Weld quality prediction and process optimization method based on welding cloud platform

Abstract

The application provides a welding seam quality prediction and process optimization method based on a welding cloud platform. By combining preset target requirements with welding seam quality data, initial optimization parameters for welding seam process parameters are determined to ensure that the optimization direction is consistent with actual production requirements, providing a scientific basis for real-time adjustment of welding equipment. The welding equipment controller performs real-time optimization according to the initial optimization parameters, and real-time collection of current welding seam quality data during the optimization process is performed. The current welding seam quality data is fed back to the welding cloud platform to update the welding seam quality prediction model, quickly respond to quality fluctuations during the welding process, avoid defect expansion, and improve welding quality stability. By feeding back the current welding seam quality data during the optimization process to the cloud platform and updating the model, the model can continuously adapt to changes in the welding scene, gradually improve the accuracy of prediction and optimization, and realize iterative upgrading of the technical scheme.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology for welding, and in particular to a method for predicting weld quality and optimizing processes based on a welding cloud platform. Background Technology

[0002] In modern equipment manufacturing industries such as steel structure manufacturing, welding quality directly determines the load-bearing capacity and service life of components. Especially in large-scale engineering construction projects, weld defects may cause serious safety hazards. Traditional welding quality control mainly relies on the experience judgment of operators and post-event inspection, which has problems such as slow response, low precision, and blind process adjustment, making it difficult to meet the needs of large-scale and intelligent production.

[0003] With the development of cloud computing and artificial intelligence technologies, welding quality control technology based on cloud platforms has gradually become a research hotspot. Existing technologies include patents proposing application schemes for welding cloud platforms. For example, Chinese invention patent CN113960114B, "An Online Analysis System and Method for Welding Process Quality Based on a Cloud Server," uses a cloud server to process and analyze electrical signals such as welding current and voltage, providing characteristic index parameters for quality analysis. However, this technology only focuses on online data analysis and does not achieve advanced prediction of weld quality or proactive optimization of process parameters, lacking a closed-loop control mechanism. Another example is Chinese invention patent CN118875566A, "A Method for Evaluating the Quality of Steel Structure Welding Process Based on Big Data Processing," which uses multimodal data fusion to construct a quality evaluation model. However, this method does not fully utilize the computing power advantages of cloud platforms, resulting in low model training efficiency and a single process optimization strategy, making it difficult to adapt to the personalized needs of different welding scenarios. Summary of the Invention

[0004] This invention provides a method for weld quality prediction and process optimization based on a welding cloud platform to solve the problems mentioned in the background art.

[0005] A method for weld quality prediction and process optimization based on a welding cloud platform includes:

[0006] S1: Based on pre-deployed sensors, welding process parameters, environmental parameters and weld feature data are collected. After data preprocessing of the welding process parameters, environmental parameters and weld feature data, multi-source data is obtained and uploaded to the welding cloud platform.

[0007] S2: Input the multi-source data into the weld quality prediction model pre-trained based on the welding cloud platform to obtain weld quality data;

[0008] S3: Based on weld quality data and combined with preset target requirements, determine the initial optimization parameters for weld process parameters;

[0009] S4: The welding equipment controller performs real-time optimization according to the initial optimization parameters and collects the current weld quality data in real time during the optimization process. The current weld quality data is then fed back to the welding cloud platform to update the weld quality prediction model.

[0010] Preferably, in step S1, after preprocessing the welding process parameters, environmental parameters, and weld characteristic data, multi-source data is obtained, including:

[0011] The welding process parameters, environmental parameters, and weld characteristic data are processed by removing abnormal data, supplementing missing values, and deduplicating duplicate data to obtain the processed data.

[0012] Data standardization is performed on the processed data to obtain multi-source data.

[0013] Preferably, in step S1, uploading multi-source data to the welding cloud platform includes:

[0014] By deploying edge computing nodes near the welding station, the multi-source data is encapsulated and formatted into data frames.

[0015] Based on the communication link in the edge computing node, the data frame is sent to the dedicated data receiving port of the welding cloud platform.

[0016] Preferably, in step S2, the multi-source data is input into a weld quality prediction model pre-trained based on a welding cloud platform to obtain weld quality data, including:

[0017] The multi-source data are grouped and integrated according to the preset time series rules to obtain an input dataset that matches the input dimension of the weld quality prediction model.

[0018] The input dataset is fed into the weld quality prediction model, which automatically performs feature mining and computational analysis on the input dataset, identifies the correlation between multi-source data and weld quality, and outputs weld quality data.

[0019] Preferably, in step S3, the initial optimization parameters for the weld process parameters are determined based on weld quality data and in conjunction with preset target requirements, including:

[0020] The preset target requirements are first decomposed to obtain the first target, which includes the quality compliance target, the production efficiency target, and the energy consumption cost target. The first target is then second decomposed to obtain quantifiable sub-targets.

[0021] Based on the scaling method, a first judgment matrix and a second judgment matrix are generated to determine the importance between any two first objectives and the importance between any two quantifiable sub-objectives, and the first maximum eigenvalue and the second maximum eigenvalue of the first judgment matrix and the second judgment matrix are obtained respectively.

[0022] Consistency verification is performed based on the first and second largest eigenvalues. If it passes, the eigenvectors corresponding to the first and second judgment matrices are calculated based on the eigenvalue method. The eigenvectors are normalized and used as the initial sovereign weight of the first target and the initial sub-weights of the quantifiable sub-targets. Based on the initial sovereign weights and the initial sub-weights, the initial weight scheme is obtained. Otherwise, the scale value is adjusted and the latest judgment matrix is ​​regenerated until the consistency verification passes.

