Friction stir welding system pressure control method and system, electronic equipment and storage medium

Abstract

The invention relates to the technical field of high-end equipment manufacturing, in particular to a friction stir welding system pressure control method and system, electronic equipment and a storage medium, and the method comprises the steps that real-time sensor data in the welding process is obtained, and a controller preset algorithm module is used for analyzing and processing to obtain a real-time pressure value; and meanwhile, a plurality of modules such as a self-adaptive control algorithm module, a pressure prediction model, an online parameter identification model, a multi-mode switching module, an abnormal prediction model and an equipment health state evaluation model are further arranged to realize dynamic adjustment and control. The effects that the working pressure value of the friction stir welding system can be continuously and dynamically adjusted, the welding process is kept stable, self-adaptive control and precise control are achieved, the welding quality stability is improved, abnormity is early warned, parameters are adjusted, and the fault-tolerant threshold value is dynamically adjusted to maintain the control precision are achieved.

Description

Technical Field

[0001] This invention relates to the field of high-end equipment manufacturing technology, and in particular to a pressure control method and system for friction stir welding systems, electronic devices, and storage media. Background Technology

[0002] Friction stir welding (FSW) technology is widely used in the aerospace, automotive, and shipbuilding industries, particularly in the welding of large aluminum alloy structural components, body frame welding, and hull assembly connections. Due to its high efficiency, environmental friendliness, and high quality, FSW has been extensively applied in the manufacture of metal structural components. It has significantly boosted the development of these industries, making the manufacturing of related products more efficient and reliable, and powerfully propelling these industries towards higher levels of development. In the field of FSW, the precision of pressure control directly affects the weld quality, and therefore has always been a key focus and research area within the industry.

[0003] In traditional friction stir welding, two main methods are used to control welding pressure. One method relies on manual pressure adjustment by operators, leading to significant human-induced variations in weld quality. This is especially problematic when welding material properties change or the workpiece surface is uneven, with pressure fluctuations reaching ±20% and a defect rate exceeding 15%. The other method employs automated control, incorporating model-based control algorithms such as PID and pole placement. However, limited by fixed model parameters and lag-based feedback mechanisms, it struggles to adapt to dynamically changing welding conditions. Control errors typically exceed ±5%, and overshoot is greater than 18%, failing to meet the stringent precision requirements of high-end manufacturing sectors like aerospace.

[0004] Furthermore, in the existing technology, the pressure control system proposed in CN201410369938.0 achieves steady-state error-free pressure control by combining an internal model controller with linear quadratic optimal control. However, it still has significant technical weaknesses. On the one hand, the parameters of the internal model controller are fixed, making it unable to adapt to the pressure control requirements of multiple materials and multiple working conditions. On the other hand, the system lacks the ability to predict sudden abnormal events and can only respond passively after an anomaly occurs, increasing the risk of equipment damage. Summary of the Invention

[0005] The purpose of this application is to overcome the above-mentioned technical problems and provide a pressure control method, system, equipment and storage medium for a friction stir welding system.

[0006] In a first aspect, this application provides a pressure control method for a friction stir welding system, employing the following technical solution: A pressure control method for a friction stir welding system includes the following steps: Acquire real-time sensor data during the welding process, wherein the real-time sensor data includes at least material thickness, material hardness, current output value, pressure, temperature and welding speed parameters; Based on real-time sensor data, the controller uses a preset algorithm module to analyze and process the real-time sensor data to obtain the real-time pressure value. The controller sends the real-time pressure value to the actuator, which then outputs instructions to adjust the working pressure value of the friction stir welding system according to the real-time pressure value to adapt to the current welding conditions. Continuously acquire real-time sensor data, and dynamically adjust the working pressure value based on the welding information fed back by the real-time sensor data to maintain the stability of the welding process.

[0007] By adopting the above technical solution, real-time sensor data during the welding process is acquired, analyzed, and processed to obtain real-time pressure values. The working pressure of the friction stir welding system is then adjusted to adapt to the current welding conditions. Furthermore, by continuously acquiring and feeding back real-time sensor data, the working pressure is dynamically adjusted to maintain the stability of the welding process.

[0008] Optionally, the controller includes an adaptive control algorithm module, which includes at least welding material characteristic data and environmental change data, for dynamically adjusting control parameters; The adaptive control algorithm module includes at least one of the following: a fuzzy control algorithm or a neural network control algorithm.

[0009] By adopting the above technical solution and utilizing the adaptive control algorithm module, the parameters of the welding process execution system are dynamically adjusted in combination with welding material characteristic data and environmental change data. By using fuzzy control algorithm or neural network control algorithm, the problem of insufficient accuracy of traditional manual adjustment and fixed model control is solved, and the system's adaptability to multiple materials, equipment aging, and sudden anomalies is improved.

[0010] Optionally, the controller further includes a pressure prediction model, and constructing the pressure prediction model and predicting the working pressure value includes: The big data analysis module is accessed to train a machine learning model using historical welding data. Real-time sensor data is also accessed to build a pressure prediction model, which predicts the optimal working pressure value under different welding materials, workpiece thicknesses, and ambient temperature conditions.

