Multi-parameter cooperative regulation and control optimization control method based on steel wire welding process
By constructing a full lifecycle tracking model for the welding quality of the steel wire welding process, and using causal relationships and deep learning to generate real-time control commands, the problems of lagging welding process control strategies and poor adaptability in existing technologies are solved, thereby improving the stability and safety of the welding process.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JIANGYIN KEYU ELECTRIC APPLIANCES
- Filing Date
- 2026-02-05
- Publication Date
- 2026-05-01
AI Technical Summary
Existing control methods for steel wire welding processes lack the ability to uncover the deep causal relationships among multiple variables, resulting in lagging control strategies. This makes it impossible to achieve quantitative diagnosis and collaborative avoidance of fault types, and the welding process control precision is insufficient with poor adaptability, leading to quality risks.
A full lifecycle tracking model for welding quality in the steel wire welding process is constructed. Through causal relationship analysis and deep learning, real-time control commands are generated. Combined with dynamic prediction and safety boundary verification, intelligent closed-loop control is achieved.
It enables interpretable diagnosis and root cause analysis of welding quality problems, significantly improves the stability and adaptability of the welding process, and ensures the reliability and safety of welding quality.
Smart Images
Figure CN121956902A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of automation control technology, specifically to a multi-parameter collaborative regulation and optimization control method for the steel wire welding process. Background Technology
[0002] Optimization control methods based on multi-parameter collaborative regulation of the steel wire welding process mostly rely on traditional PID control or simple linear models. Their core deficiency lies in the lack of exploration and utilization of the deep causal relationships among multiple variables in the welding process. They can only adjust local parameters through surface correlation, resulting in lagging control strategies and a tendency to fall into the trap of "treating the symptoms but not the root cause." At the same time, existing methods fail to achieve quantitative fault type, forward-looking diagnosis, and collaborative avoidance, making them unable to cope with complex working conditions with multiple fault coupling. Furthermore, the control command generation process lacks strict safety boundary constraints, ultimately leading to insufficient control precision, poor adaptability, and potential quality risks in the welding process. Summary of the Invention
[0003] To address the aforementioned technical problems, this technical solution provides a multi-parameter collaborative regulation and optimization control method for the steel wire welding process, thus resolving the aforementioned issues.
[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A multi-parameter collaborative regulation and optimization control method for the steel wire welding process includes: S1. Based on the historical steel wire welding process control database of known welding methods, analyze the causal relationship between historical steel wire welding process control parameters and welding quality, construct a full life cycle tracking model of steel wire welding process quality, and generate a fault source map of steel wire welding quality for known welding methods. S2. Obtain the real-time wire welding process control parameters of the welding method, and match them with the wire welding quality fault tracing map of the known welding method to determine the type of wire welding quality fault caused by the deviation of the real-time wire welding process control parameters. Construct a trend analysis model of the influence of the coupled steps of the welding quality fault type, and generate the initial compensation parameters of the wire welding process of the coupled steps of the welding quality fault type. S3. Based on the API control interface of the welding equipment, and taking the initial compensation parameters of the welding quality fault type coupling step wire welding process as input, a dynamic prediction model for the welding process is established to predict the future welding quality of the initial compensation parameters of the welding quality fault type coupling step wire welding process, and generate real-time wire welding process control instructions for executing the welding mode.
[0005] Preferably, step S1 specifically includes: Based on the historical steel wire welding process control database with known welding methods, the historical steel wire welding process control parameters and welding process image data with known welding methods are labeled. Data preprocessing is performed according to linear interpolation and wavelet threshold denoising. The associated molten pool morphology features, thermal cycle curve features and current / voltage waveform features of historical steel wire welding process control are extracted to construct a historical steel wire welding process control-quality result time series dataset. Based on a sliding window, with a unit time as the observation window, and using the historical steel wire welding process control-quality result time series dataset as the observation object, the sequence of related variables of historical steel wire welding process execution flow is extracted according to the classification criteria of control variables, process variables and quality variables. Based on the sequence of associated variables of the historical steel wire welding process, all adjacent nodes are connected to construct an undirected graph of associated variables of the historical steel wire welding process. Through hierarchical testing, the conditional independence between each node in the undirected graph is recursively verified, and non-independent nodes are eliminated to obtain an undirected skeleton graph of associated variables of the historical steel wire welding process. Based on the undirected skeleton graph of the associated variables of the historical steel wire welding process execution flow, the Bayesian posterior knowledge of each variable in the sequence of associated variables of the historical steel wire welding process execution flow is verified. As a V-structure rule and propagation rule, the edge directions between each node in the undirected skeleton graph of the associated variables of the historical steel wire welding process execution flow are determined, and a directed acyclic graph of the associated variables of the historical steel wire welding process execution flow is obtained.