[0023] Based on historical weighting schemes and their corresponding historical performance data, the theoretical performance data under the initial weighting scheme is predicted, the deviation between the theoretical performance data and the historical performance data is obtained, and the deviation is fitted and analyzed by linear regression algorithm to obtain the correction coefficient of the corresponding weight. Based on the correction coefficient, the initial main weight and initial sub-weight are corrected to obtain the target weight. Based on the first target, quantifiable sub-targets and target weight, the optimized target system is established.

[0024] Based on the defect types and defect association parameters in the weld quality data, the process parameter-defect association rule library built into the welding cloud platform is retrieved, and the process parameter set corresponding to the defect type and defect association parameter is obtained by reverse mapping. At the same time, parameter constraint boundaries are constructed based on the characteristics of steel structure welding materials and the rated parameters of welding equipment.

[0025] Based on the optimization target system and the parameter constraint boundary, the process parameter set is iteratively calculated to obtain the optimal process parameter scheme. Based on the optimal process parameter scheme, the initial optimization parameters for the weld process parameters are determined.

[0026] Preferably, consistency verification is performed based on the first and second largest eigenvalues, including:

[0027] Obtain the first difference between the first or second largest eigenvalue and the order of the corresponding judgment matrix, and the second difference between the order of the corresponding judgment matrix and 1;

[0028] The consistency index is determined based on the ratio of the first difference and the second difference. If the consistency index is less than the preset value, the consistency verification is determined to be successful; otherwise, the consistency verification is determined to be unsuccessful.

[0029] Preferably, the optimal process parameter scheme is obtained by iterative calculation of the process parameter set, including:

[0030] The process parameter set is iteratively calculated, and multiple sets of candidate process parameter schemes are generated based on the iteration results.

[0031] Virtual welding simulations were performed on multiple candidate process parameter schemes based on digital twin models, and the optimal process parameter scheme was selected based on the simulation results.

[0032] Preferably, in step S4, feeding back the current weld quality data to the welding cloud platform to update the weld quality prediction model includes:

[0033] Feature extraction is performed on the current weld quality data according to quality characteristics, parameter characteristics, scenario characteristics and operation characteristics, and a four-dimensional traceability sample of quality-parameter-scenario-operation is formed based on the extraction results;

[0034] The four-dimensional traceability sample is input into a pre-established welding Bayesian network. The root node layer is used to analyze process parameters and environmental parameters, the intermediate hidden node layer is used to analyze indirect parameters, and the leaf node layer is used to analyze defect type and defect level. Based on the analysis results, the leaf node layer is used as the cause to reason backward to the intermediate hidden node layer and the root node layer to obtain the posterior causal probability that the current parameter abnormality causes the defect to occur. Based on the sum of the posterior causal probabilities that all parameter abnormalities cause the same defect to occur, the causal contribution of the current parameter abnormality to the defect is obtained. Based on the causal contribution, the core parameters for the occurrence of the defect are determined.

[0035] The causal relationship between core parameters and defects is divided into immutable and variable relationships according to whether it is affected by the scenario. A general critical interval is determined for the continuous process parameters in the immutable relationship, and common features are extracted for the discrete process parameters in the variable relationship. Based on the general critical interval and common features, a scenario-free general rule is established.

[0036] The key positions corresponding to the core parameters are obtained from the weld quality prediction model and marked to obtain the local module for this update. The local module for this update is updated for the first time, and the general rule without scenarios is added to the data-scarce model scenario layer for the second update. Based on the results of the first and second updates, the parameters of the weld quality prediction model are optimized to obtain the latest prediction model.

[0037] Preferably, starting from the leaf node layer as the cause, backward reasoning is performed towards the intermediate hidden node layer and the root node layer to obtain the posterior causal probability that the current parameter anomaly causes the defect to occur, specifically:

[0038] ;

[0039] in, This represents the posterior causal probability that an abnormality in the current parameters leads to the occurrence of a defect. This indicates the probability of a defect occurring when the current parameter is abnormal. This indicates the probability that the current parameter is abnormal. This indicates the probability of a defect occurring.

[0040] Preferably, determining a general critical range for continuous process parameters in an invariant relationship includes:

[0041] The ratio of the specific values ​​of continuous process parameters in all scenarios to the critical reference state values ​​in the scenarios is used as the parameter deviation.

[0042] When the causal contribution of a continuous process parameter is greater than a preset contribution, obtain all values ​​of the parameter deviation, and then obtain the maximum and minimum values ​​from all values ​​to determine the general critical interval.

[0043] Compared with the prior art, the present invention has achieved the following beneficial effects:

[0044] Welding process parameters, environmental parameters, and weld feature data are collected by pre-deployed sensors. After data preprocessing, multi-source data is obtained and uploaded to the welding cloud platform. This standardization and accuracy of the multi-source data enable centralized retrieval, sharing, and subsequent reuse, providing sufficient data support for model training and parameter optimization. By inputting the multi-source data into a weld quality prediction model pre-trained on the welding cloud platform, weld quality data is obtained, improving the objectivity and accuracy of weld quality assessment. This model is suitable for large-scale, standardized production scenarios. Based on the weld quality data and pre-set target requirements, the model can be further optimized. Determining the initial optimization parameters for the weld process ensures that the optimization direction aligns with actual production needs, providing a scientific basis for real-time adjustments to the welding equipment. The welding equipment controller performs real-time optimization according to the initial optimization parameters and collects current weld quality data during the optimization process. This data is then fed back to the welding cloud platform to update the weld quality prediction model, enabling rapid response to quality fluctuations during the welding process, preventing defect expansion, and improving weld quality stability. By feeding back the current weld quality data during the optimization process to the cloud platform and updating the model, the model can continuously adapt to changes in the welding scenario, gradually improving the accuracy of prediction and optimization, and achieving iterative upgrades of the technical solution.