[0011] By adopting the above technical solution, a pressure prediction model can be trained based on historical data and real-time sensor data. This enables the pre-calculation of the optimal working pressure under different welding materials, workpiece thicknesses, and ambient temperatures, reducing the cost of manual trial and error and improving efficiency and consistency of welding quality.

[0012] Optionally, the controller includes an online parameter identification model, and the steps for real-time estimation of the actuator's execution parameters and pressure closed-loop control are as follows: The online parameter identification model is embedded into the closed-loop control loop of the actuator, and the control command signal of the actuator and real-time sensor data are used as inputs to the online parameter identification model. The extended Kalman filter algorithm is retrieved to estimate the execution parameters in real time. The initial values ​​of the motor torque constant and the sensor zero drift preset values, which are obtained offline, are used as the initial parameters of the extended Kalman filter algorithm to construct the system state equation and observation equation. Based on the input of the online parameter identification model, the extended Kalman filter algorithm is used to dynamically estimate the real-time execution parameters; The model predictive control module is invoked to process real-time execution parameters. The estimated real-time execution parameters are transmitted to the model predictive control module. Noise interference is eliminated through moving average filtering and normalization. The frequency estimated by the extended Kalman filter algorithm is consistent with the control cycle of the model predictive control module. The residual verification module is invoked to calculate the parameter deviation. When the deviation between the measured value and the input requirement of the prediction model is too large, a parameter update command is triggered. The model predictive control module performs rolling optimization prediction of pressure change trends for the next 3 to 5 control cycles based on processed real-time execution parameters, historical pressure data, and real-time operating conditions. It calculates and outputs working pressure correction values, drives the actuator to adjust the pressure, and achieves closed-loop control.

[0013] By adopting the above technical solutions, real-time acquisition of multi-dimensional sensor data and dynamic pressure adjustment can achieve real-time adaptation of welding pressure to working conditions, thereby improving the stability of the welding process. Furthermore, the adaptive control algorithm is used to dynamically adjust parameters by integrating material properties and environmental data, thereby improving control robustness. The online parameter identification model is embedded into the closed-loop control loop of the actuator, and the actuator parameters are estimated in real time through the extended Kalman filter algorithm. Combined with residual verification and rolling optimization of the model predictive control module, dynamic calibration of pressure control is achieved, reducing control error and improving control accuracy.

[0014] Optionally, the controller is equipped with a multi-modal switching module to acquire real-time workpiece spectral data collected by a spectral sensor, retrieve convolutional neural network to identify material type, and trigger control parameters for the corresponding mode. The big data analysis module is invoked to obtain real-time data on welding current, speed, and penetration depth as input data. The control parameters of each mode are optimized every fifty cycles using a reinforcement learning algorithm.

[0015] By adopting the above technical solution, the workpiece spectral data collected by the spectral sensor is acquired, and the material type is identified using a convolutional neural network to trigger the corresponding modal control parameters. This reduces the parameter switching time for welding different materials from 20 minutes manually to real-time response. A big data analysis module acquires real-time data on welding current, speed, and penetration depth. A reinforcement learning algorithm optimizes the control parameters for each modality every fifty cycles, keeping the penetration depth error within ±0.2 mm, thus improving welding quality stability. Furthermore, material control logic adapted to the current welding conditions is incorporated, and data collaboration is achieved with the material type input of the pressure prediction model.

[0016] Optionally, the controller further includes an anomaly prediction model, which achieves pressure anomaly early warning and control through the following steps: The convolutional neural network module is invoked to perform two-dimensional feature modeling on the pressure fluctuation sequence and current signal spectrum in historical welding data, and abnormal events are labeled to train and obtain an anomaly prediction model. The anomaly prediction model performs rolling predictions on real-time pressure and current data. When the predicted anomaly probability is greater than the warning threshold and the pressure value reaches the rated value, a warning signal is immediately output to trigger the controller to adjust the execution parameters. When abnormal pressure is confirmed, the controller executes a fast rollback algorithm and synchronously adjusts process parameters to reduce the pressure, welding speed, and welding rotation speed.

[0017] By employing the above technical solution, a two-dimensional feature model is created using a convolutional neural network module to analyze the pressure fluctuation sequence and current signal spectrum of historical welding data. This model trains an anomaly prediction model that performs rolling predictions on real-time pressure and current data, outputting early warning signals and triggering the controller to adjust execution parameters. This effectively reduces the scrap rate caused by welding anomalies and minimizes the risk of equipment damage. When abnormal pressure is confirmed, the controller executes a fast rollback algorithm and synchronously adjusts process parameters, reducing the pressure, welding speed, and welding rotation speed, further ensuring the stability and safety of the welding process.

