Low-stress anti-fatigue MIG welding power supply self-networking type unified control system

Abstract

The application provides a low-stress anti-fatigue MIG welding power supply self-networking type unified regulation system, which comprises an upper computer, a wireless communication module, a man-machine interaction module and a plurality of welding machines; each welding machine comprises an MCU, a storage module, a power supply module, a sampling module and a driving module; the MCU is embedded with a neural network model for establishing a welding parameter nonlinear mapping; the storage module stores a unified database, the unified database is generated by the neural network model, and the optimal welding parameter combination covering the full current interval is covered; when the parameter setting mode is set to automatic in the parameter mode setting unit, the user can input the macroscopic parameter in the automatic parameter setting mode input unit, and then the microscopic waveform parameter is automatically matched from the unified database. The system can solve the technical problems of welding waveform parameter debugging difficulty and nonlinear relationship difficult to model, automatically deduce the optimal parameter combination of the full current interval, and greatly reduce the debugging workload.

Description

Technical Field

[0001] This invention relates to the field of MIG welding equipment technology, and more specifically, to a low-stress, fatigue-resistant MIG welding power source self-organizing network unified control system. Background Technology

[0002] With the development of SiC power devices, MIG welders are now able to output composite waveforms consisting of low-frequency and high-frequency pulse currents, significantly improving the residual stress distribution and fatigue resistance of welded joints. However, the application of composite waveforms presents the following technical challenges:

[0003] (i) Parameter surge: The current parameters that need to be adjusted are increasing exponentially, resulting in a huge workload;

[0004] (ii) Nonlinear coupling: There are complex nonlinear relationships between the parameters, which are difficult to characterize using traditional empirical formulas;

[0005] (iii) Safety hazards: High-frequency current exacerbates the parasitic inductance effect of cables, welding machines must be installed near the work area, and there are safety risks to the operator's on-site observation panel.

[0006] Existing single-element welding machines mostly rely on fixed empirical formulas or simple table lookups, which are insufficient to cover the optimal parameter combinations across the entire current range and lack self-learning and remote monitoring capabilities. Therefore, there is an urgent need for a single-element MIG welding control system with automatic parameter optimization and remote monitoring functions. This system aims to achieve low residual stress and strong fatigue resistance in welded joints, autonomously deriving optimal parameters across the entire current range from limited sample data, thereby improving the intelligence and safety of MIG welding. Summary of the Invention

[0007] To overcome the shortcomings and deficiencies in the existing technology, the purpose of this invention is to provide a low-stress, fatigue-resistant MIG welding power supply self-organizing network unified control system. This system can solve the technical problems of difficult welding waveform parameter debugging and difficulty in modeling nonlinear relationships, automatically deduce the optimal parameter combination across the entire current range, and significantly reduce the debugging workload.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solution: a low-stress fatigue-resistant MIG welding power supply self-organizing network unified control system, including a host computer, a wireless communication module, a human-machine interaction module, and several welding machines; the wireless communication module adopts a self-organizing network module based on the Zigbee protocol, which is used to realize remote monitoring and start / stop control between the host computer and multiple welding machines.

[0009] Each welding machine includes an MCU, a storage module, a power supply module, a sampling module, and a driver module;

[0010] The MCU has an embedded neural network model for establishing a nonlinear mapping of welding parameters; the neural network model refers to a trained neural network model.

[0011] The storage module stores a unified database, which is generated by the neural network model and covers the optimal combination of welding parameters across the entire current range.

[0012] The human-computer interaction module includes a parameter mode setting unit for setting the parameter setting method to manual or automatic, and various parameter setting method input units. When the parameter setting method is set to automatic in the parameter mode setting unit, the user can input macroscopic parameters in the automatic parameter setting method input unit, and then automatically match microscopic waveform parameters from the unified database. When the parameter setting method is set to manual in the parameter mode setting unit, the user can customize all parameters in the manual parameter setting method input unit and choose whether to send the parameter combination to the training sample database of the neural network model.