[0045] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0046] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0047] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0048] Figure 1This is a flowchart of the weld quality prediction and process optimization method based on a welding cloud platform in an embodiment of the present invention;

[0049] Figure 2 This is a flowchart illustrating the process of obtaining weld quality data in an embodiment of the present invention. Detailed Implementation

[0050] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0051] Example 1: This embodiment of the invention provides a method for weld quality prediction and process optimization based on a welding cloud platform, such as... Figure 1 As shown, it includes:

[0052] S1: Based on pre-deployed sensors, welding process parameters, environmental parameters and weld feature data are collected. After data preprocessing of the welding process parameters, environmental parameters and weld feature data, multi-source data is obtained and uploaded to the welding cloud platform.

[0053] S2: Input the multi-source data into the weld quality prediction model pre-trained based on the welding cloud platform to obtain weld quality data;

[0054] S3: Based on weld quality data and combined with preset target requirements, determine the initial optimization parameters for weld process parameters;

[0055] S4: The welding equipment controller performs real-time optimization according to the initial optimization parameters and collects the current weld quality data in real time during the optimization process. The current weld quality data is then fed back to the welding cloud platform to update the weld quality prediction model.

[0056] In this embodiment, welding process parameters refer to the core technical parameters that can be adjusted during the welding process, including welding current, voltage, welding speed, shielding gas flow rate, and wire feed speed.

[0057] In this embodiment, environmental parameters refer to the environmental conditions at the welding site, including ambient temperature, humidity, wind speed, etc.

[0058] In this embodiment, weld feature data refers to parameters that describe the appearance and internal state of the weld, including weld appearance dimensions (such as weld width, weld depth, and weld reinforcement height) and internal defect information (such as the location and size of defects such as porosity, cracks, and incomplete penetration).

[0059] In this embodiment, the weld quality prediction model is a smart analysis model (such as a neural network, machine learning model, etc.) pre-trained with historical welding data. It can take in multi-source welding data (process, environment, weld characteristics) and output weld quality-related results (such as defect type, pass / fail level, risk level, etc.).

[0060] In this embodiment, the initial optimization parameters are initial values ​​of process parameters determined based on weld quality prediction results and preset targets (such as quality, efficiency, and cost targets) to guide the adjustment of welding equipment.

[0061] In this embodiment, the welding equipment controller refers to the core component that controls the operation of the welding equipment. It can receive optimization parameters from the cloud platform and adjust the operating status of the welding equipment in real time.

[0062] In this embodiment, the pre-deployed sensors include voltage, current, temperature, speed, air pressure, laser, ultrasonic sensors, etc.

[0063] In this embodiment, the preset target requirements are, for example, quality compliance, efficiency improvement, and cost reduction.

[0064] The beneficial effects of the above design scheme are as follows: By pre-deploying sensors to collect welding process parameters, environmental parameters, and weld characteristic data, and after preprocessing these data to obtain multi-source data, this data is uploaded to the welding cloud platform. This achieves standardization and accuracy of the multi-source data, enabling centralized retrieval, sharing, and subsequent reuse, providing sufficient data support for model training and parameter optimization. By inputting the multi-source data into a pre-trained weld quality prediction model based on the welding cloud platform, weld quality data is obtained, improving the objectivity and accuracy of weld quality assessment. This approach is suitable for large-scale, standardized production scenarios. Furthermore, by combining weld quality data with… Pre-defined target requirements determine the initial optimization parameters for the weld process, ensuring that the optimization direction aligns with actual production needs. This provides a scientific basis for real-time adjustments to the welding equipment. The welding equipment controller performs real-time optimization according to the initial optimization parameters and collects current weld quality data during the optimization process. This data is then fed back to the welding cloud platform to update the weld quality prediction model, enabling rapid response to quality fluctuations during the welding process, preventing defect expansion, and improving weld quality stability. By feeding back the current weld quality data during the optimization process to the cloud platform and updating the model, the model can continuously adapt to changes in the welding scenario, gradually improving the accuracy of prediction and optimization, and achieving iterative upgrades of the technical solution.

[0065] Example 2: Based on Example 1, this embodiment of the invention provides a method for weld quality prediction and process optimization based on a welding cloud platform. In step S1, after preprocessing the welding process parameters, environmental parameters, and weld characteristic data, multi-source data is obtained, including:

[0066] The welding process parameters, environmental parameters, and weld characteristic data are processed by removing abnormal data, supplementing missing values, and deduplicating duplicate data to obtain the processed data.

[0067] Data standardization is performed on the processed data to obtain multi-source data.

[0068] In this embodiment, abnormal data removal is achieved based on the process threshold range and statistical regularity of steel structure welding process parameters, environmental parameters, and weld characteristic data.

[0069] In this embodiment, missing value supplementation is based on the characteristic that the welding process is a sequential continuous process, and is achieved by using linear interpolation or nearest neighbor filling method, which are highly adaptable to industrial scenarios.

[0070] In this embodiment, duplicate data deduplication is achieved based on a dual criterion of timestamp and core features.

[0071] In this embodiment, the data is standardized for the following reasons: the reasonable range of welding current is 200-300A, and a certain collected value is 250A. Then the standardized value is (250-200) / (300-200)=0.5.

[0072] The beneficial effects of the above design scheme are as follows: By removing abnormal data, supplementing missing values, and deduplicating duplicate data in welding process parameters, environmental parameters, and weld characteristic data, processed data is obtained, which solves the inherent defects of the original data at the welding site and ensures the authenticity and integrity of the data. By standardizing the processed data, multi-source data is obtained, eliminating the dimensional differences between multi-source data and ensuring the fairness and effectiveness of model calculation. The pre-processed data can truly and accurately reflect the actual impact of each parameter on the weld quality. The weld quality prediction results obtained by the cloud platform based on this data are more in line with the actual site conditions, and the process optimization parameters determined in conjunction with the prediction results will be more targeted.