[0006] Preferably, step S2 further includes: Based on the directed acyclic graph of the associated variables of the historical steel wire welding process execution flow, the causal triple array of the associated variables of the historical steel wire welding process execution flow is extracted according to the path between each node in the directed acyclic graph. Using causal triple arrays of historical steel wire welding process execution flow correlation variables, an LSTM long short-term memory network is trained to construct a full life cycle tracking model for steel wire welding quality. The control variables and process variables in the historical steel wire welding process execution flow correlation variable sequences in each causal triple array are used as inputs, and the steel wire welding process execution flow quality variables are used as outputs. Identify several known welding quality failure types during the steel wire welding process; Using PAC principal component analysis, we extracted a vector of known welding quality fault types in the steel wire welding process and a causal triple array vector of related variables in the historical steel wire welding process execution flow. Using the Euclidean distance formula, the spatial distance between the known welding quality fault type vectors in the steel wire welding process and the causal triple array vector of the historical steel wire welding process execution flow association variables is calculated, and the welding quality fault type to which the causal triple array vector of the historical steel wire welding process execution flow association variables belongs is determined. Based on the full life cycle tracking model of welding quality in the wire welding process, SHAP value analysis was used to verify the contribution values of control variables and process variables to the quality variables of the wire welding process execution flow in the historical wire welding process execution flow correlation variable sequence in each causal ternary array. Assign the contribution values of the control variables and process variables in the sequence of associated variables of the historical steel wire welding process to the edge weights of the associated variables of the historical steel wire welding process, and obtain the directed weighted acyclic graph of the associated variables of the historical steel wire welding process. Based on the directed weighted acyclic graph of the associated variables in the historical wire welding process, the triplet arrays of each path are assigned to the welding quality fault type, and a source map of wire welding quality faults with known welding methods is established.
[0007] Preferably, step S2 specifically includes: Data preprocessing is performed based on real-time wire welding process control parameters of the welding method. According to the sliding window, the real-time wire welding process control parameters for the welding method are sliced, and the real-time wire welding process control feature vector in each data slice is extracted. Based on the source map of steel wire welding quality faults with known welding methods, Gaussian analysis is used to calculate the conditional probability distribution of steel wire welding quality fault types under different path-related nodes, thus obtaining the learning network for each path relative to the steel wire welding quality fault type in the source map of steel wire welding quality faults with known welding methods. Based on the learning network for the relative fault types of wire welding quality in the fault tracing map of known welding methods, the network uses the real-time wire welding process control feature vector in each data slice as input to calculate the posterior probability distribution of each path node in the fault tracing map of wire welding quality under a given real-time wire welding process control feature vector, and determines the fault type of wire welding quality indicative of the deviation of real-time wire welding process control parameters.
[0008] Preferably, step S2 further includes: Based on the source map of steel wire welding quality faults with known welding methods, the influence intensity between steel wire welding quality faults with known welding methods is verified by multiple linear regression, and a coupling influence correlation matrix of steel wire welding quality faults is constructed. Based on the correlation matrix of the coupling effects of steel wire welding quality faults, the analytic hierarchy process is used to assign the contribution degree of each type of steel wire welding quality fault to the coupling steel wire welding steps, and to assign weights to each real-time steel wire welding process control parameter. The posterior probability distribution of each path node in the fault tracing map of steel wire welding quality is normalized to obtain the state vector of each path node in the fault tracing map of steel wire welding quality. Based on Kalman filtering, a trend analysis model for the coupling steps of welding quality fault types is constructed. Using the state vectors, state transition matrices, and standardized parameters of the wire welding process control of each path node in the source map of wire welding quality faults, a state transition complement equation is constructed. Taking the deviation of the real-time wire welding process control parameters pointing to the wire welding quality fault type as input, the state transition complement equation is solved recursively to generate the initial compensation parameters of the wire welding process for the coupling steps of welding quality fault types.
