Welding parameter intelligent acquisition and process optimization management system

Abstract

The invention relates to the technical field of welding process optimization, and discloses a welding parameter intelligent acquisition and process optimization management system. The system comprises a multi-source data acquisition module for synchronously acquiring various welding data to generate a multi-dimensional data set; the signal preprocessing module is used for carrying out noise reduction processing on the data; the dynamic parameter regulation and control module dynamically adjusts welding parameters based on a reinforcement learning model; the heterogeneous network communication module is used for designing a dual-channel protocol to guarantee data transmission; and the process optimization decision module is used for generating an optimal welding parameter scheme by using a deep neural network. And a welding quality simulation module and an abnormal process mode library are also arranged. Intelligent collection, real-time regulation and control and process optimization of welding parameters are achieved, the welding quality and the production efficiency are improved, the cost is reduced, and it is guaranteed that welding production is conducted stably.

Description

Technical Field

[0001] The present invention relates to the technical field of welding process optimization, and particularly to an intelligent acquisition and process optimization management system for welding parameters. Background Art

[0002] Welding plays a crucial role in modern industrial production. From large-scale infrastructure construction to the manufacturing of precision electronic equipment, welding processes are everywhere. However, with the continuous development of industrial technology, the disadvantages of traditional welding technology in parameter acquisition and process control have become increasingly prominent, making it difficult to meet the current high requirements for welding quality, production efficiency, and intelligent level.

[0003] In the welding parameter acquisition link, there are many deficiencies in the traditional method. On the one hand, the types of data collected are limited. Usually, only basic parameters such as welding current and voltage can be obtained, and it is difficult to effectively collect key information such as temperature changes and workpiece deformation during the welding process. Take the welding of an automobile engine cylinder block as an example. Due to the lack of accurate monitoring of the welding temperature field, local overheating or overcooling areas cannot be detected in time, which easily leads to uneven weld microstructure and affects the strength and sealing performance of the cylinder block. On the other hand, the acquisition accuracy and real-time performance are poor. The data collected manually or by simple sensors often have errors and cannot reflect the dynamic changes of parameters during the welding process in real time, making it difficult to meet the requirements for real-time monitoring of welding quality.

[0004] The setting of welding process parameters has long relied on the experience of operators, lacking scientific basis and precise control means. Due to the differences in the skill levels and experience of different operators, when facing the same welding task, the parameter settings are different, resulting in uneven welding quality. In the aerospace field, the requirements for welding quality are extremely high, and even a tiny process deviation may cause serious safety hazards. For example, when welding aircraft parts, if the welding energy input is improper, it may lead to weld cracks or lack of fusion defects, threatening flight safety.

[0005] At the same time, during the welding process, due to the influence of external environmental factors (such as power grid voltage fluctuations, electromagnetic interference in the workshop, etc.), the welding parameters are prone to fluctuate, and the traditional system lacks a real-time monitoring and dynamic adjustment mechanism. This makes it impossible to optimize the parameters of the welding process in a timely manner according to the actual situation. Once parameter anomalies occur, it will lead to the generation of welding defects, increasing production costs and product scrap rates.

[0006] In addition, the optimization of welding processes has always been a difficult problem in the industry. Traditional process optimization methods mainly rely on a large number of experiments and trial-and-error, which not only consume a large amount of human, material, and time resources, but also, due to the lack of effective management and in-depth mining of historical process data, it is difficult to find the key factors and parameter combinations for truly optimizing the process. In today's era of accelerating product updates, such an inefficient process optimization method severely restricts the market competitiveness of enterprises.

[0007] With the booming development of emerging technologies such as the Internet of Things, big data, and artificial intelligence, the industrial field is rapidly moving towards the direction of intelligence. However, in the welding industry, the application of these advanced technologies is still in its infancy, and a mature and perfect intelligent acquisition and process optimization management system for welding parameters has not yet been formed. Therefore, developing an efficient and intelligent welding parameter acquisition and process optimization management system has important practical significance for promoting the technological upgrading and sustainable development of the welding industry. Summary of the Invention

[0008] The purpose of the present invention is to provide an intelligent acquisition and process optimization management system for welding parameters to solve the problems raised in the above-mentioned background technology.