[0073] Example 3: Based on Example 1, this embodiment of the invention provides a method for weld quality prediction and process optimization based on a welding cloud platform. In step S1, uploading multi-source data to the welding cloud platform includes:

[0074] By deploying edge computing nodes near the welding station, the multi-source data is encapsulated and formatted into data frames.

[0075] Based on the communication link in the edge computing node, the data frame is sent to the dedicated data receiving port of the welding cloud platform.

[0076] In this embodiment, the data frame format is, for example,: welding station number + sensor type + millisecond-level timestamp + standardized value.

[0077] In this embodiment, the communication link is an industrial-grade communication link, which adopts a dual-link redundancy architecture with wired transmission as the main method and wireless transmission as the auxiliary method, and is adapted to the complex working conditions of steel structure welding sites.

[0078] In this embodiment, encapsulating multi-source data into data frames ensures that a single data frame corresponds to a complete set of multi-source welding process data, thereby achieving data uniqueness and traceability.

[0079] The beneficial effects of the above design scheme are as follows: By deploying edge computing nodes near the welding station, the multi-source data is encapsulated and formatted uniformly, encapsulating the multi-source data into data frames. This encapsulation and format unification can be completed at the edge computing node, eliminating the need to transmit fragmented data to remote devices for processing. This significantly shortens the time required for data preprocessing and encapsulation, adapting to the real-time uploading requirements of multi-source data during the welding process. Furthermore, the uniform format encapsulation into data frames ensures the integrity and traceability of data transmission. Through the communication link in the edge computing node, the data frames are sent to the dedicated data receiving port of the welding cloud platform, reducing the risks of loose link interfaces and interference, ensuring stable transmission of data frames to the dedicated receiving port in the cloud, avoiding data transmission interruptions due to link failures, and guaranteeing continuous uploading of multi-source data. This provides uninterrupted data input for subsequent weld quality prediction models. The welding cloud platform's dedicated data receiving port, which only connects to data frames sent by the edge computing nodes at each station, effectively isolates interference and unauthorized access from external irrelevant data, reducing the risk of industrial data leakage and tampering, and ensuring the security of core welding data.

[0080] Example 4: Based on Example 1, this embodiment of the invention provides a method for weld quality prediction and process optimization based on a welding cloud platform, such as... Figure 2 As shown, in step S2, multi-source data is input into a weld quality prediction model pre-trained based on a welding cloud platform to obtain weld quality data, including:

[0081] The multi-source data are grouped and integrated according to the preset time series rules to obtain an input dataset that matches the input dimension of the weld quality prediction model.

[0082] The input dataset is fed into the weld quality prediction model, which automatically performs feature mining and computational analysis on the input dataset, identifies the correlation between multi-source data and weld quality, and outputs weld quality data.

[0083] In this embodiment, the input dataset is associated with the corresponding welding station number and a millisecond-level timestamp.

[0084] In this embodiment, the weld quality prediction model is deployed on a welding cloud platform server cluster and is trained and generated based on historical welding data and corresponding weld quality inspection results. It has the ability to extract features from multiple sources and predict quality.

[0085] In this embodiment, the preset time rule refers to the data grouping rules pre-set by the welding cloud platform that are adapted to the characteristics of welding operations and the requirements of model input. The core purpose is to classify and integrate the time-series multi-source data uploaded to the structured database according to the welding operation cycle, process stage or time window to form a complete data sequence that conforms to the model input dimension, while preserving the temporal correlation and traceability of the data.

[0086] The beneficial effects of the above design scheme are as follows: by grouping and integrating multi-source data according to preset time sequence rules, an input dataset matching the input dimension of the weld quality prediction model is obtained. Scattered standardized data are associated with complete data sequences corresponding to a single welding operation or a specific process stage, avoiding data mixing at different times. By inputting the input dataset into the weld quality prediction model, feature mining and computational analysis are automatically performed on the input dataset to identify the correlation between multi-source data and weld quality, and weld quality data is output. The grouped and integrated data has been pre-matched with the model input dimension, avoiding secondary grouping or dimension correction after the model receives the data, reducing the ineffective computing power consumption of the welding cloud platform, and improving prediction efficiency and standardization level.

[0087] Example 5: Based on Example 1, this embodiment of the invention provides a method for weld quality prediction and process optimization based on a welding cloud platform. In step S3, based on weld quality data and combined with preset target requirements, initial optimization parameters for weld process parameters are determined, including:

[0088] The preset target requirements are first decomposed to obtain the first target, which includes the quality compliance target, the production efficiency target, and the energy consumption cost target. The first target is then second decomposed to obtain quantifiable sub-targets.

[0089] Based on the scaling method, a first judgment matrix and a second judgment matrix are generated to determine the importance between any two first objectives and the importance between any two quantifiable sub-objectives, and the first maximum eigenvalue and the second maximum eigenvalue of the first judgment matrix and the second judgment matrix are obtained respectively.

[0090] Consistency verification is performed based on the first and second largest eigenvalues. If it passes, the eigenvectors corresponding to the first and second judgment matrices are calculated based on the eigenvalue method. The eigenvectors are normalized and used as the initial sovereign weight of the first target and the initial sub-weights of the quantifiable sub-targets. Based on the initial sovereign weights and the initial sub-weights, the initial weight scheme is obtained. Otherwise, the scale value is adjusted and the latest judgment matrix is ​​regenerated until the consistency verification passes.