[0009] Preferably, step S3 specifically includes: Based on the NARX neural network, a dynamic prediction model for the welding process is constructed. The real-time wire welding process control command of the welding mode is used as input, and the future welding quality of the initial compensation parameters of the wire welding process coupled with the welding quality fault type is used as output to determine the compensation control sequence of the wire welding process coupled with the welding quality fault type.
[0010] Preferably, step S3 further includes: Based on the source map of steel wire welding quality faults, SVR supports regression vector machine to determine the triggerable range hyperplane boundary of the steel wire welding process control parameters for each type of steel wire welding quality fault. This boundary serves as the constraint condition for the steel wire welding process control parameters of the steel wire welding quality fault type. The compensation control sequence of the steel wire welding process for the coupled steps of the welding quality fault type is then filtered to obtain the real-time steel wire welding process control command for the execution welding mode.
[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes a multi-parameter collaborative regulation and optimization control scheme for the steel wire welding process. By constructing a fault source map that integrates causal discovery and dynamic tracking, interpretable diagnosis and root cause analysis of welding quality problems are achieved. On this basis, the system uses Bayesian inference to quantify the probability of fault risk in real time, predicts the development trend of fault coupling through Kalman filtering, and generates forward-looking initial compensation parameters. Finally, by combining nonlinear model predictive control and support vector regression safety boundary verification, the system outputs collaboratively optimized and absolutely safe real-time control commands, forming a complete intelligent closed loop from causal analysis and dynamic diagnosis to safety control, which significantly improves the stability of the steel wire welding process. Attached Figure Description
[0012] Figure 1 The flowchart shows the optimization control method based on multi-parameter collaborative regulation of the steel wire welding process. Detailed Implementation
[0013] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.
[0014] Reference Figure 1 As shown, the multi-parameter collaborative regulation and optimization control method for the steel wire welding process includes: S1. Based on the historical steel wire welding process control database of known welding methods, analyze the causal relationship between historical steel wire welding process control parameters and welding quality, construct a full life cycle tracking model of steel wire welding process, and generate a fault source map of steel wire welding quality for known welding methods. Step S1 specifically includes: Based on the historical steel wire welding process control database with known welding methods, the historical steel wire welding process control parameters and welding process image data with known welding methods are labeled. Data preprocessing is performed according to linear interpolation and wavelet threshold denoising. The associated molten pool morphology features, thermal cycle curve features and current / voltage waveform features of historical steel wire welding process control are extracted to construct a historical steel wire welding process control-quality result time series dataset. Based on a sliding window, with a unit time as the observation window, and using the historical steel wire welding process control-quality result time series dataset as the observation object, the sequence of related variables of historical steel wire welding process execution flow is extracted according to the classification criteria of control variables, process variables and quality variables. Based on the sequence of associated variables of the historical steel wire welding process, all adjacent nodes are connected to construct an undirected graph of associated variables of the historical steel wire welding process. Through hierarchical testing, the conditional independence between each node in the undirected graph is recursively verified, and non-independent nodes are eliminated to obtain an undirected skeleton graph of associated variables of the historical steel wire welding process. Based on the undirected skeleton graph of the associated variables of the historical steel wire welding process execution flow, the Bayesian posterior knowledge of each variable in the sequence of associated variables of the historical steel wire welding process execution flow is verified. As a V-structure rule and propagation rule, the edge directions between each node in the undirected skeleton graph of the associated variables of the historical steel wire welding process execution flow are determined, and a directed acyclic graph of the associated variables of the historical steel wire welding process execution flow is obtained.
[0015] Step S2 also includes: Based on the directed acyclic graph of the associated variables of the historical steel wire welding process execution flow, the causal triple array of the associated variables of the historical steel wire welding process execution flow is extracted according to the path between each node in the directed acyclic graph. Using causal triple arrays of historical steel wire welding process execution flow correlation variables, an LSTM long short-term memory network is trained to construct a full life cycle tracking model for steel wire welding quality. The control variables and process variables in the historical steel wire welding process execution flow correlation variable sequences in each causal triple array are used as inputs, and the steel wire welding process execution flow quality variables are used as outputs. Identify several known welding quality failure types during the steel wire welding process; Using PAC principal component analysis, we extracted a vector of known welding quality fault types in the steel wire welding process and a causal triple array vector of related variables in the historical steel wire welding process execution flow. Using the Euclidean distance formula, the spatial distance between the known welding quality fault type vectors in the steel wire welding process and the causal triple array vector of the historical steel wire welding process execution flow association variables is calculated, and the welding quality fault type to which the causal triple array vector of the historical steel wire welding process execution flow association variables belongs is determined. Based on the full life cycle tracking model of welding quality in the wire welding process, SHAP value analysis was used to verify the contribution values of control variables and process variables to the quality variables of the wire welding process execution flow in the historical wire welding process execution flow correlation variable sequence in each causal ternary array. Assign the contribution values of the control variables and process variables in the sequence of associated variables of the historical steel wire welding process to the edge weights of the associated variables of the historical steel wire welding process, and obtain the directed weighted acyclic graph of the associated variables of the historical steel wire welding process. Based on the directed weighted acyclic graph of the associated variables in the historical wire welding process, the triplet arrays of each path are assigned to the welding quality fault type, and a source map of wire welding quality faults with known welding methods is established.