[0018] Optionally, the controller further includes a device health status assessment model. Using data obtained from the online parameter identification model, the controller's fault tolerance threshold is automatically adjusted to construct the device health status assessment model. Specifically, dynamic fault tolerance threshold adjustment is achieved through the following steps: Obtaining the motor torque constant drift rate and sensor zero drift change output by the online parameter identification model, and obtaining the pressure fluctuation standard deviation and current ripple coefficient from real-time sensor data, denoted as ∆Kt Kt0, ∆b, σp, and γi respectively, constructs a four-dimensional health vector H: H = [∆Kt Kt0 , ∆b, σp, γi ]; Establishing a mapping relationship between health characteristics and fault tolerance thresholds based on historical data, using a linear regression model: Tnew = T0 + α× ∆Kt Kt0 + β × ∆b + δ × σp + ϵ × γi, where T0 is the initial threshold, and α, β, δ, and ϵ are weighting coefficients; When the health vector H... When the threshold adjustment is triggered, the controller parameter self-tuning is started simultaneously, with the pressure control error and step response overshoot ≤5% and the adjustment time ≤200ms.

[0019] By adopting the above technical solution, and combining the motor torque constant drift rate, sensor zero drift change, and pressure fluctuation standard deviation and current ripple coefficient output by the online parameter identification model, a four-dimensional health vector is constructed. Based on historical data, a linear regression model is used to establish the relationship between health characteristics and fault tolerance threshold, automatically adjust the controller fault tolerance threshold, and then start the controller parameter self-tuning. This ensures that the pressure control error and step response overshoot remain at a low level after the equipment has been running continuously for a long time, with a short adjustment time, and solves the problem that traditional fixed thresholds cannot cope with equipment aging.

[0020] Secondly, this application provides a pressure control system for a friction stir welding system, which adopts the following technical solution: A pressure control system for a friction stir welding system, comprising: The sensor module includes a spectral sensor, a pressure sensor, a current sensor, and a temperature sensor, used to acquire workpiece spectral data, welding pressure, drive current, and ambient temperature in real time. The actuator is connected to the controller, receives pressure control signals, and realizes automatic adjustment of working pressure and rapid retraction in case of abnormality. The big data analysis module stores historical welding data and uses machine learning algorithms to train a pressure prediction model. The controller has multiple built-in modules: The adaptive control algorithm module is used to dynamically adjust control parameters; A pressure prediction model is used to achieve adaptive control of the friction stir welding system. The multimodal switching module uses a convolutional neural network to identify the material type of spectral data and triggers the control parameters of the corresponding mode. An online parameter identification model is used to accurately estimate execution parameters in real time, dynamically optimize control, and achieve precise pressure control. The model predictive control module is used to achieve precise pressure control; The multi-modal switching module is used to identify material type in real time, automatically switch control parameters, dynamically optimize PID parameters, and improve welding quality stability. Anomaly prediction models are used to provide early warnings of welding anomalies and pressure anomalies, and to adjust parameters accordingly. Equipment health status assessment model is used to dynamically adjust fault tolerance thresholds and maintain control accuracy.

[0021] By adopting the above technical solutions, the sensor module collects workpiece spectral data, welding pressure, driving current, and ambient temperature in real time; the actuator can receive pressure control signals to achieve automatic adjustment of working pressure and rapid retraction in case of abnormalities; the big data analysis module can store historical welding data and train a pressure prediction model; the adaptive control algorithm module can dynamically adjust control parameters, and the pressure prediction model can realize adaptive control of the friction stir welding system; the multi-modal switching module can identify material types, trigger control parameters for the corresponding mode, and dynamically optimize PID parameters to improve welding quality stability; the online parameter identification model can accurately estimate execution parameters in real time and dynamically optimize control, achieving precise pressure control in conjunction with the model predictive control module; the anomaly prediction model can provide early warning of welding and pressure anomalies and adjust parameters synchronously; the equipment health status assessment model can dynamically adjust the fault tolerance threshold to maintain control accuracy, ultimately achieving full-process automation from data acquisition, model analysis, parameter adjustment to execution control, improving welding efficiency and overall yield.

[0022] Thirdly, this application provides a computer electronic device, which adopts the following technical solution: A computer electronic device includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the method as described in any of the first aspects.

[0023] Fourthly, this application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method as described in any of the first aspects.

[0024] In summary, this application has the following beneficial effects: 1. By collecting multi-dimensional sensor data in real time and dynamically adjusting the pressure, the welding pressure and working conditions can be matched in real time, thereby improving the stability of the welding process and avoiding quality fluctuations caused by parameter lag. 2. By using an adaptive control algorithm to dynamically adjust parameters by integrating material properties and environmental data, the problem that traditional fixed parameters cannot adapt to material differences and environmental changes can be solved; a multi-input pressure prediction model is trained based on historical data and real-time sensor data to achieve optimal pressure pre-calculation under different working conditions, reduce manual trial and error costs, and improve process planning efficiency and welding quality consistency. 3. By estimating actuator parameters in real time through extended Kalman filtering algorithm, and combining residual verification and rolling optimization of model predictive control module, dynamic calibration of pressure control is achieved, reducing control error to a low level and improving accuracy; based on spectral recognition and convolutional neural network, automatic switching control mode for material type is realized, and PID parameters are dynamically optimized by reinforcement learning, so that parameter switching during welding of different materials is more timely and the penetration error is controlled within a smaller range. 4. By modeling the two-dimensional features of pressure fluctuations and current spectra using convolutional neural networks, early warning and emergency response to abnormal pressure can be achieved, reducing the scrap rate and equipment damage risk caused by welding anomalies; 5. Based on the four-dimensional health vector and linear regression model, the fault tolerance threshold is dynamically adjusted, and combined with PID parameter self-tuning, to ensure that the pressure control error and overshoot of the equipment during long-term operation are within a small range, thus solving the control problems caused by equipment aging; 6. By coordinating multiple modules including sensor modules, controllers, and actuators, the entire process from data acquisition, model analysis, parameter adjustment to execution control is automated, thereby improving welding efficiency and overall yield. Attached Figure Description