[0009] To achieve the above purpose, the present invention provides the following technical solution: An intelligent acquisition and process optimization management system for welding parameters, the system includes: A multi-source data acquisition module for synchronously acquiring welding current, voltage, temperature, and workpiece deformation data through a multi-modal sensor array to generate a multi-dimensional spatio-temporal process data set; A signal preprocessing module for performing non-linear noise reduction processing on the multi-dimensional spatio-temporal process data set to eliminate high-frequency interference components and generate a time-domain stationary signal sequence; A dynamic parameter regulation module for constructing an optimization model based on reinforcement learning, dynamically adjusting the pulse frequency and energy input parameters according to real-time welding quality indicators, and generating an adaptive process regulation instruction; A heterogeneous network communication module for designing a dual-channel transmission protocol and alternately uploading data packets through a wireless local area network and a cellular network to ensure the redundancy and fault tolerance of the communication link; A process optimization decision module for establishing a process feature mapping relationship based on a deep neural network, performing multi-objective optimization modeling on the historical process data set, and generating an optimal welding parameter configuration plan.

[0010] Preferably, the non-linear noise reduction processing uses a combined algorithm of wavelet packet threshold denoising and Kalman filtering to separate transient noise and steady-state process signals, including: Performing wavelet packet multi-scale decomposition on the welding current signal to extract the energy ratio distribution of each sub-band; Construct a Kalman filter state equation, use the high-frequency noise component as the observation variable, and iteratively update the signal estimate value; Identify abnormal current fluctuation segments through the analysis of variance method, and segment the normal welding process signal interval.

[0011] Preferably, the construction of the optimization model includes: Define the state space as a combined vector of current stability, molten pool morphology score, and energy utilization rate; Construct a deep deterministic policy gradient algorithm framework, and use an experience replay mechanism to optimize the convergence speed of the policy network; Design the reward function as the weighted difference between the weld formation quality score and the energy consumption penalty term to drive the model to generate a dynamic regulation strategy.

[0012] Preferably, the design of the dual-channel transmission protocol includes: Define the data packet priority classification rule, and mark the real-time monitoring data as a high-priority transmission queue; Adopt time-division multiplexing technology to alternately switch the wireless local area network and cellular network channels, and calculate the link quality assessment index; Enhance the anti-interference ability of the data packet through the forward error correction coding algorithm, and generate a redundant check code and append it to the end of the transmission frame.

[0013] Preferably, the establishment of the process feature mapping relationship includes: Introduce dilated convolutional kernels in the feature extraction layer to capture the long-range process parameter correlation, and output a multi-scale process feature tensor; In the feature fusion layer, adopt a gated attention mechanism to weighted aggregate spatio-temporal features, and introduce skip connections to avoid gradient dispersion; Optimize the parameters of the feature encoder and decoder through the adversarial generation strategy, and minimize the process parameter prediction error boundary.

[0014] Preferably, the system further includes: A welding quality simulation module for establishing a multi-physical field coupling model based on the heat conduction equation to simulate the dynamic evolution process of the molten pool; Input the real-time process parameters into the simulation model to generate a virtual weld morphology map, and trigger a process deviation warning signal through difference comparison.

[0015] Preferably, the generation of the multi-dimensional spatio-temporal process data set includes: Synchronously collect the welder output signal, the infrared thermal imager temperature field data, and the laser displacement sensor deformation reading; Adopt a tensor decomposition algorithm to extract features from heterogeneous data and construct a low-dimensional approximate data matrix; Eliminate the sensor baseline drift error through the median filtering algorithm and enhance the signal-to-noise ratio of weak feature signals.

[0016] Preferably, the generation of the adaptive process control instruction includes: Establish a multi-objective constraint optimization model with process stability, equipment life, and production efficiency as boundary conditions; Use the sequential quadratic programming algorithm to solve the convex optimization problem with non-linear constraints and generate a feasible solution set of pulse parameters; Evaluate the multi-index matching degree of the solution set through the grey relational analysis method and output the optimal process adjustment plan.