[0016] When using it, please refer to the steps outlined above: As a further development, a comprehensive method for extracting causal knowledge from historical data on steel wire welding was constructed through the deep integration of data-driven approaches and causal inference. Its core lies in using the PC algorithm to test conditional independence, constructing a directed acyclic graph (DAG) that reveals the true driving relationships between parameters, and then leveraging the dynamic characteristics of the LSTM network learning process to form a full lifecycle tracking model for welding quality. Finally, the SHAP value is used to quantify the strength of causal influence, generating a fault source map that combines qualitative and quantitative information. This achieves a leap from correlation analysis to causal discovery, establishing an interpretable knowledge graph and providing a reliable causal logic basis for subsequent real-time diagnosis and intelligent control.
[0017] Furthermore, as an embodiment of step S1: Scenario: A steel wire skeleton layer welding production line for automobiles, using MIG welding. The historical database contains tens of thousands of welding records.
[0018] Implementation process: Data preprocessing and feature extraction: Control parameters: Welding current (I) = 150-250A, voltage (U) = 22-30V, welding speed (S) = 0.8-1.2m / min.
[0019] Process feature extraction: Molten pool image: The molten pool length (L_pool) is 8.5 mm and the molten pool width (W_pool) is 5.2 mm, extracted in real time from the high-speed camera image.
[0020] Thermal cycling curve: The peak temperature (T_max) was calculated to be 1650℃ from the thermocouple data, and the residence time (t_800) above 800℃ was 12 seconds.
[0021] Electrical signal: The average current (I_avg) is calculated to be 200A and the standard deviation of current (σ_I) is 15 (characterizing arc stability) from the waveform of the current sensor.
[0022] Quality results: According to UT testing, the porosity of this weld is 0.1% (qualified) and the tensile strength (TS) is 550MPa (qualified).
[0023] Results: A time series data was generated: [I=200A, U=26V, S=1.0m / min, L_pool=8.5, W_pool=5.2, T_max=1650, t_800=12, I_avg=200, σ_I=15, Porosity=0.1%, TS=550MPa].
[0024] Causal discovery and DAG construction: Conditional independence test: The algorithm found that when I_avg and U are fixed, the correlation between S (welding speed) and W_pool (melt pool width) becomes insignificant. Therefore, the direct edge between S and W_pool is removed.
[0025] V-Structure Orientation: The algorithm finds that I_avg and U are not connected, but they both affect T_max, and I_avg and U are conditionally independent given T_max. Therefore, the structure is determined to be I_avg->T_max<-U.
[0026] Result: A DAG is finally obtained, with the following partial structure: [I_avg, U]->T_max->W_pool; S->t_800; σ_I->Porosity.
[0027] Full lifecycle model training and graph generation: Causal triples: Extracting triples such as ([I_avg, U], T_max), (T_max, W_pool), ([W_pool, σ_I], Porosity) from a DAG.
[0028] LSTM Training: Taking ([I_avg, U], T_max) as an example, the sequence [I_avg, U] of a past time window is input into the LSTM to train it to predict T_max at the next time step. All triples are trained in this way to form a full lifecycle model.
[0029] Fault Classification and SHAP Quantification: Determine the fault: Define "high porosity" as fault F1.
[0030] PCA and Euclidean distance: Reduce the dimensionality of all historical data vectors that resulted in F1 scores and calculate their centers. New data vectors are classified as F1 type if the distance to these centers is less than a threshold.
[0031] SHAP analysis: For a sample with high porosity, SHAP analysis showed that the contribution value (SHAP value) of σ_I on the path leading to porosity σ_I->Porosity was 0.6; and the contribution value of W_pool on the path W_pool->Porosity was 0.25.