[0025] Figure 1 This is a flowchart of a pressure control method for a friction stir welding system disclosed in an embodiment of this application.

[0026] Figure 2 This is a schematic diagram of a pressure control system for a friction stir welding system disclosed in an embodiment of this application.

[0027] Figure 3 This is a schematic diagram of a device according to one embodiment of this application. Detailed Implementation

[0028] The present application will be further described in detail below with reference to the accompanying drawings.

[0029] This application discloses a pressure control method for a friction stir welding system, including: S1: The sensor module acquires real-time sensor data during the welding process. The sensor module is used to acquire parameters such as material thickness, material hardness, current output value, pressure, temperature and welding speed.

[0030] Specifically, the sensor module includes a pressure sensor, a temperature sensor, a spectral sensor, a thickness sensor, and a hardness sensor.

[0031] Pressure and temperature sensors are installed near the welding area to accurately measure relevant parameters.

[0032] In this embodiment, the spectral sensor is selected from fiber optic spectral sensors, Raman spectral sensors, or ultraviolet-visible spectral sensors according to the application scenario.

[0033] In this embodiment, the thickness sensor is an ultrasonic sensor, which measures the material thickness by emitting ultrasonic waves and receiving the reflected waves. In other embodiments, it can be a laser displacement sensor, which obtains the material thickness by emitting a laser to measure the distance.

[0034] In this embodiment, the hardness sensor is an indentation hardness sensor, which determines the hardness by applying a certain pressure to create an indentation and measuring the indentation size. In other embodiments, a dynamic hardness sensor can be used, which measures the hardness by impacting the material with an impactor. The above sensors collectively collect parameters such as material thickness, material hardness, current output value, pressure, temperature, and welding speed.

[0035] The sensor module continuously collects multi-dimensional data during the welding process, including material thickness, hardness, thermal conductivity, welding speed, rotation frequency, current pressure feedback value, welding zone temperature, and equipment operating status. The sensor module collects 50 to 100 sets of data per second to ensure the capture of instantaneous changes in material properties.

[0036] Furthermore, Kalman filtering is used to filter and denoise the raw sensor data, eliminating outliers caused by electromagnetic interference or mechanical vibration.

[0037] The steps for collecting parameters are as follows: The system monitors the rate of change in material thickness in real time and calculates Δthickness / Δtime through continuous sampling to determine the surface flatness of the workpiece. In addition, it acquires the hardness gradient of the material and identifies the transition zone of the material interface based on ultrasonic sensor data. Furthermore, it calculates the equivalent heat input by using welding speed and rotation frequency to evaluate the influence of the temperature field on pressure.

[0038] The sensor module acquires key information during the welding process and transmits the collected data to the controller, thus providing an accurate data basis for subsequent pressure calculation and adjustment.

[0039] S2: Based on real-time sensor data, the controller uses a preset algorithm module to analyze and process the real-time sensor data to obtain the real-time pressure value.

[0040] Specifically, the controller includes an adaptive control algorithm module, which contains at least welding material characteristic data and environmental change data, used to dynamically adjust control parameters, i.e., the parameters of the welding process execution system. The adaptive control algorithm module contains at least one of a fuzzy control algorithm or a neural network control algorithm.

[0041] The controller automatically activates the corresponding algorithm module based on the real-time data type: In the scenario of fuzzy control algorithm, when the material properties exhibit nonlinear changes, i.e., composite material welding, the continuous data, i.e. the material hardness value, is transformed into fuzzy variables, i.e., high hardness, medium hardness, or low hardness, through the fuzzy control algorithm. The pressure adjustment is calculated based on the preset fuzzy control rules. When the hardness is high and the thickness is large, the pressure increment is +10%; when the hardness is low and the thickness is small, the pressure reduction is -10%. In the context of neural network control algorithms, when welding multiple materials, such as joining dissimilar metals like aluminum alloy and titanium alloy, a pre-trained neural network model is invoked. A backpropagation (BP) neural network is used. Specifically, the number of nodes in the input layer is determined based on the dimension of the input parameters. When there are 5 dimensions of input parameters, such as material type, workpiece thickness, welding speed, current pressure, and heat input, it is set to 5. Two hidden layers are set: the first layer has 8 neurons, which is 1.6 times the input dimension, and the second layer has 6 neurons. The activation function for both layers is ReLU. The output layer has 1 node, which is the pressure value, and linear activation is used.