[0013] When the training sample database receives a new combination of parameters, the neural network model updates the weights, thereby achieving dynamic optimization of the unified database.

[0014] Preferably, the macroscopic parameters include base material, wire diameter, average current, and current mode; the microscopic waveform parameters include wire feed speed, low-frequency duty cycle, low-frequency frequency, low-frequency base value, low-frequency peak value, high-frequency frequency, high-frequency base value, and high-frequency peak value.

[0015] Preferably, the neural network model includes an input layer, a hidden layer, and an output layer;

[0016] Input layer parameters include base material, wire diameter, and average current;

[0017] Output layer parameters include wire feed speed, low frequency duty cycle, low frequency, low frequency base value, low frequency peak value, high frequency, high frequency base value, and high frequency peak value;

[0018] The hidden layer adopts a single-layer structure, consisting of four neurons, with ReLU as the activation function, mean squared error as the loss function, and gradient descent as the optimization algorithm.

[0019] Preferably, the calculation process of the neural network model is as follows:

[0020] S1. Perform a weighted summation on the input layer data to obtain the total input value for each neuron in the hidden layer. n=1,2,3,4:

[0021] ;

[0022] in, This is the deviation term; i=1,2,3 represents the various parameters of the input layer: base material, wire diameter and average current, where the base material uses unique thermal coding; This represents the weight value passed from the i-th parameter of the input layer to the n-th neuron in the hidden layer;

[0023] S2. Calculate the output value of each neuron in the hidden layer using the ReLU activation function. :

[0024] ;

[0025] S3. Perform a weighted summation of the output values ​​of the hidden layer to obtain the input values ​​of the output layer neurons. :

[0026] ;

[0027] in, For deviation terms, This represents the weight value passed from the nth neuron in the hidden layer to the mth neuron in the output layer;

[0028] S4. Calculate the output value of the output layer neurons using the ReLU activation function. :

[0029] ;

[0030] The current mode includes a low-frequency mode and a composite mode. When the current mode is set to low-frequency mode, only the wire feed speed, low-frequency duty cycle, low-frequency frequency, low-frequency base value, and low-frequency peak value are calculated for the output layer parameters. When the current mode is set to composite mode, all output layer parameters are calculated.

[0031] Preferably, when the parameter setting mode is set to manual in the parameter mode setting unit, the condition for selecting to send the parameter combination to the training sample database of the neural network model is:

[0032] After welding was completed using this parameter combination, the weld passed the fatigue test and residual stress test results.

[0033] Determine if a parameter combination with the same macroscopic parameter value exists in the training sample database. If not, send the parameter combination to the training sample database. If it exists, determine if the fatigue test and residual stress detection results of the current parameter combination are better. If so, send the parameter combination to the training sample database and overwrite the original parameter combination with the same macroscopic parameter value.

[0034] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0035] 1. High level of intelligence: Through neural network model, the system can automatically deduce the optimal parameter combination for the entire current range from a limited sample, greatly reducing the amount of debugging work;

[0036] 2. High security: Wireless self-organizing network technology allows operators to remotely monitor multiple welding machines in a safe area, avoiding the safety hazards of high-frequency welding;

[0037] 3. Flexible operation: It adopts a dual-mode human-computer interaction with manual and automatic parameter setting, which takes into account both the ease of operation and the flexibility of process development.

[0038] 4. Continuous evolution capability: The data overwrite mechanism enables the system to continuously absorb new process experience and optimize the database. Attached Figure Description

[0039] Figure 1 This is an overall block diagram of the low-stress, fatigue-resistant MIG welding power source self-organizing network unified control system of the present invention.

[0040] Figure 2 This is a block diagram of the power supply module of the self-organizing network unified control system for low-stress fatigue-resistant MIG welding power supply of the present invention.

[0041] Figure 3 This invention relates to the current mode of the self-organizing network unified control system for low-stress, fatigue-resistant MIG welding power supply.

[0042] Figure 4 This is a welding control flowchart of the low-stress, fatigue-resistant MIG welding power source self-organizing network unified control system of the present invention.