[0032] Generate a weighted graph: In the fault source graph, the edge from node σ_I to Porosity is assigned a weight of 0.6; the edge from node W_pool to Porosity is assigned a weight of 0.25.
[0033] S2. Obtain the real-time wire welding process control parameters of the welding method, and match them with the wire welding quality fault tracing map of the known welding method to determine the type of wire welding quality fault caused by the deviation of the real-time wire welding process control parameters. Construct a trend analysis model of the influence of the coupled steps of the welding quality fault type, and generate the initial compensation parameters of the wire welding process of the coupled steps of the welding quality fault type. Step S2 specifically includes: Data preprocessing is performed based on real-time wire welding process control parameters of the welding method. According to the sliding window, the real-time wire welding process control parameters for the welding method are sliced, and the real-time wire welding process control feature vector in each data slice is extracted. Based on the source map of steel wire welding quality faults with known welding methods, Gaussian analysis is used to calculate the conditional probability distribution of steel wire welding quality fault types under different path-related nodes, thus obtaining the learning network for each path relative to the steel wire welding quality fault type in the source map of steel wire welding quality faults with known welding methods. Based on the learning network for the relative fault types of wire welding quality in the fault tracing map of known welding methods, the network uses the real-time wire welding process control feature vector in each data slice as input to calculate the posterior probability distribution of each path node in the fault tracing map of wire welding quality under a given real-time wire welding process control feature vector, and determines the fault type of wire welding quality indicative of the deviation of real-time wire welding process control parameters.
[0034] Step S2 also includes: Based on the source map of steel wire welding quality faults with known welding methods, the influence intensity between steel wire welding quality faults with known welding methods is verified by multiple linear regression, and a coupling influence correlation matrix of steel wire welding quality faults is constructed. Based on the correlation matrix of the coupling effects of steel wire welding quality faults, the analytic hierarchy process is used to assign the contribution degree of each type of steel wire welding quality fault to the coupling steel wire welding steps, and to assign weights to each real-time steel wire welding process control parameter. The posterior probability distribution of each path node in the fault tracing map of steel wire welding quality is normalized to obtain the state vector of each path node in the fault tracing map of steel wire welding quality. Based on Kalman filtering, a trend analysis model for the coupling steps of welding quality fault types is constructed. Using the state vectors, state transition matrices, and standardized parameters of the wire welding process control of each path node in the source map of wire welding quality faults, a state transition complement equation is constructed. Taking the deviation of the real-time wire welding process control parameters pointing to the wire welding quality fault type as input, the state transition complement equation is solved recursively to generate the initial compensation parameters of the wire welding process for the coupling steps of welding quality fault types.
[0035] When using it, please refer to the steps outlined above: As a further step, Bayesian network inference is used to achieve accurate quantitative diagnosis of real-time faults. Gaussian analysis is used to calculate the conditional probability of faults, mapping real-time process parameters to the posterior probability distribution of specific fault types. Furthermore, a fault coupling influence correlation matrix is constructed based on multiple linear regression and analytic hierarchy process (AHP) to quantify the interaction strength and contribution weight between different faults. Finally, a dynamic trend prediction model is built using Kalman filtering, and the collaborative compensation parameters are obtained by recursively solving the state-space equations. This achieves closed-loop optimization from fault diagnosis to compensation decision-making, providing a basis for the control of the welding process.
[0036] Furthermore, as an embodiment of step S2: Scenario: Continuing with the example in S1, the MIG welding production line for automotive steel wire skeleton layers is running in real time.
[0037] Implementation process: Real-time feature extraction and fault probability calculation: Real-time data: The system reads the real-time feature vector of the current window: [I_avg=245A, U=29V, W_pool=6.5mm, σ_I=8].
[0038] Bayesian network inference: Input the vector into the constructed Bayesian network (fault source map).
[0039] Calculation results: Network output posterior probability: P(burn-through | real-time evidence) = 0.90 (extremely high risk); P(porosity | real-time evidence) = 0.10 (low risk); P(Unfused | Real-time Evidence) = 0.05 (Very Low Risk); Diagnostic conclusion: The deviation in real-time control parameters clearly points to a "burn-through" fault type. The root cause is excessive heat input due to excessive current and voltage.