[0091] Based on historical weighting schemes and their corresponding historical performance data, the theoretical performance data under the initial weighting scheme is predicted, the deviation between the theoretical performance data and the historical performance data is obtained, and the deviation is fitted and analyzed by linear regression algorithm to obtain the correction coefficient of the corresponding weight. Based on the correction coefficient, the initial main weight and initial sub-weight are corrected to obtain the target weight. Based on the first target, quantifiable sub-targets and target weight, the optimized target system is established.

[0092] Based on the defect types and defect association parameters in the weld quality data, the process parameter-defect association rule library built into the welding cloud platform is retrieved, and the process parameter set corresponding to the defect type and defect association parameter is obtained by reverse mapping. At the same time, parameter constraint boundaries are constructed based on the characteristics of steel structure welding materials and the rated parameters of welding equipment.

[0093] Based on the optimization target system and the parameter constraint boundary, the process parameter set is iteratively calculated to obtain the optimal process parameter scheme. Based on the optimal process parameter scheme, the initial optimization parameters for the weld process parameters are determined.

[0094] In this embodiment, quantifiable sub-objectives include, for example: the quality compliance target is broken down into the defect rate control sub-objective and the weld grade compliance sub-objective; the production efficiency target is broken down into the welding speed improvement sub-objective and the stable operation cycle time sub-objective; and the energy consumption cost target is broken down into the power consumption reduction sub-objective and the consumable loss control sub-objective.

[0095] In this embodiment, the initial scaling method is set to a 1-9 scale, which is combined with the steel structure welding scenario to define the comparison standard. 1 indicates that the two targets are of equal importance, 3 indicates that one target is slightly more important than the other, 5 indicates that one target is significantly more important than the other, 7 indicates that one target is strongly more important than the other, and 9 indicates that one target is extremely more important than the other. 2, 4, 6, and 8 are the median values ​​of the corresponding scales. The reciprocals of 1-9 indicate that the two targets are of opposite importance. For example, 1 / 3 indicates that one target is slightly less important than the other. The welding cloud platform automatically pops up a pairwise comparison interface. The operator inputs the scale value based on the welding scenario requirements, and the system automatically generates an upper triangular or lower triangular judgment matrix.

[0096] In this embodiment, the welding cloud platform automatically retrieves historical optimization data from the past three years that matches the current welding scenario (workpiece type, welding material, equipment model, and environmental conditions), and filters out the weighting scheme and corresponding actual application effect data.

[0097] In this embodiment, the core feature parameters corresponding to the first and second judgment matrices are key indicators for measuring the consistency of the judgment matrices. They are solved through matrix operations. If the judgment matrices are completely consistent, the largest eigenvalue is equal to the matrix order. The greater the deviation, the greater the difference between the largest eigenvalue and the matrix order.

[0098] In this embodiment, the linear regression algorithm is used to fit and analyze historical weight schemes, effect data and current deviation values ​​to generate weight correction coefficients. The core of the algorithm is to build a linear correlation model between deviation values ​​and weights to achieve accurate correction of deviations.

[0099] In this embodiment, the process parameter-defect association rule base is the core database built into the welding cloud platform. It integrates industry standards, historical data, and defect mechanisms, and stores rules according to defect type, core influencing parameter, and parameter adjustment trend. This is used to reverse locate the process parameters that affect the defect. For example, the rule base stores porosity defects with the core parameter being shielding gas flow rate and welding speed with the adjustment trend being increasing shielding gas flow rate (15→22L / min) and decreasing welding speed (5→3.8mm / s). For predicted porosity defects, the corresponding parameters are directly mapped.

[0100] In this embodiment, the parameter constraint boundary is the range of process parameter values ​​constructed based on hardware, process, and compliance requirements, which is used to limit the boundary of iterative calculation.

[0101] The beneficial effects of the above design scheme are as follows: By first decomposing the preset target requirements, the first targets of quality compliance, production efficiency, and energy cost are obtained. The first targets are then further decomposed to obtain quantifiable sub-targets. This transforms the vague preset target requirements into a clear hierarchical target system, taking into account the three core demands of quality, efficiency, and cost. This adapts to the actual production needs of balancing multiple targets in steel structure welding. The quantification of sub-targets provides clear indicators for subsequent weight calculation and effect verification, avoiding deviations in optimization direction caused by target ambiguity. A first judgment matrix and a second judgment matrix are generated based on the importance between any two first targets and between any two quantifiable sub-targets using a scaling method. The first and second maximum eigenvalues ​​of the first and second judgment matrices are obtained respectively. Consistency verification is performed based on the first and second maximum eigenvalues. If successful, the eigenvectors corresponding to the first and second judgment matrices are calculated using the eigenvalue method. After normalization, the eigenvectors are used as the initial sovereign weights of the first targets and the initial sub-weights of the quantifiable sub-targets. Based on the initial sovereign weights and initial sub-weights, an initial weight scheme is obtained. Otherwise... Then, the scale value is adjusted, and the latest judgment matrix is ​​regenerated until the consistency check is passed. The deviation of subjective judgment is forcibly corrected to ensure the rationality of the initial sovereign weight and initial sub-weights, laying a reliable foundation for the subsequent construction of the optimization target system and avoiding process parameter optimization errors caused by weight distortion. Based on the historical weight scheme and its corresponding historical effect data, the theoretical effect data under the initial weight scheme is predicted, and the deviation value between the theoretical effect data and the historical effect data is obtained. The deviation value is fitted and analyzed by the linear regression algorithm to obtain the correction coefficient of the corresponding weight. Based on the correction coefficient, the initial sovereign weight and initial sub-weights are corrected to obtain the target weight. Based on the first target, quantifiable sub-targets and target weights, the optimization target system is established. Based on the defect type and defect association parameters in the weld quality data, the process parameter-defect association rule library built into the welding cloud platform is called to obtain the process parameter set corresponding to the defect type and defect association parameters through reverse mapping. At the same time, based on the characteristics of steel structure welding materials and the rated parameters of welding equipment, parameter constraint boundaries are constructed to directly and accurately optimize the defect type in the weld quality data, avoiding blindly traversing all parameters and greatly improving the efficiency of iterative calculation.Simultaneously, a triple constraint boundary is constructed by combining material properties, equipment rated parameters, and industry standards. This limits the parameter range from three dimensions: hardware, process, and compliance, preventing optimized parameters from exceeding equipment capacity, violating material properties, or adhering to industry standards. This ensures that the initial optimized parameters can be directly applied, reducing trial-and-error costs. Based on the aforementioned optimization target system and the parameter constraint boundary, the process parameter set is iteratively calculated to obtain the optimal process parameter scheme. Based on this optimal scheme, the initial optimized parameters for the weld process are determined. The generated initial optimized parameters not only conform to theoretical optimality but also adapt to the actual conditions of the welding equipment, materials, and processes on site, providing a high-quality benchmark for subsequent real-time optimization of the welding equipment.