[0017] Preferably, the system further includes: Construct an abnormal process pattern library based on the variational autoencoder and perform unsupervised feature learning on historical welding defect data; Use the dynamic time warping algorithm to extract the transient abnormal patterns of real-time signals and generate the classification results of potential defect types.

[0018] Preferably, the present invention further includes a method for intelligent acquisition of welding parameters and process optimization management. The method includes the following steps: Step 1: Synchronously collect welding current, voltage, temperature, and workpiece deformation data through a multi-modal sensor array to generate a multi-dimensional spatio-temporal process data set; Step 2: Perform non-linear noise reduction processing on the multi-dimensional spatio-temporal process data set to generate a time-domain stationary signal sequence; Step 3: Construct an optimization model based on reinforcement learning and dynamically adjust the pulse frequency and energy input parameters according to real-time welding quality indicators; Step 4: Design a dual-channel transmission protocol and alternately upload data packets through a wireless local area network and a cellular network; Step 5: Establish a process feature mapping relationship based on a deep neural network, perform multi-objective optimization modeling on the historical process data set, and generate an optimal welding parameter configuration plan.

[0019] Compared with the prior art, the beneficial effects of the present invention are: The intelligent acquisition system and method for welding parameters and process optimization management proposed by the present invention provide a comprehensive and effective solution to the long-existing problems in the welding industry and bring significant benefits in many aspects. In terms of data acquisition, the multi-source data acquisition module synchronously collects welding current, voltage, temperature, and workpiece deformation data through a multi-modal sensor array to generate a multi-dimensional spatio-temporal process data set. This all-round data acquisition method can accurately capture the real-time changes of various physical quantities during the welding process. For example, in the welding of bridge steel structures, the precise monitoring of temperature and workpiece deformation can timely master the range of the heat-affected zone and the structural deformation situation during welding, prevent structural problems caused by thermal stress and deformation in advance, ensure the reliability of the bridge welding quality, and guarantee the long-term safe use of the bridge.

[0020] The signal preprocessing module uses a joint algorithm of wavelet packet threshold denoising and Kalman filtering to perform non-linear noise reduction on the collected data, generating a time-domain stationary signal sequence. This processing effectively removes the noise interference in the data, making the subsequent data analysis and processing more accurate and reliable. The data after noise reduction can truly reflect the actual situation of the welding process, providing a solid data basis for the accurate evaluation of welding quality and the scientific adjustment of process parameters, and avoiding misjudgment and wrong decisions caused by noise.

[0021] The dynamic parameter regulation module constructs an optimization model based on reinforcement learning, dynamically adjusts the pulse frequency and energy input parameters according to the real-time welding quality indicators, and generates adaptive process regulation instructions. In actual welding production, in the face of complex and changeable welding conditions, this module can optimize the welding parameters in real time. For example, when welding the pins of an electronic chip, it can accurately adjust the energy input and pulse frequency according to the material, size of the pins and the real-time welding quality feedback, ensuring the stable quality of the solder joints, avoiding problems such as false soldering and short circuits, and improving the production yield and reliability of electronic products.

[0022] The heterogeneous network communication module designs a dual-channel transmission protocol, which alternately uploads data packets using a wireless local area network and a cellular network, ensuring the redundancy and fault tolerance of the communication link. In an industrial production environment, network signals are easily interfered with and unstable. This communication method can effectively avoid data loss or transmission interruption problems caused by network failures. In the remote welding monitoring scenario, regardless of the on-site network conditions, it can ensure the stable transmission of welding data, providing timely and accurate data support for remote operators, and realizing real-time monitoring and remote control of the welding process.

[0023] The process optimization decision module establishes a process feature mapping relationship based on a deep neural network, performs multi-objective optimization modeling on the historical process data set, and generates an optimal welding parameter configuration plan. This innovative technology breaks the traditional mode of optimizing processes relying on experience, and realizes the efficient optimization of welding processes by leveraging the powerful capabilities of big data and deep learning. During the new product R & D or process improvement process, through in-depth analysis of historical data and model prediction, it can quickly find suitable combinations of welding parameters, greatly shortening the R & D cycle, reducing the R & D cost, and improving the market response speed and competitiveness of the enterprise.