[0042] Furthermore, the controller is equipped with a multi-modal switching module, which acquires real-time workpiece spectral data from a spectral sensor, calls up a convolutional neural network to identify the material type, and calls up a big data analysis module to acquire real-time data on welding current, speed, and penetration depth as input data. The controller optimizes the control parameters of each mode every fifty cycles through a reinforcement learning algorithm, and simultaneously triggers the control parameters of the corresponding mode.

[0043] Specifically, when the spectral sensor detects a change in material from aluminum alloy to magnesium alloy, the adaptive algorithm module automatically retrieves the control parameters for the corresponding mode from the convolutional neural network that identifies the material type and the big data analysis module, and sends the control parameters for the corresponding mode to the actuator. The steps are as follows: The convolutional neural network is invoked to identify the material type. For example, if the material type is identified as aluminum alloy, the aluminum alloy mode is switched and sent to the actuator. The corresponding control parameters are: proportional gain Kp=1.2, integral gain Ki=0.3, and pressure adjustment step size ΔP=5N. The convolutional neural network is invoked to identify the material type. For example, if the material type is identified as magnesium alloy, the magnesium alloy mode is switched and sent to the actuator. The corresponding control parameters are: Kp=0.8, Ki=0.5, ΔP=3N. Based on the plastic deformation threshold of the material type, the control parameters are adjusted. When the plastic deformation threshold is high, the control parameter values ​​of the actuator are increased; when the plastic deformation threshold is low, the control parameter values ​​of the actuator are decreased. This achieves adaptive adaptation to material properties, reducing the risk of workpiece damage due to excessive pressure.

[0044] Every 50 control cycles, with the control cycle ranging from 15 to 25 ms, the current modal control parameters, the four-dimensional health vector H, and real-time pressure data are collected, and then the control parameters for each modality are optimized.

[0045] S3: The controller sends the real-time pressure value to the actuator. The actuator outputs instructions based on the real-time pressure value to adjust the working pressure value of the friction stir welding system to adapt to the current welding conditions.

[0046] Specifically, the controller also includes a pressure prediction model and an online parameter identification model. It accesses a big data analysis module, uses historical welding data to train a machine learning model, and retrieves real-time sensor data to construct a pressure prediction model. This model predicts the optimal working pressure value of the friction stir welding system under different welding materials, workpiece thicknesses, and ambient temperature conditions. The steps are as follows: At least 1,000 sets of historical welding data were retrieved from the big data analytics module. This data included welding pressure, welding speed, rotation frequency, material type, workpiece thickness and hardness, as well as motor current, pressure sensor output, and ambient temperature. Outliers were removed from the retrieved historical welding data, and random noise was eliminated using a moving average filter.

[0047] Machine learning model training was conducted, using a convolutional neural network to extract features from the material interface transition zone. Data on workpiece thickness, hardness gradient change rate, current heat input distribution, and historical pressure fluctuations were obtained. The data was divided: 80% for training, 10% for validation, and 10% for testing. Using these data as the input layer of the convolutional neural network, and under conditions of heat input deviation ≤ ±2% and hardness gradient fluctuation ≤ ±1.5%, the convolutional neural network output predicted the optimal working pressure value for the next 3 to 5 control cycles with an accuracy of ±1.5%, thus constructing a pressure prediction model. The time range for each control cycle was 15 to 25 ms.

[0048] In addition, the online parameter identification model is embedded into the closed-loop control process of the actuator. The control command signal of the actuator and real-time sensor data are retrieved as input data for the online parameter identification model. The extended Kalman filter algorithm is invoked to obtain the initial value of the offline calibrated motor torque constant and the preset value of sensor zero drift as the initial parameters of the extended Kalman filter algorithm. The real-time execution parameters of the actuator are then estimated in real time using the extended Kalman filter algorithm. The steps are as follows: Specifically, the initial value of the motor torque constant is obtained through no-load testing; under no-load conditions, the average value of the sensor output is obtained to obtain the sensor zero-drift preset value.

[0049] Based on the input data of the online parameter identification model, the real-time execution parameters of the actuator, namely the motor torque constant and the sensor zero drift parameters, are dynamically estimated by the extended Kalman filter algorithm.

[0050] The model predictive control module (MMCC) filters, normalizes, and denoises the obtained real-time execution parameters. These parameters are then transmitted to the MMCC, which performs filtering and normalization to eliminate noise interference. Furthermore, the frequency estimated by the extended Kalman filter algorithm is consistent with the control cycle of the MMCC. Then, the residual verification module is invoked to verify the matching degree between the processed real-time execution parameters and the input requirements of the prediction model. When the parameter deviation is greater than 2%, the model prediction control module triggers an update command to update the real-time execution parameters.

[0051] Using the above scheme, a model relating execution parameters to pressure is established. A correlation matrix between motor torque constant, sensor zero drift, and pressure control accuracy is constructed using offline experimental data. In the model predictive control module, the coefficient matrix of the pressure prediction model is adjusted based on real-time execution parameters, thereby improving the accuracy and stability of pressure prediction.