[0102] Example 6: Based on Example 5, this embodiment of the invention provides a method for weld quality prediction and process optimization based on a welding cloud platform, which performs consistency verification based on a first maximum eigenvalue and a second maximum eigenvalue, including:

[0103] Obtain the first difference between the first or second largest eigenvalue and the order of the corresponding judgment matrix, and the second difference between the order of the corresponding judgment matrix and 1;

[0104] The consistency index is determined based on the ratio of the first difference and the second difference. If the consistency index is less than the preset value, the consistency verification is determined to be successful; otherwise, the consistency verification is determined to be unsuccessful.

[0105] The beneficial effects of the above design scheme are: it eliminates the need for table lookup and matching processes for random consistency indices, greatly simplifying the algorithm logic; and through prior consistency verification, it lays a reliable foundation for subsequent weight correction, optimization of the target system construction, and iterative calculation of process parameters, thereby improving the adaptability of the initial optimization parameters from the source.

[0106] Example 7: Based on Example 5, this embodiment of the invention provides a method for weld quality prediction and process optimization based on a welding cloud platform. Iterative calculations are performed on a set of process parameters to obtain the optimal process parameter scheme, including:

[0107] The process parameter set is iteratively calculated, and multiple sets of candidate process parameter schemes are generated based on the iteration results.

[0108] Virtual welding simulations were performed on multiple candidate process parameter schemes based on digital twin models, and the optimal process parameter scheme was selected based on the simulation results.

[0109] The beneficial effects of the above design scheme are: multiple candidate schemes are generated through iterative calculation, taking into account both global optimization and parameter constraint boundaries, avoiding getting trapped in local optima; digital twin virtual simulation does not require physical welding tests, can accurately simulate the welding process, predict weld quality and defects, shorten the optimization cycle; and combined with simulation results screening, it can accurately match the optimization target system, ensuring that the optimal scheme not only meets the requirements of quality, efficiency and cost, but also adapts to the characteristics of on-site equipment and materials, and improves the reliability of initial optimization parameters.

[0110] Example 8: Based on Example 1, this embodiment of the invention provides a method for weld quality prediction and process optimization based on a welding cloud platform. In step S4, the current weld quality data is fed back to the welding cloud platform to update the weld quality prediction model, including:

[0111] Feature extraction is performed on the current weld quality data according to quality characteristics, parameter characteristics, scenario characteristics and operation characteristics, and a four-dimensional traceability sample of quality-parameter-scenario-operation is formed based on the extraction results;

[0112] The four-dimensional traceability sample is input into a pre-established welding Bayesian network. The root node layer is used to analyze process parameters and environmental parameters, the intermediate hidden node layer is used to analyze indirect parameters, and the leaf node layer is used to analyze defect type and defect level. Based on the analysis results, the leaf node layer is used as the cause to reason backward to the intermediate hidden node layer and the root node layer to obtain the posterior causal probability that the current parameter abnormality causes the defect to occur. Based on the sum of the posterior causal probabilities that all parameter abnormalities cause the same defect to occur, the causal contribution of the current parameter abnormality to the defect is obtained. Based on the causal contribution, the core parameters for the occurrence of the defect are determined.

[0113] The causal relationship between core parameters and defects is divided into immutable and variable relationships according to whether it is affected by the scenario. A general critical interval is determined for the continuous process parameters in the immutable relationship, and common features are extracted for the discrete process parameters in the variable relationship. Based on the general critical interval and common features, a scenario-free general rule is established.

[0114] The key positions corresponding to the core parameters are obtained from the weld quality prediction model and marked to obtain the local module for this update. The local module for this update is updated for the first time, and the general rule without scenarios is added to the data-scarce model scenario layer for the second update. Based on the results of the first and second updates, the parameters of the weld quality prediction model are optimized to obtain the latest prediction model.

[0115] In this embodiment, quality characteristics include weld defect type (porosity / crack), defect level, defect location, etc.; parameter characteristics include process parameter time sequence curves, environmental parameter fluctuation data, and equipment operating status data (such as welding machine current stability) during the weld welding process; scenario characteristics include welding station, steel grade, welding method (such as submerged arc welding / gas shielded welding), etc.; and operation characteristics include welder operation habit labels (such as welding torch angle changes and walking speed fluctuations).

[0116] In this embodiment, the welding Bayesian network is a causal-specific architecture for the welding process. The network nodes are arranged in layers according to the causal transmission logic, without any messy associations. All directed edges only represent the physical logic of welding cause and effect, rather than meaningless correlation connections. The node hierarchy is fixed at 3 layers.