[0024] The welding quality simulation module establishes a multi - physical - field coupling model based on the heat conduction equation, simulates the dynamic evolution process of the molten pool, and triggers a process deviation warning signal through the difference between the virtual weld appearance map and the actual situation. This enables operators to preview the welding process before welding, discover potential process problems in advance, adjust welding parameters in a timely manner, and effectively avoid the generation of welding defects. In the welding of pressure vessels, through simulation prediction, the welding process can be optimized in advance to ensure that the weld quality meets the safety requirements under high - pressure environments, reducing the risk of safety accidents caused by welding defects.

[0025] The abnormal process pattern library constructed based on variational auto - encoders and the transient abnormal pattern extraction of real - time signals using the dynamic time warping algorithm can timely detect potential welding defects and generate classification results of potential defect types. This helps operators quickly locate the root cause of problems and take targeted measures for repair, ensuring the continuity and stability of welding production. In pipeline welding production, welding defects can be detected and processed in a timely manner, avoiding a large number of product rejections caused by the accumulation of defects, improving production efficiency, and reducing production costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is the working principle diagram of the intelligent welding parameter acquisition and process optimization management system described in the present invention; Figure 2 is the working principle diagram for constructing the optimization model; Figure 3 is the working principle diagram for establishing the process feature mapping relationship; Figure 4 is the working principle diagram of the welding quality simulation module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0028] Please refer to Figures 1 - 4 , the present invention provides an intelligent welding parameter acquisition and process optimization management system, and the system includes: Multi - source data acquisition module: Simultaneously collect welding current, voltage, temperature, and workpiece deformation data through a multi - modal sensor array. For example, use a high - precision current sensor to collect welding current, a voltage sensor to collect voltage, an infrared thermal imager to obtain temperature data, and a laser displacement sensor to measure workpiece deformation. Integrate the collected data to generate a multi - dimensional spatio - temporal process data set, providing basic data for subsequent analysis and processing.

[0029] Signal preprocessing module: For the multi-source data acquisition module-generated multi-dimensional spatio-temporal process data set, this module performs non-linear noise reduction processing. By eliminating high-frequency interference components, the original data set is transformed into a time-domain stationary signal sequence, improving the quality and stability of the data for subsequent modules to perform accurate analysis.

[0030] Dynamic parameter regulation module: Construct an optimization model based on reinforcement learning. This model dynamically adjusts the pulse frequency and energy input parameters according to real-time welding quality indicators. During the actual welding process, the welding quality indicators change in real time with the welding process. The optimization model can adjust the parameters in a timely manner according to these changes, generating adaptive process regulation instructions to ensure welding quality.

[0031] Heterogeneous network communication module: Design a dual-channel transmission protocol, and use wireless local area network and cellular network to alternately upload data packets. In the complex communication environment of the industrial site, a single network may have problems such as signal interruption. The dual-channel transmission protocol can ensure the redundancy and fault tolerance of the communication link and guarantee the reliability of data transmission.

[0032] Process optimization decision-making module: Establish a process feature mapping relationship based on a deep neural network, and perform multi-objective optimization modeling on the historical process data set. Through the learning and analysis of a large amount of historical data, an optimal welding parameter configuration scheme is generated to provide decision-making support for the optimization of the welding process.

[0033] The following further describes the implementation of the present invention in combination with Embodiments 1 to 5.

[0034] Embodiment 1: When the signal preprocessing module performs non-linear noise reduction processing on the multi-dimensional spatio-temporal process data set, a joint algorithm of wavelet packet threshold denoising and Kalman filtering is used to separate transient noise and steady-state process signals. The specific steps are as follows: Perform wavelet packet multi-scale decomposition on the welding current signal to extract the energy ratio distribution of each sub-band. Assume the welding current signal is , and decompose it into different sub-bands through wavelet packet transform. The energy of the th layer and the th sub-band is , then the calculation formula for the energy ratio of this sub-band is: Among them, represents the number of layers of wavelet packet decomposition, and represents the sub-band serial number in this layer. By analyzing these energy ratio distributions, the energy distribution of the signal in different frequency components can be understood, providing a basis for subsequent noise separation.