[0052] In addition, the model predictive control module obtains historical pressure data and real-time operating conditions based on the processed real-time execution parameters, performs rolling optimization prediction of pressure change trends for the next three to five control cycles, calculates and outputs a pre-adjusted working pressure correction value based on the prediction results, and sends the working pressure correction value to the actuator for adjustment, thereby realizing closed-loop pressure control.

[0053] In addition, when the temperature sensor detects that the temperature of the welding area exceeds the preset threshold, in this embodiment, the temperature threshold is set to 300°C. The adaptive algorithm automatically reduces the pressure adjustment rate to prevent overpressure deformation caused by material softening at high temperatures. When the equipment runs for more than 500 hours, the extended Kalman filter algorithm is invoked online to identify and dynamically compensate for the pressure lag based on the motor torque attenuation model. The lead time of the predicted signal ranges from 10 to 20 ms.

[0054] S4: Continuously acquires real-time sensor data and dynamically adjusts the working pressure value based on the welding information fed back by the real-time sensor data to maintain the stability of the welding process.

[0055] Specifically, the steps for dynamically adjusting the welding working pressure based on real-time sensor data are as follows: Welding data, including pressure, temperature, and displacement, is acquired, filtered for noise reduction, and normalized to a standard format. Machine learning algorithms are then used to extract key features such as welding current fluctuations and penetration depth changes, and the current welding status is assessed by comparing it with process standards. When abnormalities such as insufficient penetration or excessive spatter occur, a pressure adjustment is calculated based on a PID controller and a model predictive control module. If the pressure is insufficient, it is gradually increased proportionally, and adjustment commands are sent to the actuator.

[0056] Continuously monitor sensor data to verify the adjustment effect. If the target is not met, repeat the above steps to ensure the stability of the welding process.

[0057] By adopting the above scheme, welding data such as pressure and temperature are collected at high frequency, filtered and normalized, and then machine learning is used to extract key features such as current fluctuations and penetration changes. The working pressure is dynamically adjusted by combining PID and model predictive control algorithms, and the pressure deviation is gradually corrected proportionally. This ensures that the welding state is consistent with the process standard and improves the real-time response accuracy of the welding process.

[0058] Furthermore, the controller also includes an anomaly prediction model, which achieves pressure anomaly early warning and control through the following steps: The convolutional neural network module is invoked to perform two-dimensional feature modeling on the pressure fluctuation sequence and current signal spectrum in historical welding data, and abnormal events are labeled to train and obtain an anomaly prediction model. The anomaly prediction model performs rolling predictions on real-time pressure and current data. When the model predicts an anomaly probability exceeding the warning threshold, in this embodiment, the warning threshold is when the anomaly probability is greater than 85% and the pressure value reaches 1.1 times the rated value, a warning signal is immediately output.

[0059] When abnormal pressure is confirmed, the controller executes a fast backoff algorithm, reducing the actuator's downward pressure, welding speed, and welding rotation speed, while simultaneously adjusting process parameters. Specifically, the PID controller fine-tunes parameters, reducing the downward pressure displacement to 0.7d, the welding speed to 0.8v, and the rotation speed to 0.8ω, achieving a smooth parameter transition. If the abnormality persists, the parameters are further reduced to 0.5d, 0.6v, and 0.7ω. Simultaneously, a real-time data verification mechanism is introduced, combining real-time sensor data to cross-confirm the abnormality and avoid misjudgments.

[0060] The anomaly prediction model, through rolling predictions of real-time pressure and current data, can output early warning signals and trigger the controller to adjust execution parameters, effectively reducing the scrap rate caused by welding anomalies and minimizing the risk of equipment damage. When abnormal pressure is confirmed, the controller executes a fast rollback algorithm and synchronously adjusts process parameters, reducing the pressure, welding speed, and welding rotation speed, further ensuring the stability and safety of the welding process.

[0061] Furthermore, the controller integrates a device health status assessment model. By using data obtained from the online parameter identification model, it automatically adjusts the controller's fault tolerance threshold and constructs the device health status assessment model.

[0062] Specifically, the dynamic fault tolerance threshold adjustment is achieved through the following steps: Specifically, dynamic fault tolerance threshold adjustment is achieved through the following steps: Obtain the motor torque constant drift rate and sensor zero drift change output from the online parameter identification model, as well as the pressure fluctuation standard deviation and current ripple coefficient from real-time sensor data, denoted as ∆Kt Kt0, ∆b, σp, and γi, respectively, and construct a four-dimensional health vector H: H = [∆Kt Kt0, ∆b, σp, γi]; Establish a mapping relationship between health characteristics and fault tolerance thresholds based on historical data, and dynamically generate an adaptive threshold Tnew using a linear regression model: Tnew = T0 + α × ∆Kt Kt0 + β × ∆b + δ × σp + ϵ × γi, where T0 is the initial threshold, and α, β, δ, and ϵ are weight coefficients, determined through least squares training; and update the model parameters based on the latest 1000 sets of sampled data after each 50-hour run.