[0040] Fault Coupling and Weight Analysis: Coupling matrix: Historical data shows that severe burn-through is sometimes accompanied by weld depression, but it is not strongly coupled with other faults in this case.
[0041] AHP weights: According to expert evaluation, "burn-through" is a serious defect with a global weight w_burn-through = 0.7; "porosity" weight w_porosity = 0.2; and "lack of fusion" weight w_lack of fusion = 0.1.
[0042] Kalman filter prediction and compensation generation: State vector: Normalize the failure probability and assign weights to construct the state vector x(k)=[0.9*0.7, 0.1*0.2, 0.05*0.1]^T=[0.63, 0.02, 0.005]^T.
[0043] State prediction: According to the Kalman filter model (whose A matrix already contains the coupling relationship between faults), without intervention, the burn-through risk component in the next state x(k+1|k) will continue to increase to 0.68.
[0044] Solving for compensation: The optimization algorithm begins to solve for how much the current I and voltage U need to be reduced to significantly reduce the burn-through risk in the predicted x(k+1|k)? At the same time, the compensation amounts ΔI and ΔU must meet the equipment constraints (e.g., ΔI>-50A) and must not cause the risk of "non-fusion" to increase sharply due to low heat input (controlled by the B*u(k) term and constraints).
[0045] Generate initial compensation parameters: Through recursive calculation, the system outputs the optimal initial compensation parameters. ΔI*=-25A (reducing the current from 245A to 220A); ΔU*=-2V (the voltage is reduced from 29V to 27V).
[0046] S3. Based on the API control interface of the welding equipment, the initial compensation parameters of the welding process of the coupled steps of welding quality fault type are used as input to establish a dynamic prediction model of the welding process, predict the future welding quality of the initial compensation parameters of the coupled steps of welding quality fault type, and generate real-time control instructions for the welding process of the wire. Step S3 specifically includes: Based on the NARX neural network, a dynamic prediction model for the welding process is constructed. The real-time wire welding process control command of the welding mode is used as input, and the future welding quality of the initial compensation parameters of the wire welding process coupled with the welding quality fault type is used as output to determine the compensation control sequence of the wire welding process coupled with the welding quality fault type.
[0047] Step S3 also includes: Based on the source map of steel wire welding quality faults, SVR supports regression vector machine to determine the triggerable range hyperplane boundary of the steel wire welding process control parameters for each type of steel wire welding quality fault. This boundary serves as the constraint condition for the steel wire welding process control parameters of the steel wire welding quality fault type. The compensation control sequence of the steel wire welding process for the coupled steps of the welding quality fault type is then filtered to obtain the real-time steel wire welding process control command for the execution welding mode.
[0048] When using it, please refer to the steps outlined above: As a further development, a dynamic prediction model for the welding process is constructed using a NARX neural network. With model predictive control as the core algorithm, multi-step rolling optimization is performed based on the initial compensation parameters for welding quality faults to generate a collaborative compensation control sequence. Furthermore, a support vector regression mechanism is used to construct safety boundaries for control parameters of each fault type as hard constraints to perform safety screening and correction on the predictive control sequence. Finally, real-time control commands that satisfy the dynamic optimization objectives and absolutely guarantee process safety are output through the equipment API interface, realizing a fully closed-loop adaptive optimization control of the steel wire welding process from intelligent diagnosis to safe execution.
[0049] Furthermore, as an embodiment of step S3: Scenario: Continuing from the example in S2, the system has diagnosed a high risk of "burn-through" and generated initial compensation parameters: ΔI = -25A, ΔU = -2V.
[0050] Implementation process: NARX Model Prediction and MPC Optimization: Current status: Current I(k) = 245A, voltage U(k) = 29V.
[0051] Optimization objective: The cost function J of MPC is set to drive the system to reach the target area suggested by S2 (current ~220A, voltage ~27V) quickly and smoothly, while minimizing control jitter.
[0052] Dynamic prediction: The NARX model begins simulation. It may predict that a sudden drop of 25A current, while rapidly reducing heat input, could lead to instability in the molten pool or even arc interruption. Conversely, a two-step approach, reducing the current by 15A first and then 10A, would result in a smoother process.
[0053] Generating the compensation sequence: The MPC solver calculates and outputs an optimal compensation control sequence. u*(k) = [I(k) + (-15A), U(k) + (-1.5V)] = [230A, 27.5V]; u*(k+1) = [230A + (-10A), 27.5V + (-0.5V)] = [220A, 27V]; SVR security boundary verification and instruction filtering: Boundary Model Invocation: The system inputs the first step instruction [230A, 27.5V] of the MPC plan execution into the SVR boundary model for all fault types.