[0035] Construct the Kalman filter state equation, use the high-frequency noise component as the observation variable, and iteratively update the signal estimation value. Let the state variable represent the signal state at time, and the state transition equation is , where is the state transition matrix, is the process noise. The observation equation is , is the observation value at time, is the observation matrix, is the observation noise. In this embodiment, the high-frequency noise component is used as the observation value , and through continuous iteration and update by the Kalman filter algorithm, a more accurate signal estimation value is obtained, thereby effectively suppressing the high-frequency noise in the signal.

[0036] Identify abnormal current fluctuation segments through the analysis of variance method, and segment the normal welding process signal interval. Calculate the variance of the current signal in different time periods. The time period with a larger variance may have abnormal current fluctuations. When the variance exceeds the set threshold, it is determined that there are abnormal current fluctuations in this time period, and then the normal welding process signal interval is segmented, which provides convenience for accurately analyzing the process signal in the normal welding process later.

[0037] In actual application scenarios, such as in the welding production line of an automobile manufacturing workshop, the welding current signal is easily affected by various electromagnetic interferences. Using the method of this embodiment, after wavelet packet multi-scale decomposition, it is found that the energy ratio of the high-frequency sub-band is too high, confirming the existence of a large amount of high-frequency noise. After processing the signal using the Kalman filter algorithm, the stability of the signal is significantly improved. By accurately identifying the abnormal current fluctuation segments through the analysis of variance method and separating them from the normal welding process signal interval, the subsequent analysis of the normal welding process signal is more accurate and reliable, providing strong support for ensuring the welding quality.

[0038] Embodiment 2: This embodiment is used to describe the construction process of the optimization model in the dynamic parameter regulation module. When constructing an optimization model based on reinforcement learning in the dynamic parameter regulation module, the specific operations are as follows: Define the state space as a combined vector of current stability, molten pool morphology score, and energy utilization rate. Let the current stability be , the molten pool morphology score is obtained by analyzing the molten pool image, etc., denoted as , and the energy utilization rate is , then the state space vector . Among them, the current stability It can be measured by calculating the standard deviation of the welding current over a period of time. The smaller the standard deviation, the higher the current stability; the molten pool morphology score Score according to the characteristics of the molten pool shape, size, etc. The higher the score, the better the molten pool morphology; the energy utilization rate It is the ratio of the energy effectively used for welding to the total input energy during the welding process.

[0039] Construct a deep deterministic policy gradient algorithm framework and adopt an experience replay mechanism to optimize the convergence speed of the policy network. The deep deterministic policy gradient algorithm learns the optimal policy by constructing a policy network and a value network. During the training process, the experience replay mechanism stores the experience samples generated by the agent in the environment into the experience replay buffer, where is the state at time is the action taken at time is the reward obtained after executing the action and is the state at time. Randomly sample a batch of samples from the experience replay buffer for training, which avoids the correlation between samples and speeds up the convergence speed of the policy network.

[0040] Design the reward function as the weighted difference between the weld forming quality score and the energy consumption penalty term to drive the model to generate a dynamic regulation strategy. Let the weld forming quality score be and the energy consumption penalty term be with weights and respectively. Then the calculation formula of the reward function is . The weld forming quality score is scored according to factors such as the appearance quality and internal defects of the weld. The higher the score, the better the weld forming quality; the energy consumption penalty term is proportional to the energy consumption during the welding process. The higher the energy consumption, the larger the penalty term. Through this reward function, while optimizing the model to pursue good weld forming quality, it will try to reduce energy consumption as much as possible, thereby generating a more reasonable dynamic regulation strategy.

[0041] ​Taking the welding of large structural parts in shipbuilding as an example, during the welding process, due to the large size of the structural parts and the long welding time, there are high requirements for current stability, molten pool shape, and energy utilization rate. By using the optimization model constructed in this embodiment, at the initial stage of welding, the model selects an action (such as adjusting the pulse frequency and energy input parameters) through the policy network according to the initial state space vector. As welding progresses, the reward value is calculated based on the real-time feedback of the weld formation quality score and energy consumption. The policy network is trained using the experience replay mechanism to continuously optimize the policy. After multiple iterations, the model can accurately adjust the pulse frequency and energy input parameters according to different welding states, achieving a high standard of weld formation quality while effectively reducing energy consumption and improving welding efficiency and quality.