[0043] Figure 5 This is a schematic diagram of the neural network model of the self-organizing network type unified control system for low-stress fatigue-resistant MIG welding power supply of the present invention. Detailed Implementation

[0044] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0045] Example

[0046] This embodiment presents a low-stress, fatigue-resistant MIG welding power supply self-organizing network type unified control system, such as... Figure 1 As shown, it includes a host computer, a wireless communication module, a human-computer interaction module, and several welding machines.

[0047] Each welding machine includes an MCU, a storage module, a power supply module, a sampling module, a drive module, and an external control module.

[0048] The MCU can use existing chips, such as the STM32F405RGT6 chip, with embedded FreeRTOS to control the functions of each module; and it also has an embedded neural network model for establishing nonlinear mapping of welding parameters; the neural network model refers to a trained neural network model.

[0049] The storage module uses existing chips, such as the AT24C32 EEPROM chip, which is connected to the IIC communication port of the MCU. The MCU stores the unified database in the EEPROM chip in the form of a multidimensional array through IIC communication. The unified database is generated by the neural network model and covers the optimal welding parameter combination for the entire current range.

[0050] The power supply module includes 380V industrial AC power and a voltage regulator circuit, such as... Figure 2 As shown; the voltage regulator circuit is used to convert 380V industrial AC power into various graded voltages to power other modules; the power supply grades include 24VDC, ±15VDC, 5VDC and 3.3VDC; the voltage regulator circuit can provide isolation between important grades, limit noise to the local area and prevent it from propagating to other modules through the shared power path.

[0051] The sampling module is used to sample high-frequency pulse current, low-frequency pulse current, and welding voltage, and is connected to the ADC channel I / O port of the MCU respectively; the driving module is used to drive the full-bridge inverse converter, high-frequency chopping, and wire feed of the welding power supply, and is connected to the PWM drive I / O port of the MCU respectively; the external control module is used to control the gas valve and the robot. The gas valve is used for protective gas control, and the robot is used for transferring the welding machine. The MCU sends switching signals to control the corresponding I / O ports.

[0052] The wireless communication module adopts a self-organizing network module based on the Zigbee protocol to realize remote monitoring and start / stop control between the host computer and multiple welding machines. Remote monitoring includes monitoring of status, average current, voltage, and wire feed speed; the status refers to the current state of the welding machine, including stopped, no-load, and welding; the start / stop control is used to start or stop the welding machine output. Specifically, the host computer can remotely monitor the welding machine status, average current, voltage, and wire feed speed through a communication protocol, and can start and stop the welding machine via a button. Communication between the host computer and the welding machine is achieved using a wireless communication module. This module is a DL-LN33 wireless self-organizing network module based on the Zigbee protocol, connected as follows: DL-LN33 wireless self-organizing network modules that need to communicate with each other are configured to share the same channel and network ID. The host computer connects to the DL-LN33 wireless self-organizing network module via CP2102, and the control system connects to another DL-LN33 wireless self-organizing network module via a serial port. If multiple welding machines need to be monitored, the control system of each welding machine must be connected to a separate DL-LN33 wireless self-organizing network module.

[0053] The human-computer interaction module uses a display screen based on the RS485 communication protocol, including a parameter mode setting unit for setting the parameter setting mode to manual or automatic, and various parameter setting mode input units.

[0054] The human-computer interaction module may also include a PID setting unit, which can set the PID parameters for low-frequency pulse current and high-frequency pulse current respectively; the PID parameters are used to adjust the overshoot, settling time and response speed of the current waveform.

[0055] The welding machine's current modes include low-frequency mode and composite mode; low-frequency mode outputs only low-frequency pulse current; composite mode outputs a superposition of low-frequency and high-frequency pulse currents, such as... Figure 3 As shown.

[0056] The welding control process of the system of the present invention is as follows: Figure 4 As shown:

[0057] When the parameter setting mode is set to automatic in the parameter mode setting unit, the user can input macroscopic parameters in the automatic parameter setting mode input unit. Macroscopic parameters include base material, wire diameter, average current, and current mode. Then, the MCU automatically matches microscopic waveform parameters from a unified database via 485 communication. Microscopic waveform parameters include wire feed speed, low-frequency duty cycle, low-frequency frequency, low-frequency base value, low-frequency peak value, high-frequency frequency, high-frequency base value, and high-frequency peak value. The welding operation is then performed using this parameter combination.

