current welding process's health status.

[0072] When the similarity falls below a preset threshold of 90%, or shows a continuous downward trend, the system's judgment process malfunctions, triggering a more refined root cause diagnosis procedure. At this point, the system calculates a real-time offset vector, which is the difference vector between the current real-time parameters and the baseline in the health parameter database. Subsequently, the system performs directional similarity matching between this real-time offset vector and each baseline offset vector in the fault mode vector library constructed by S2. The matching algorithm calculates the cosine of the angle between the vectors; the closer the cosine value is to 1, the more consistent the directions of the two vectors, meaning the more similar the current abnormal mode is to a known fault mode.

[0073] The system filters out the fault modes with the highest matching degree (over 90%) and combines this with the previously calculated overall similarity to generate an accurate early warning report. The report no longer provides vague system anomalies but specific diagnostic results. For example, the report shows a 92% match with the wire feeding obstruction mode, with the similarity dropping to 85%. The recommended action is to immediately check the wire feeding hose and the clamping force of the wire feeding rollers. This frees maintenance personnel from tedious troubleshooting, directly targeting the most likely root cause, greatly improving maintenance efficiency, avoiding production interruptions due to misdiagnosis, and achieving true predictive intelligent maintenance.

[0074] A welding defect system based on welding parameter stability includes a frequency data acquisition unit configured to acquire the raw waveforms of welding current and arc voltage in real time at a sampling rate of not less than 10 kHz.

[0075] The data processing and storage unit is connected to a frequency data acquisition unit of a welding defect system and its monitoring method based on welding parameter stability, and is configured to store a health parameter library and a fault mode vector library, and to run feature extraction, vector calculation and similarity matching algorithms.

[0076] The real-time diagnosis and early warning unit is integrated into or connected to the data processing and storage unit and is configured to perform real-time similarity calculation, offset vector matching, and generate an early warning report containing diagnostic results and processing suggestions.

[0077] The human-computer interaction unit is connected to the real-time diagnosis and early warning unit and is configured to display system status, real-time monitoring data, early warning information and diagnostic reports.

[0078] In a preferred embodiment of the present invention, the frequency data acquisition unit includes a welding current sensor, an arc voltage sensor, and an automated welding robot or special machine equipped with an EtherCAT or Profinet communication interface.

[0079] It should be noted that the high-frequency data acquisition unit consists of welding current sensors (Hall effect sensors), arc voltage sensors, and automated welding equipment with real-time industrial Ethernet communication interfaces, namely EtherCAT or Profinet. The requirement of a high sampling rate of ≥10kHz dictates that these sensors and interfaces must have high bandwidth and low latency characteristics to ensure that the original waveform information can be captured and transmitted without distortion.

[0080] The data processing and storage unit is an industrial-grade computer or a high-performance edge computing server, mainly used to run complex signal processing algorithms and extract multi-dimensional features from raw waveforms in real time; it securely stores two core knowledge bases, a health parameter base and a fault mode vector base, which contain the system's knowledge and experience; it carries all the core computing algorithms, including feature extraction, similarity calculation, vector matching, etc. The computing power and reliability of this unit directly determine the speed and accuracy of system diagnosis.

[0081] The real-time diagnosis and early warning unit is integrated into the data processing and storage unit as a software module. This unit continuously receives real-time feature values ​​from the data processing unit and sequentially executes logical processes such as similarity calculation, threshold judgment, offset vector calculation and matching. Finally, it is responsible for generating an early warning report containing specific diagnostic conclusions and handling suggestions, and sending it to operators via a network interface. It is a key link in transforming data into actionable information.

[0082] The human-machine interface unit (HMI) is an industrial touchscreen HMI or a computer client connected to the office network. This unit clearly displays the system status, real-time monitoring data, historical records, and most importantly, early warning and diagnostic reports in a visual manner, including trend curves, dashboards, and alarm lists. This allows operators and maintenance personnel to intuitively understand the health status of the production process and take precise actions based on the system's intelligent diagnostic results.

[0083] In this invention, with the welding equipment, process parameters, and base material all in a confirmed ideal state, a high-frequency data acquisition unit is activated. This unit continuously acquires the raw dynamic waveforms of the arc voltage and welding current during the welding process at a sampling frequency far higher than conventional control, specifically 100 kHz. Subsequently, the feature extraction engine processes the massive amount of raw waveform data to calculate a set of multidimensional statistical features defining process stability. These features include not only time-domain indicators such as mean, standard deviation, skewness, and kurtosis, but also frequency-domain features such as energy distribution in specific frequency bands and advanced indicators like waveform symmetry. The resulting multidimensional feature set, representing an absolute healthy state, is defined as the digital health baseline of the welding process and persistently stored in the system database.

[0084] After initialization, various known typical faults are artificially and systematically introduced, including but not limited to equipment components such as wire feeding mechanisms and progressive wear of contact tips, deliberate deviations in process parameters such as protective gas flow rates and arc voltage settings, and simulations of material conditions such as oil stains and oxide films on the workpiece surface. For each induced fault, the system fully records the electrical parameter waveforms during its occurrence and calculates the feature offset vector relative to the healthy baseline. This vector precisely quantifies the direction and magnitude of deviation of each multidimensional feature under this specific fault. All validated fault modes and their corresponding feature offset vectors together constitute the system's diagnostic knowledge base.