[0042] Example 3: This embodiment mainly elaborates on the design details of the dual-channel transmission protocol in the heterogeneous network communication module. When designing the dual-channel transmission protocol in the heterogeneous network communication module, the specific implementation method is as follows: Define the data packet priority classification rule and mark the real-time monitoring data as the high-priority transmission queue. During the welding process, real-time monitoring data (such as key parameters like welding current and voltage) is crucial for promptly grasping the welding state, so it is marked as high-priority. Let the data packet type be , when is the real-time monitoring data type, mark its priority as (high priority); for other non-real-time data, such as historical process data backup, etc., the priority is marked as (low priority). In this way, during data transmission, high-priority real-time monitoring data can be processed and transmitted first, ensuring the timeliness of the data.

[0043] Adopt time-division multiplexing technology to alternately switch the wireless local area network and cellular network channels, and calculate the link quality assessment index. Let the channel of the wireless local area network be , and the channel of the cellular network be . Within each time slice , select the channel for data transmission according to a certain switching rule. By calculating the link quality assessment index in real time, a channel with better quality can be dynamically selected for data transmission, improving the reliability of data transmission.

[0044] Enhance the anti-interference ability of data packets through the forward error correction coding algorithm, and generate a redundant check code to append to the end of the transmission frame. If the received data is interfered and errors occur during transmission, as long as the number of errors is within the error correction ability of the error correction code, the correct data can be restored through the redundant check code, thus enhancing the anti-interference ability of the data packet.

[0045] In an actual industrial environment, such as the welding workshop in a steel mill, there is a large amount of electromagnetic interference and the communication environment is complex. By using the dual-channel transmission protocol designed in this embodiment, the real-time monitoring data is preferentially transmitted, ensuring the real-time monitoring of the welding process. Through time-division multiplexing technology, the wireless local area network and cellular network channels are alternately used, and the channel is dynamically selected according to the link quality assessment index, effectively avoiding the problem of data transmission interruption caused by poor quality of a single network channel. The redundant check code generated by the forward error correction coding algorithm enables the data packet to be accurately restored when it is interfered, ensuring the integrity and accuracy of data transmission and providing a strong guarantee for the reliable transmission of welding parameters.

[0046] Embodiment 4: When establishing the process feature mapping relationship based on the deep neural network in the process optimization decision-making module, the specific implementation is as follows: At the feature extraction layer, dilated convolutional kernels are introduced to capture the long-range process parameter correlation, and a multi-scale process feature tensor is output. By convolutional kernels with different dilation rates, the correlation between process parameters can be captured at different scales, and these multi-scale features are combined to form a multi-scale process feature tensor , providing richer feature information for subsequent analysis.

[0047] At the feature fusion layer, a gated attention mechanism is used to weighted aggregate spatio-temporal features, and skip connections are introduced to avoid gradient dispersion. The gated attention mechanism weighted aggregates the spatio-temporal features by calculating the importance weights of different features. Let the input spatio-temporal feature be , and the gated weight be , then the weighted aggregated feature The calculation formula is . The skip connection directly connects the input of the feature extraction layer to the output of the feature fusion layer, enabling the gradient to be transmitted more smoothly during backpropagation, avoiding the problem of gradient dispersion, and ensuring the effective training of the deep neural network.

[0048] By using the adversarial generation strategy to optimize the parameters of the feature encoder and decoder, the prediction error boundary of the process parameters is minimized. A generator and a decoder are constructed. The generator tries to generate samples similar to the real process parameters, and the decoder is used to judge whether the generated samples are real. Let the real process parameter be , the generated process parameter be , and the prediction error be . Through adversarial training, the generator and the decoder parameters to minimize the prediction error boundary, thereby improving the accuracy of process parameter prediction and providing a more reliable basis for generating an optimal welding parameter configuration plan.