[0058] When the parameter setting mode is set to manual in the parameter mode setting unit, users can customize all parameters in the manual parameter setting mode input unit, including the aforementioned macroscopic parameters and microscopic waveform parameters. This parameter combination is then used for welding operations, and the user can choose whether to send this parameter combination to the training sample database of the neural network model. Specifically, the conditions for choosing to send this parameter combination to the training sample database of the neural network model are:

[0059] After welding was completed using this parameter combination, the weld passed the fatigue test and residual stress test results.

[0060] The system determines whether a parameter combination with the same macroscopic parameter values ​​exists in the training sample database. If not, the parameter combination is sent to the training sample database. If it exists, the system determines whether the fatigue test and residual stress test results of the current parameter combination are superior. If so, the parameter combination is sent to the training sample database, overwriting the original parameter combination with the same macroscopic parameter values. Otherwise, the parameter combination is not sent to the training sample database. If it is only a process adjustment stage, the parameter combination can also be used only for welding operations without being sent to the training sample database.

[0061] When the training sample database receives a new combination of parameters, the neural network model updates the weights, thereby achieving dynamic optimization of the unified database.

[0062] like Figure 5 As shown, the neural network model includes an input layer, a hidden layer, and an output layer;

[0063] Input layer parameters include base material, wire diameter, and average current;

[0064] Output layer parameters include wire feed speed, low frequency duty cycle, low frequency, low frequency base value, low frequency peak value, high frequency, high frequency base value, and high frequency peak value;

[0065] The hidden layer adopts a single-layer structure, consisting of four neurons, with ReLU as the activation function, mean squared error as the loss function, and gradient descent as the optimization algorithm.

[0066] The calculation process of the neural network model is as follows:

[0067] S1. Perform a weighted summation on the input layer data to obtain the total input value for each neuron in the hidden layer. n=1,2,3,4:

[0068] ;

[0069] in, This is the deviation term; i=1,2,3 represents the various parameters of the input layer: base material, welding wire diameter and average current. The base material uses unique thermal codes (0 0 1), (0 1 0), and (1 0 0) to replace Q series, stainless steel and titanium alloy materials as input values, respectively. This represents the weight value passed from the i-th parameter of the input layer to the n-th neuron in the hidden layer;

[0070] S2. Calculate the output value of each neuron in the hidden layer using the ReLU activation function. :

[0071] ;

[0072] S3. Perform a weighted summation of the output values ​​of the hidden layer to obtain the input values ​​of the output layer neurons. :

[0073] ;

[0074] in, For deviation terms, This represents the weight value passed from the nth neuron in the hidden layer to the mth neuron in the output layer;

[0075] S4. Calculate the output value of the output layer neurons using the ReLU activation function. :

[0076] ;

[0077] The current modes include a low-frequency mode and a composite mode; the low-frequency mode outputs only low-frequency pulse current; the composite mode outputs both low-frequency and high-frequency pulse currents. When the current mode is set to low-frequency mode, the output layer parameters only include wire feed speed, low-frequency duty cycle, low-frequency frequency, low-frequency base value, and low-frequency peak value. When the current mode is set to composite mode, all output layer parameters are calculated.

[0078] The neural network model training uses the mean squared error as the loss function L:

[0079] ;

[0080] in, This represents the fitted value of each neuron in the output layer calculated using discrete points. This represents the true value of the discrete point;

[0081] Calculate the partial derivatives of the loss function with respect to the weights and biases using gradient descent:

[0082] ;

[0083] in, The learning rate is adjusted based on the actual data until the fitted value is close to the true value, which means the fitting of the current data is complete.