[0054] The “burn-through” boundary model returned the following result: [230A, 27.5V] Inside the safe conduit, passed.
[0055] The “unfused” boundary model returned the result: [230A, 27.5V] is also located within the safety conduit, passing through.
[0056] The "porosity" boundary model returned the following result: [230A, 27.5V] Also safe, passed.
[0057] Special case: Assuming the instruction is [210A, 26V], the SVR's "unfused" model may determine that this point is close to its fault boundary (too low heat input). The system will fine-tune this instruction, for example, by correcting it to [215A, 26.2V] to ensure safety.
[0058] Final instruction generation: Since the instruction [230A, 27.5V] generated by MPC passed all safety checks, the system sends this real-time wire welding process control instruction to the welding power source for execution via the welding equipment API.
[0059] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.
Claims
1. A multi-parameter collaborative regulation and optimization control method for steel wire welding process, characterized in that, include: S1. Based on the historical steel wire welding process control database of known welding methods, analyze the causal relationship between historical steel wire welding process control parameters and welding quality, construct a full life cycle tracking model of steel wire welding process quality, and generate a fault source map of steel wire welding quality for known welding methods. S2. Obtain the real-time wire welding process control parameters of the welding method, and match them with the wire welding quality fault tracing map of the known welding method to determine the type of wire welding quality fault caused by the deviation of the real-time wire welding process control parameters. Construct a trend analysis model of the influence of the coupled steps of the welding quality fault type, and generate the initial compensation parameters of the wire welding process of the coupled steps of the welding quality fault type. S3. Based on the API control interface of the welding equipment, and taking the initial compensation parameters of the welding quality fault type coupling step wire welding process as input, a dynamic prediction model for the welding process is established to predict the future welding quality of the initial compensation parameters of the welding quality fault type coupling step wire welding process, and generate real-time wire welding process control instructions for executing the welding mode.
2. The multi-parameter collaborative regulation and optimization control method for steel wire welding process according to claim 1, characterized in that, Step S1 specifically includes: Based on the historical steel wire welding process control database with known welding methods, the historical steel wire welding process control parameters and welding process image data with known welding methods are labeled. Data preprocessing is performed according to linear interpolation and wavelet threshold denoising. The associated molten pool morphology features, thermal cycle curve features and current / voltage waveform features of historical steel wire welding process control are extracted to construct a historical steel wire welding process control-quality result time series dataset. Based on a sliding window, with a unit time as the observation window, and using the historical steel wire welding process control-quality result time series dataset as the observation object, the sequence of related variables of historical steel wire welding process execution flow is extracted according to the classification criteria of control variables, process variables and quality variables. Based on the sequence of associated variables of the historical steel wire welding process, all adjacent nodes are connected to construct an undirected graph of associated variables of the historical steel wire welding process. Through hierarchical testing, the conditional independence between each node in the undirected graph is recursively verified, and non-independent nodes are eliminated to obtain an undirected skeleton graph of associated variables of the historical steel wire welding process. Based on the undirected skeleton graph of the associated variables of the historical steel wire welding process execution flow, the Bayesian posterior knowledge of each variable in the sequence of associated variables of the historical steel wire welding process execution flow is verified. As a V-structure rule and propagation rule, the edge directions between each node in the undirected skeleton graph of the associated variables of the historical steel wire welding process execution flow are determined, and a directed acyclic graph of the associated variables of the historical steel wire welding process execution flow is obtained.