[0049] In the precision welding scenario in the field of mechanical manufacturing, the correlation between welding process parameters is complex, and the accuracy requirements for process parameter prediction are relatively high. By using the process feature mapping relationship established in this embodiment, the long-range process parameter correlation is successfully captured through the dilated convolution kernel, and the multi-scale process feature tensor contains rich information. The gated attention mechanism effectively aggregates spatio-temporal features, and the skip connection ensures the stable training of the deep neural network. After being optimized by the adversarial generation strategy, the process parameter prediction error is significantly reduced, and the generated optimal welding parameter configuration plan can better meet the requirements of precision welding, improving the quality and production efficiency of welded products.

[0050] Example 5: This embodiment mainly focuses on the implementation of other functional modules in the system, including the construction of a welding quality simulation module and an abnormal process mode library. In this system, in addition to the above core modules, the following functional modules are also available: Welding quality simulation module: Establish a multi-physical field coupling model based on the heat conduction equation to simulate the dynamic evolution process of the molten pool. Substitute various physical parameters in the welding process (such as the heat source intensity converted from welding current and voltage) into this equation, and combine other relevant physical field equations (such as fluid mechanics equations, etc.) to construct a multi-physical field coupling model. Input the real-time process parameters into the simulation model to generate a virtual weld appearance map, and trigger a process deviation warning signal through difference comparison. Let a certain characteristic parameter of the virtual weld appearance be , and the corresponding characteristic parameter of the actual weld appearance be . When ( is the set deviation threshold), trigger a process deviation warning signal to prompt the operator to adjust the welding process parameters in time to ensure the welding quality.

[0051] Construction of abnormal process mode library: Construct an abnormal process mode library based on variational autoencoders for unsupervised feature learning of historical welding defect data. The variational autoencoder encodes the input data into a latent vector and then decodes and reconstructs the data from the latent vector. Let the input historical welding defect data be , the encoding function be , the decoding function be , and learn the feature representation of the data by minimizing the reconstruction error . Use the dynamic time warping algorithm to extract the transient abnormal mode of the real-time signal and generate the classification result of potential defect types. Let the real-time signal be , and the historical abnormal signal template be , the dynamic time warping algorithm calculates the distance between the two , when is less than a certain threshold, the real-time signal is classified into the corresponding potential defect type, providing support for timely discovery and solution of potential problems in the welding process.

[0052] In the welding process of aerospace components, the welding quality requirements are extremely high. By using the welding quality simulation module, before welding, simulations are carried out by inputting the estimated process parameters to discover possible process problems in advance and adjust the parameters. During the welding process, the process parameters are input into the simulation model in real time. Once there is a large deviation between the virtual weld appearance and the actual weld appearance, a warning signal is triggered in a timely manner. At the same time, through the abnormal process mode library, transient abnormal modes in the real-time signal can be quickly identified, potential defect types can be judged, and corresponding measures can be taken for repair, effectively ensuring the welding quality and safety of aerospace components.

[0053] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0054] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent welding parameter acquisition and process optimization management system, characterized in that The system includes: A multi-source data acquisition module, which is used to synchronously acquire welding current, voltage, temperature and workpiece deformation data through a multi-modal sensor array, and generate a multi-dimensional spatio-temporal process data set; A signal preprocessing module, which is used to perform non-linear noise reduction processing on the multi-dimensional spatio-temporal process data set, eliminate high-frequency interference components, and generate a time-domain stationary signal sequence; A dynamic parameter regulation module, which is used to build an optimization model based on reinforcement learning, dynamically adjust the pulse frequency and energy input parameters according to real-time welding quality indicators, and generate an adaptive process regulation instruction; A heterogeneous network communication module, which is used to design a dual-channel transmission protocol, alternately upload data packets through a wireless local area network and a cellular network, and ensure the redundancy and fault tolerance of the communication link; A process optimization decision-making module, which is used to establish a process feature mapping relationship based on a deep neural network, perform multi-objective optimization modeling on the historical process data set, and generate an optimal welding parameter configuration scheme.

2. The intelligent acquisition and process optimization management system for welding parameters according to claim 1, characterized in that The non-linear noise reduction processing uses a joint algorithm of wavelet packet threshold denoising and Kalman filtering to separate transient noise and steady-state process signals, including: Perform wavelet packet multi-scale decomposition on the welding current signal, and extract the energy ratio distribution of each sub-band; Build a Kalman filter state equation, use the high-frequency noise component as the observation variable, and iteratively update the signal estimate value; Identify abnormal current fluctuation segments through analysis of variance, and segment the normal welding process signal interval.