[0063] When H triggers the threshold adjustment condition for three consecutive sampling cycles, the controller parameter self-tuning is initiated synchronously to ensure that the pressure control error is ≤3%, the step response overshoot is ≤3%, and the settling time is ≤150ms. This ensures that within 1000 hours of continuous operation, the rated pressure error is ≤±1.5%, the step response overshoot is ≤5%, and the settling time is ≤200ms, achieving coordinated optimization of health status and control parameters.

[0064] Using the above scheme, a four-dimensional health vector is constructed by combining the motor torque constant drift rate, sensor zero drift change, and pressure fluctuation standard deviation and current ripple coefficient output by the online parameter identification model. Based on historical data, a linear regression model is used to establish the relationship between health characteristics and fault tolerance threshold, automatically adjust the controller fault tolerance threshold, and then start controller parameter self-tuning. This ensures that the pressure control error and step response overshoot are maintained at a low level after the equipment has been running continuously for a long time, with a short adjustment time, and solves the problem that traditional fixed thresholds cannot cope with equipment aging.

[0065] Secondly, embodiments of this application disclose a pressure control system for a friction stir welding system, comprising: The sensor module includes a spectral sensor, a pressure sensor, a current sensor, and a temperature sensor, used to acquire workpiece spectral data, welding pressure, drive current, and ambient temperature in real time. The actuator is connected to the controller, receives pressure control signals, and realizes automatic adjustment of working pressure and rapid retraction in case of abnormality. The big data analysis module stores historical welding data and uses machine learning algorithms to train a pressure prediction model. The controller has multiple built-in modules: The adaptive control algorithm module is used to dynamically adjust control parameters; A pressure prediction model is used to achieve adaptive control of the friction stir welding system. The multimodal switching module uses a convolutional neural network to identify the material type of spectral data and triggers the control parameters of the corresponding mode. An online parameter identification model is used to accurately estimate execution parameters in real time, dynamically optimize control, and achieve precise pressure control. The model predictive control module is used to achieve precise pressure control; The multi-modal switching module is used to identify material type in real time, automatically switch control parameters, dynamically optimize PID parameters, and improve welding quality stability. Anomaly prediction models are used to provide early warnings of welding anomalies and pressure anomalies, and to adjust parameters accordingly. The equipment health status assessment model is used to dynamically adjust the fault tolerance threshold and maintain control accuracy. The entire process, from data acquisition and model analysis to parameter adjustment and execution control, is automated, improving welding efficiency and overall yield.

[0066] Thirdly, this embodiment also discloses a computer electronic device. A computer electronic device includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as in any of the methods described above.

[0067] The electronic device can be a desktop computer, laptop computer, or cloud server, and includes, but is not limited to, a processor and memory. For example, the electronic device may also include input / output devices, network access devices, and buses. The processor in this application may include one or more processing cores. The processor executes instructions, programs, code sets, or instruction sets stored in memory, calls data stored in memory, and performs various functions and processes data in this application. The processor can be at least one of the following: Application-Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), controller, microcontroller, and microprocessor. It is understood that for different devices, the electronic device used to implement the above-described processor functions can be other types, and this application does not specifically limit the embodiments.

[0068] The memory can be an internal storage unit of an electronic device, such as a hard disk or RAM, or an external storage device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD), or flash memory card (FC) equipped on the electronic device. Furthermore, the memory can be a combination of an internal storage unit and an external storage device. The memory is used to store computer programs and other programs and data required by the electronic device. The memory can also be used to temporarily store data that has been output or will be output. This application does not limit this.

[0069] Fourthly, this embodiment also discloses a computer-readable storage medium. A computer-readable storage medium stores a computer program that can be loaded by a processor and executed as in any of the methods described above.

[0070] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0071] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete the above description.

Claims

1. A pressure control method for a friction stir welding system, characterized in that, Includes the following steps: Acquire real-time sensor data during the welding process, wherein the real-time sensor data includes at least material thickness, material hardness, current output value, pressure, temperature and welding speed parameters; Based on real-time sensor data, the controller uses a preset algorithm module to analyze and process the real-time sensor data to obtain the real-time pressure value. The controller sends the real-time pressure value to the actuator, which then outputs instructions to adjust the working pressure value of the friction stir welding system according to the real-time pressure value to adapt to the current welding conditions. Continuously acquire real-time sensor data, and dynamically adjust the working pressure value based on the welding information fed back by the real-time sensor data to maintain the stability of the welding process.

2. The method according to claim 1, characterized in that, The controller includes an adaptive control algorithm module, which contains at least welding material characteristic data and environmental change data, and is used to dynamically adjust control parameters. The adaptive control algorithm module includes at least one of the following: a fuzzy control algorithm or a neural network control algorithm.