3. The multi-parameter collaborative regulation and optimization control method for steel wire welding process according to claim 2, characterized in that, Step S2 also includes: Based on the directed acyclic graph of the associated variables of the historical steel wire welding process execution flow, the causal triple array of the associated variables of the historical steel wire welding process execution flow is extracted according to the path between each node in the directed acyclic graph. Using causal triple arrays of historical steel wire welding process execution flow correlation variables, an LSTM long short-term memory network is trained to construct a full life cycle tracking model for steel wire welding quality. The control variables and process variables in the historical steel wire welding process execution flow correlation variable sequences in each causal triple array are used as inputs, and the steel wire welding process execution flow quality variables are used as outputs. Identify several known welding quality failure types during the steel wire welding process; Using PAC principal component analysis, we extracted a vector of known welding quality fault types in the steel wire welding process and a causal triple array vector of related variables in the historical steel wire welding process execution flow. Using the Euclidean distance formula, the spatial distance between the known welding quality fault type vectors in the steel wire welding process and the causal triple array vector of the historical steel wire welding process execution flow association variables is calculated, and the welding quality fault type to which the causal triple array vector of the historical steel wire welding process execution flow association variables belongs is determined. Based on the full life cycle tracking model of welding quality in the wire welding process, SHAP value analysis was used to verify the contribution values of control variables and process variables to the quality variables of the wire welding process execution flow in the historical wire welding process execution flow correlation variable sequence in each causal ternary array. Assign the contribution values of the control variables and process variables in the sequence of associated variables of the historical steel wire welding process to the edge weights of the associated variables of the historical steel wire welding process, and obtain the directed weighted acyclic graph of the associated variables of the historical steel wire welding process. Based on the directed weighted acyclic graph of the associated variables in the historical wire welding process, the triplet arrays of each path are assigned to the welding quality fault type, and a source map of wire welding quality faults with known welding methods is established.
4. The multi-parameter collaborative regulation and optimization control method for steel wire welding process according to claim 3, characterized in that, Step S2 specifically includes: Data preprocessing is performed based on real-time wire welding process control parameters of the welding method. According to the sliding window, the real-time wire welding process control parameters for the welding method are sliced, and the real-time wire welding process control feature vector in each data slice is extracted. Based on the source map of steel wire welding quality faults with known welding methods, Gaussian analysis is used to calculate the conditional probability distribution of steel wire welding quality fault types under different path-related nodes, thus obtaining the learning network for each path relative to the steel wire welding quality fault type in the source map of steel wire welding quality faults with known welding methods. Based on the learning network for the relative fault types of wire welding quality in the fault tracing map of known welding methods, the network uses the real-time wire welding process control feature vector in each data slice as input to calculate the posterior probability distribution of each path node in the fault tracing map of wire welding quality under a given real-time wire welding process control feature vector, and determines the fault type of wire welding quality indicative of the deviation of real-time wire welding process control parameters.
5. The multi-parameter collaborative regulation and optimization control method for steel wire welding process according to claim 4, characterized in that, Step S2 also includes: Based on the source map of steel wire welding quality faults with known welding methods, the influence intensity between steel wire welding quality faults with known welding methods is verified by multiple linear regression, and a coupling influence correlation matrix of steel wire welding quality faults is constructed. Based on the correlation matrix of the coupling effects of steel wire welding quality faults, the analytic hierarchy process is used to assign the contribution degree of each type of steel wire welding quality fault to the coupling steel wire welding steps, and to assign weights to each real-time steel wire welding process control parameter. The posterior probability distribution of each path node in the fault tracing map of steel wire welding quality is normalized to obtain the state vector of each path node in the fault tracing map of steel wire welding quality. Based on Kalman filtering, a trend analysis model for the coupling steps of welding quality fault types is constructed. Using the state vectors, state transition matrices, and standardized parameters of the wire welding process control of each path node in the source map of wire welding quality faults, a state transition complement equation is constructed. Taking the deviation of the real-time wire welding process control parameters pointing to the wire welding quality fault type as input, the state transition complement equation is solved recursively to generate the initial compensation parameters of the wire welding process for the coupling steps of welding quality fault types.
6. The multi-parameter collaborative regulation and optimization control method for steel wire welding process according to claim 5, characterized in that, Step S3 specifically includes: Based on the NARX neural network, a dynamic prediction model for the welding process is constructed. The real-time wire welding process control command of the welding mode is used as input, and the future welding quality of the initial compensation parameters of the wire welding process coupled with the welding quality fault type is used as output to determine the compensation control sequence of the wire welding process coupled with the welding quality fault type.
7. The multi-parameter collaborative regulation and optimization control method for steel wire welding process according to claim 6, characterized in that, Step S3 also includes: Based on the source map of steel wire welding quality faults, SVR supports regression vector machine to determine the triggerable range hyperplane boundary of the steel wire welding process control parameters for each type of steel wire welding quality fault. This boundary serves as the constraint condition for the steel wire welding process control parameters of the steel wire welding quality fault type. The compensation control sequence of the steel wire welding process for the coupled steps of the welding quality fault type is then filtered to obtain the real-time steel wire welding process control command for the execution welding mode.