3. The intelligent acquisition and process optimization management system for welding parameters according to claim 1, characterized in that The construction of the optimization model includes: Define the state space as a combined vector of current stability, molten pool morphology score and energy utilization rate; Build a deep deterministic policy gradient algorithm framework, and use an experience replay mechanism to optimize the convergence speed of the policy network; Design the reward function as the weighted difference between the weld formation quality score and the energy consumption penalty term, and drive the model to generate a dynamic regulation strategy.

4. The intelligent acquisition and process optimization management system for welding parameters according to claim 3, characterized in that The design of the dual-channel transmission protocol includes: Define the data packet priority classification rule, and mark the real-time monitoring data as a high-priority transmission queue; Use time-division multiplexing technology to alternately switch the wireless local area network and cellular network channels, and calculate the link quality evaluation index; Enhance the anti-interference ability of the data packet through the forward error correction coding algorithm, and generate a redundant check code and append it to the end of the transmission frame.

5. The intelligent acquisition and process optimization management system for welding parameters according to claim 4, wherein The establishment of the process feature mapping relationship includes: Introduce dilated convolutional kernels in the feature extraction layer to capture the long-range process parameter correlation, and output a multi-scale process feature tensor; Adopt a gated attention mechanism in the feature fusion layer to weighted aggregate spatio-temporal features, and introduce skip connections to avoid gradient dispersion; Optimize the parameters of the feature encoder and decoder through an adversarial generation strategy, and minimize the process parameter prediction error boundary.

6. The intelligent acquisition and process optimization management system for welding parameters according to claim 1, wherein The system further includes: A welding quality simulation module, which is used to establish a multi-physics field coupling model based on the heat conduction equation and simulate the dynamic evolution process of the molten pool; Input the real-time process parameters into the simulation model to generate a virtual weld morphology map, and trigger a process deviation warning signal through difference comparison.

7. The intelligent acquisition and process optimization management system for welding parameters according to claim 2, characterized in that, The generation of the multi-dimensional spatio-temporal process data set includes: Synchronously acquire the welding machine output signal, the temperature field data of the infrared thermal imager and the deformation reading of the laser displacement sensor; Use a tensor decomposition algorithm to extract features from heterogeneous data and construct a low-dimensional approximate data matrix; Eliminate the baseline drift error of the sensor through the median filtering algorithm and enhance the signal-to-noise ratio of weak feature signals.

8. The intelligent acquisition and process optimization management system for welding parameters according to claim 7, characterized in that, The generation of the adaptive process control instruction includes: Establish a multi-objective constraint optimization model with process stability, equipment life, and production efficiency as boundary conditions; Use the sequential quadratic programming algorithm to solve the convex optimization problem with nonlinear constraints and generate a feasible solution set of pulse parameters; Evaluate the multi-index matching degree of the solution set through the grey relational analysis method and output the optimal process adjustment plan.

9. The intelligent acquisition and process optimization management system for welding parameters according to claim 1, characterized in that, The system further includes: Construct an abnormal process pattern library based on the variational autoencoder and perform unsupervised feature learning on historical welding defect data; Use the dynamic time warping algorithm to extract the transient abnormal pattern of the real-time signal and generate the classification result of potential defect types.

10. An intelligent acquisition and process optimization management method for welding parameters, characterized in that The method includes the following steps: Step 1: Synchronously collect welding current, voltage, temperature, and workpiece deformation data through a multi-modal sensor array to generate a multi-dimensional spatio-temporal process data set; Step 2: Perform nonlinear noise reduction processing on the multi-dimensional spatio-temporal process data set to generate a time-domain stationary signal sequence; Step 3: Construct an optimization model based on reinforcement learning and dynamically adjust the pulse frequency and energy input parameters according to the real-time welding quality index; Step 4: Design a dual-channel transmission protocol and alternately upload data packets through the wireless local area network and the cellular network; Step 5: Establish a process feature mapping relationship based on a deep neural network, perform multi-objective optimization modeling on the historical process data set, and generate an optimal welding parameter configuration plan.

Images