[0084] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A low-stress, fatigue-resistant MIG welding power supply self-organizing network type unified control system, characterized in that: It includes a host computer, a wireless communication module, a human-machine interaction module, and several welding machines; the wireless communication module adopts a self-organizing network module based on the Zigbee protocol, which is used to realize remote monitoring and start / stop control between the host computer and multiple welding machines; Each welding machine includes an MCU, a storage module, a power supply module, a sampling module, and a driver module; The MCU has an embedded neural network model for establishing a nonlinear mapping of welding parameters; the neural network model refers to a trained neural network model. The storage module stores a unified database, which is generated by the neural network model and covers the optimal combination of welding parameters across the entire current range. The human-computer interaction module includes a parameter mode setting unit for setting the parameter setting method to manual or automatic, and various parameter setting method input units. When the parameter setting method is set to automatic in the parameter mode setting unit, the user can input macroscopic parameters in the automatic parameter setting method input unit, and then automatically match microscopic waveform parameters from the unified database. When the parameter setting method is set to manual in the parameter mode setting unit, the user can customize all parameters in the manual parameter setting method input unit and choose whether to send the parameter combination to the training sample database of the neural network model. When the training sample database receives a new combination of parameters, the neural network model updates the weights, thereby achieving dynamic optimization of the unified database.

2. The low-stress, fatigue-resistant MIG welding power supply self-organizing network unified control system according to claim 1, characterized in that: The macroscopic parameters include base material, wire diameter, average current, and current mode; the microscopic waveform parameters include wire feed speed, low-frequency duty cycle, low-frequency frequency, low-frequency base value, low-frequency peak value, high-frequency frequency, high-frequency base value, and high-frequency peak value.

3. The low-stress, fatigue-resistant MIG welding power supply self-organizing network unified control system according to claim 2, characterized in that: The neural network model includes an input layer, a hidden layer, and an output layer; Input layer parameters include base material, wire diameter, and average current; Output layer parameters include wire feed speed, low frequency duty cycle, low frequency frequency, low frequency base value, low frequency peak value, high frequency frequency, high frequency base value, and high frequency peak value; The hidden layer adopts a single-layer structure, consisting of four neurons, with ReLU as the activation function, mean squared error as the loss function, and gradient descent as the optimization algorithm.

4. The low-stress, fatigue-resistant MIG welding power supply self-organizing network unified control system according to claim 3, characterized in that: The calculation process of the neural network model is as follows: S1. Perform a weighted summation on the input layer data to obtain the total input value for each neuron in the hidden layer. n=1,2,3,4: ； in, This is the deviation term; i=1,2,3 represents the various parameters of the input layer: base material, wire diameter and average current, where the base material uses unique thermal coding; This represents the weight value passed from the i-th parameter of the input layer to the n-th neuron in the hidden layer; S2. Calculate the output value of each neuron in the hidden layer using the ReLU activation function. : ； S3. Perform a weighted summation of the output values ​​of the hidden layer to obtain the input values ​​of the output layer neurons. : ； in, For deviation terms, This represents the weight value passed from the nth neuron in the hidden layer to the mth neuron in the output layer; S4. Calculate the output value of the output layer neurons using the ReLU activation function. : ； The current mode includes a low-frequency mode and a composite mode. When the current mode is set to low-frequency mode, only the wire feed speed, low-frequency duty cycle, low-frequency frequency, low-frequency base value, and low-frequency peak value are calculated for the output layer parameters. When the current mode is set to composite mode, all output layer parameters are calculated.

5. The low-stress, fatigue-resistant MIG welding power supply self-organizing network unified control system according to claim 1, characterized in that: When the parameter setting mode is set to manual in the parameter mode setting unit, the condition for selecting to send this parameter combination to the training sample database of the neural network model is: After welding was completed using this parameter combination, the weld passed the fatigue test and residual stress test results. Determine if a parameter combination with the same macroscopic parameter value exists in the training sample database. If not, send the parameter combination to the training sample database. If it exists, determine if the fatigue test and residual stress detection results of the current parameter combination are better. If so, send the parameter combination to the training sample database and overwrite the original parameter combination with the same macroscopic parameter value.

Images