3. The method according to claim 1, characterized in that, The controller also includes a pressure prediction model, and constructing the pressure prediction model and predicting working pressure values ​​includes: The big data analysis module is accessed to train a machine learning model using historical welding data. Real-time sensor data is also accessed to build a pressure prediction model, which predicts the optimal working pressure value under different welding materials, workpiece thicknesses, and ambient temperature conditions.

4. The method according to claim 2, characterized in that, The controller includes an online parameter identification model, and the steps for real-time estimation of the actuator's execution parameters and pressure closed-loop control are as follows: The online parameter identification model is embedded into the closed-loop control loop of the actuator, and the control command signal of the actuator and real-time sensor data are used as inputs to the online parameter identification model. The extended Kalman filter algorithm is retrieved to estimate the execution parameters in real time. The initial values ​​of the motor torque constant and the sensor zero drift preset values, which are obtained offline, are used as the initial parameters of the extended Kalman filter algorithm to construct the system state equation and observation equation. Based on the input of the online parameter identification model, the extended Kalman filter algorithm is used to dynamically estimate the real-time execution parameters; The model predictive control module is invoked to process real-time execution parameters. The estimated real-time execution parameters are transmitted to the model predictive control module. Noise interference is eliminated through moving average filtering and normalization. The frequency estimated by the extended Kalman filter algorithm is consistent with the control cycle of the model predictive control module. The residual verification module is invoked to calculate the parameter deviation. When the deviation between the measured value and the input requirement of the prediction model is too large, a parameter update command is triggered. The model predictive control module performs rolling optimization prediction of pressure change trends for the next 3 to 5 control cycles based on processed real-time execution parameters, historical pressure data, and real-time operating conditions. It calculates and outputs working pressure correction values, drives the actuator to adjust the pressure, and achieves closed-loop control.

5. The method according to claim 1, characterized in that, The controller is equipped with a multi-modal switching module, which acquires real-time workpiece spectral data from a spectral sensor, retrieves convolutional neural network to identify material type, and triggers control parameters for the corresponding mode. The big data analysis module is invoked to obtain real-time data on welding current, speed, and penetration depth as input data. The control parameters of each mode are optimized every fifty cycles using a reinforcement learning algorithm.

6. The method according to claim 1, characterized in that, The controller also includes an anomaly prediction model, which achieves pressure anomaly early warning and control through the following steps: The convolutional neural network module is invoked to perform two-dimensional feature modeling on the pressure fluctuation sequence and current signal spectrum in historical welding data, and abnormal events are labeled to train and obtain an anomaly prediction model. The anomaly prediction model performs rolling predictions on real-time pressure and current data. When the predicted anomaly probability is greater than the warning threshold and the pressure value reaches the rated value, a warning signal is immediately output to trigger the controller to adjust the execution parameters. When abnormal pressure is confirmed, the controller executes a fast rollback algorithm and synchronously adjusts process parameters to reduce the pressure, welding speed, and welding rotation speed.

7. The method according to claim 1, characterized in that, The controller also includes a device health status assessment model. The controller's fault tolerance threshold is automatically adjusted based on the data obtained from the online parameter identification model to construct the device health status assessment model. Specifically, the dynamic fault tolerance threshold adjustment is achieved through the following steps: The drift rate of the motor torque constant and the change in zero drift of the sensor are obtained from the output of the online parameter identification model, and the standard deviation of pressure fluctuation and current ripple coefficient are obtained from the real-time sensor data. A four-dimensional health vector H is constructed using Δb, σp, and γi: H = [ΔKt Kt0, Δb, σp, γi]. A mapping relationship between health characteristics and tolerance thresholds is established based on historical data, using a linear regression model: Tnew = T0 + α × ΔKt Kt0 + β × Δb + δ × σp + ϵ × γi, where T0 is the initial threshold, and α, β, δ, and ϵ are weighting coefficients. When the health vector H triggers threshold adjustment, controller parameter self-tuning is initiated synchronously, with pressure control error and step response overshoot ≤5% and adjustment time ≤200ms.

8. A pressure control system for a friction stir welding system, characterized in that, include: The sensor module includes a spectral sensor, a pressure sensor, a current sensor, and a temperature sensor, used to acquire workpiece spectral data, welding pressure, drive current, and ambient temperature in real time. The actuator is connected to the controller, receives pressure control signals, and realizes automatic adjustment of working pressure and rapid retraction in case of abnormality. The big data analysis module stores historical welding data and uses machine learning algorithms to train a pressure prediction model. The controller has multiple built-in modules: The adaptive control algorithm module is used to dynamically adjust control parameters; A pressure prediction model is used to achieve adaptive control of the friction stir welding system. An online parameter identification model is used to accurately estimate execution parameters in real time, dynamically optimize control, and achieve precise pressure control. The model predictive control module is used to achieve precise pressure control; The multi-modal switching module is used to identify material type in real time, automatically switch control parameters, dynamically optimize PID parameters, and improve welding quality stability. Anomaly prediction models are used to provide early warnings of welding anomalies and pressure anomalies, and to adjust parameters accordingly. Equipment health status assessment model is used to dynamically adjust fault tolerance thresholds and maintain control accuracy.

9. A computer electronic device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images