[0117] In this embodiment, the immutable relationship is an objectively existing and unchanging physical relationship in welding metallurgy, fluid mechanics, and materials science, while the variable relationship is a relationship that is strongly bound to a specific welding scenario and whose values, thresholds, states, conditions, etc. change with the scenario.

[0118] In this embodiment, the causal contribution of the current parameter abnormality to the occurrence of the defect is the ratio of the posterior causal probability of the current parameter abnormality causing the defect to occur to the sum of the posterior causal probabilities of all parameter abnormalities causing the same defect to occur.

[0119] In this embodiment, an indirect parameter is, for example, that the probability of a sudden drop in the protective gas flow rate leading to a deterioration in the molten pool protection effect is 92%.

[0120] The beneficial effects of the above design scheme are as follows: By extracting four-dimensional features of quality, parameters, scenario, and operation and associating them with traceability samples, it provides real and traceable input data for subsequent causal inference, eliminating analytical bias caused by missing data dimensions. By inferring the posterior causal probability of parameter anomalies from defect results, and by differentiating core parameters through causal contribution metric, it breaks through the limitations of conventional correlation analysis, accurately locates the dominant cause of defects, avoids indiscriminate optimization of all parameters, significantly improves the targeting of model updates, and reduces computational waste. By splitting the variable and immutable relationships of causality, it determines the general critical interval of continuous parameters, extracts common features of discrete parameters, and refines general rules that are independent of specific scenarios. This solves the model optimization problem in scenarios with scarce data, such as special steel and niche welding processes, and enables cross-scenario knowledge reuse. It marks the corresponding modules of core parameters for local updates, avoids knowledge forgetting caused by full training, and improves update efficiency. By embedding general rules into the data-scarce scenario layer, it fills data gaps, and the dual updates collaboratively optimize model parameters, which not only strengthens the defect prediction accuracy of core parameters but also enhances the model's generalization ability in niche scenarios, forming a closed-loop optimization.

[0121] Example 9: Based on Example 8, this embodiment of the invention provides a weld quality prediction and process optimization method based on a welding cloud platform. Taking the leaf node layer as the cause, it performs reverse reasoning towards the intermediate hidden node layer and the root node layer to obtain the posterior causal probability of defects caused by current parameter anomalies. Specifically:

[0122] ;

[0123] in, This represents the posterior causal probability that an abnormality in the current parameters leads to the occurrence of a defect. This indicates the probability of a defect occurring when the current parameters are abnormal. This indicates the probability that the current parameter is abnormal. This indicates the probability of a defect occurring.

[0124] The beneficial effect of the above design scheme is that it can deduce the posterior causal probability of parameter anomalies from the defect results, thus providing a basis for the calculation of causal contribution.

[0125] Example 10: Based on Example 8, this embodiment of the invention provides a method for weld quality prediction and process optimization based on a welding cloud platform, which determines a general critical interval for continuous process parameters in an invariant relationship, including:

[0126] The ratio of the specific values ​​of continuous process parameters in all scenarios to the critical reference state values ​​in the scenarios is used as the parameter deviation.

[0127] When the causal contribution of a continuous process parameter is greater than a preset contribution, obtain all values ​​of the parameter deviation, and then obtain the maximum and minimum values ​​from all values ​​to determine the general critical interval.

[0128] In this embodiment, when the deviation of a certain parameter is greater than a preset contribution, it indicates that the parameter is the dominant cause of the defect.

[0129] The beneficial effects of the above design scheme are: by stripping scene attributes through dimensionless deviation, filtering the deviation range with the causal contribution threshold, accurately locking the general critical interval of core parameters, improving the cross-scene adaptability of rules, providing a reliable basis for targeted model updates, and taking into account both accuracy and universality.

[0130] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this application and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for weld quality prediction and process optimization based on a welding cloud platform, characterized in that, include: S1: Based on pre-deployed sensors, welding process parameters, environmental parameters and weld feature data are collected. After data preprocessing of the welding process parameters, environmental parameters and weld feature data, multi-source data is obtained and uploaded to the welding cloud platform. S2: Input the multi-source data into the weld quality prediction model pre-trained based on the welding cloud platform to obtain weld quality data; S3: Based on weld quality data and combined with preset target requirements, determine the initial optimization parameters for weld process parameters; S4: The welding equipment controller performs real-time optimization according to the initial optimization parameters and collects the current weld quality data in real time during the optimization process. The current weld quality data is then fed back to the welding cloud platform to update the weld quality prediction model. In step S1, uploading multi-source data to the welding cloud platform includes: By deploying edge computing nodes near the welding station, the multi-source data is encapsulated and formatted into data frames. Based on the communication link in the edge computing node, the data frame is sent to the dedicated data receiving port of the welding cloud platform; In step S2, multi-source data is input into a weld quality prediction model pre-trained based on a welding cloud platform to obtain weld quality data, including: The multi-source data are grouped and integrated according to the preset time series rules to obtain an input dataset that matches the input dimension of the weld quality prediction model. The input dataset is fed into the weld quality prediction model, which automatically performs feature mining and computational analysis on the input dataset, identifies the correlation between multi-source data and weld quality, and outputs weld quality data. In step S3, based on weld quality data and in conjunction with preset target requirements, initial optimization parameters for the weld process parameters are determined, including: The preset target requirements are first decomposed to obtain the first target, which includes the quality compliance target, the production efficiency target, and the energy consumption cost target. The first target is then second decomposed to obtain quantifiable sub-targets. Based on the scaling method, a first judgment matrix and a second judgment matrix are generated to determine the importance between any two first objectives and the importance between any two quantifiable sub-objectives, and the first maximum eigenvalue and the second maximum eigenvalue of the first judgment matrix and the second judgment matrix are obtained respectively. Consistency verification is performed based on the first and second largest eigenvalues. If it passes, the eigenvectors corresponding to the first and second judgment matrices are calculated based on the eigenvalue method. The eigenvectors are normalized and used as the initial sovereign weight of the first target and the initial sub-weights of the quantifiable sub-targets. Based on the initial sovereign weights and the initial sub-weights, the initial weight scheme is obtained. Otherwise, the scale value is adjusted and the latest judgment matrix is ​​regenerated until the consistency verification passes. Based on historical weighting schemes and their corresponding historical performance data, the theoretical performance data under the initial weighting scheme is predicted, the deviation between the theoretical performance data and the historical performance data is obtained, and the deviation is fitted and analyzed by linear regression algorithm to obtain the correction coefficient of the corresponding weight. Based on the correction coefficient, the initial main weight and initial sub-weight are corrected to obtain the target weight. Based on the first target, quantifiable sub-targets and target weight, the optimized target system is established. Based on the defect types and defect association parameters in the weld quality data, the process parameter-defect association rule library built into the welding cloud platform is retrieved, and the process parameter set corresponding to the defect type and defect association parameter is obtained by reverse mapping. At the same time, parameter constraint boundaries are constructed based on the characteristics of steel structure welding materials and the rated parameters of welding equipment. Based on the optimization target system and the parameter constraint boundary, the process parameter set is iteratively calculated to obtain the optimal process parameter scheme. Based on the optimal process parameter scheme, the initial optimization parameters for the weld process parameters are determined.