[0085] During actual welding production, the system enters real-time monitoring mode, the high-frequency data acquisition unit works continuously, and the feature extraction engine synchronously calculates the multi-dimensional statistical features of the real-time welding current and voltage waveforms. The stability assessment module then calculates the similarity or deviation of these real-time features from the healthy baseline. When the deviation exceeds the preset threshold of 90%, it indicates that the process stability is impaired and there is a quality risk.

[0086] At this point, the intelligent diagnostic engine is triggered. It calculates the current real-time feature offset vector and then matches this vector with various reference vectors stored in the fault mode vector library. The core of the matching algorithm is to analyze the consistency of the vector direction, not just the magnitude. By calculating the cosine of the angle between the vectors or similar directional similarity measures, the engine identifies the known fault type that best matches the current offset mode.

[0087] The diagnostic results will clearly identify the potential root causes of the stability degradation, including equipment, processes, or materials, and can be further refined to specific reasons, such as poor wire feeding or suspected protective gas abnormalities. These results, along with confidence level indicators, are output to the human-machine interface, providing maintenance personnel with precise intervention guidance. This achieves a leap from anomaly warning to root cause identification, realizing the goals of predictive maintenance and intelligent quality control.

[0088] Based on the preferred embodiments of the present invention described above, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A welding defect monitoring method based on stability of welding parameters, characterized by, The method comprises the following steps: S1: Obtain welding parameter waveform features in normal and fault states of the equipment respectively, and construct a health parameter library and a fault parameter library according to the welding parameter waveform features; S2: Compare the health parameter library and the fault parameter library obtained in S1 to obtain a feature offset vector between the health parameters and the fault parameters, and obtain a fault mode vector library according to the feature offset vector; S3: Real-time monitor welding parameters in the production process, obtain welding parameter waveforms in the welding process, extract real-time welding features, and generate real-time welding parameters; S4: Calculate the similarity of the real-time welding parameters obtained in S3 and the health parameter library obtained in S1, and when the similarity is lower than a preset threshold, calculate a real-time offset vector, match the real-time offset vector with the fault mode vector library obtained in S2, and diagnose the cause of the welding defect.

2. The welding defect monitoring method based on welding parameter stability according to claim 1, characterized in that: In the S1, the health parameter library is established by collecting original waveforms of welding current and arc voltage through a frequency data acquisition system based on welding parameter stability when the welding equipment is in good condition; Multi-dimensional statistical features and spectral features are extracted from the original waveforms; Based on the feature data set of multiple qualified weldings, the mean vector and the covariance matrix are calculated to form the parameter library as the health benchmark.

3. The method of claim 1, wherein: In the S2, the fault mode vector library is established by simulating multiple typical fault modes artificially and welding; Welding parameter waveforms are collected for each fault mode and features are extracted to obtain fault parameters; The difference vector of each fault parameter and the health parameter is calculated as the feature offset vector of the fault; all feature offset vectors and corresponding fault descriptions are stored.

4. The method of claim 1, wherein: In the S3, the sampling rate of the welding parameter waveform collection is not less than 10 kHz.

5. The method of claim 1, wherein: In the S3, the extraction of real-time features includes calculation of at least one of the statistical features of the effective value, standard deviation, skewness, kurtosis of the welding parameter waveform, and the frequency spectrum energy distribution feature obtained by fast Fourier transform.

6. The method of claim 1, wherein: In the S4, the similarity is calculated by using Mahalanobis distance or cosine similarity algorithm to generate the overall score of process stability.

7. The method of claim 1, wherein: In the S4, the cause of the welding defect is diagnosed by calculating the directional similarity of the real-time offset vector and each feature offset vector in the fault mode vector library, matching by the vector angle cosine value; screening the fault mode with the highest matching degree, and generating a warning report in combination with the similarity.

8. The method of claim 3, wherein: The typical fault modes include poor wire feeding, electrode nozzle wear, abnormal shielding gas, and power grid fluctuation.

9. A welding defect system based on stability of welding parameters based on the monitoring method of any one of claims 1-8, characterized by, The frequency data acquisition unit is configured to collect original waveforms of welding current and arc voltage in real time at a sampling rate of not less than 10 kHz; The data processing and storage unit is connected to the frequency data acquisition unit and is configured to store the health parameter library and the fault mode vector library, and run feature extraction, vector calculation and similarity matching algorithms. a real-time diagnosis and early warning unit integrated with or connected to the data processing and storage unit, and configured to perform real-time similarity calculation, offset vector matching, and generate early warning reports containing diagnosis results and processing suggestions; a human-computer interaction unit connected to the real-time diagnosis and early warning unit, and configured to display system status, real-time monitoring data, early warning information, and diagnosis reports.

10. The welding defect system based on stability of welding parameters as claimed in claim 9 wherein: The frequency data acquisition unit includes a welding current sensor, an arc voltage sensor, and an automated welding robot or a special machine with an EtherCAT or Profinet communication interface.

Images