2. The method for weld quality prediction and process optimization based on a welding cloud platform according to claim 1, characterized in that, In step S1, after preprocessing the welding process parameters, environmental parameters, and weld characteristic data, multi-source data is obtained, including: The welding process parameters, environmental parameters, and weld characteristic data are processed by removing abnormal data, supplementing missing values, and deduplicating duplicate data to obtain the processed data. Data standardization is performed on the processed data to obtain multi-source data.

3. The method for weld quality prediction and process optimization based on a welding cloud platform according to claim 1, characterized in that, Consistency verification is performed based on the first and second largest eigenvalues, including: Obtain the first difference between the first or second largest eigenvalue and the order of the corresponding judgment matrix, and the second difference between the order of the corresponding judgment matrix and 1; The consistency index is determined based on the ratio of the first difference and the second difference. If the consistency index is less than the preset value, the consistency verification is determined to be successful; otherwise, the consistency verification is determined to be unsuccessful.

4. The method for weld quality prediction and process optimization based on a welding cloud platform according to claim 1, characterized in that, The optimal process parameter scheme is obtained by iterative calculation of the set of process parameters, including: The process parameter set is iteratively calculated, and multiple sets of candidate process parameter schemes are generated based on the iteration results. Virtual welding simulations were performed on multiple candidate process parameter schemes based on digital twin models, and the optimal process parameter scheme was selected based on the simulation results.

5. The method for weld quality prediction and process optimization based on a welding cloud platform according to claim 1, characterized in that, In step S4, the current weld quality data is fed back to the welding cloud platform to update the weld quality prediction model, including: Feature extraction is performed on the current weld quality data according to quality characteristics, parameter characteristics, scenario characteristics and operation characteristics, and a four-dimensional traceability sample of quality-parameter-scenario-operation is formed based on the extraction results; The four-dimensional traceability sample is input into a pre-established welding Bayesian network. The root node layer is used to analyze process parameters and environmental parameters, the intermediate hidden node layer is used to analyze indirect parameters, and the leaf node layer is used to analyze defect type and defect level. Based on the analysis results, the leaf node layer is used as the cause to reason backward to the intermediate hidden node layer and the root node layer to obtain the posterior causal probability that the current parameter abnormality causes the defect to occur. Based on the sum of the posterior causal probabilities that all parameter abnormalities cause the same defect to occur, the causal contribution of the current parameter abnormality to the defect is obtained. Based on the causal contribution, the core parameters for the occurrence of the defect are determined. The causal relationship between core parameters and defects is divided into immutable and variable relationships according to whether it is affected by the scenario. A general critical interval is determined for the continuous process parameters in the immutable relationship, and common features are extracted for the discrete process parameters in the variable relationship. Based on the general critical interval and common features, a scenario-free general rule is established. The key positions corresponding to the core parameters are obtained from the weld quality prediction model and marked to obtain the local module for this update. The local module for this update is updated for the first time, and the general rule without scenarios is added to the data-scarce model scenario layer for the second update. Based on the results of the first and second updates, the parameters of the weld quality prediction model are optimized to obtain the latest prediction model.

6. The method for weld quality prediction and process optimization based on a welding cloud platform according to claim 5, characterized in that, Starting from the leaf node layer as the cause, we reason backwards towards the intermediate hidden node layer and the root node layer to obtain the posterior causal probability that the current parameter anomaly causes the defect to occur, specifically: ； in, This represents the posterior causal probability that an abnormality in the current parameters leads to the occurrence of a defect. This indicates the probability of a defect occurring when the current parameters are abnormal. This indicates the probability that the current parameter is abnormal. This indicates the probability of a defect occurring.

7. The method for weld quality prediction and process optimization based on a welding cloud platform according to claim 5, characterized in that, Determining the general critical interval for continuous process parameters in invariant relationships includes: The ratio of the specific values ​​of continuous process parameters in all scenarios to the critical reference state values ​​in the scenarios is used as the parameter deviation. When the causal contribution of a continuous process parameter is greater than a preset contribution, obtain all values ​​of the parameter deviation, and then obtain the maximum and minimum values ​​from all values ​​to determine the general critical interval.

Images