Welding gun parameter full function programming and display system based on touch screen and menu navigation

Abstract

The application relates to the technical field of welding gun parameter setting, and discloses a welding gun parameter full-function programming and display system based on a touch screen and menu navigation, which integrates a touch and a rocker module to perform multi-level menu navigation, combines a parameter full-function programming unit, uses a prediction model to output a basic prediction value based on a current welding process, generates a parameter recommended value in combination with an offset value of a corresponding working condition domain, and after user confirmation, uses a prediction residual to perform exponential smoothing closed-loop updating on the offset value, and through a display control unit, standby or welding states are automatically switched to display modes and error operation locking is provided. The application can realize full-function parameter programming of a welding gun end, improve the continuity of field operation, and in combination with a parameter intelligent prediction module, recommended values that match user operation habits are given, so that a welder can determine welding parameters by only one-key confirmation or rocker fine adjustment, thereby reducing the frequency of parameter wide-range adjustment, and improving the convenience and operation efficiency of parameter setting.

Description

Technical Field

[0001] This invention relates to the field of welding torch parameter setting technology, specifically to a full-function programming and display system for welding torch parameters based on a touch screen and menu navigation. Background Technology

[0002] Welding, as a key fundamental process in modern manufacturing, is widely used in heavy equipment manufacturing, shipbuilding, bridge steel structures, and automobile manufacturing. The precise setting and real-time monitoring of welding parameters have a decisive impact on weld formation quality, penetration control, and mechanical properties. Welding parameters typically include various process variables such as welding current, welding voltage, wire feed speed, welding speed, shielding gas flow rate, pulse frequency, and duty cycle. These parameters have complex coupling constraints and require targeted adjustments based on operating conditions such as base material, plate thickness, welding wire diameter, and welding position. To adapt to the flexible production needs of multi-variety, small-batch production, welding operators need to frequently switch and fine-tune these parameters at the welding station. Therefore, high demands are placed on the convenience, functional completeness, and response efficiency of parameter setting devices.

[0003] Currently, the setting and display of welding parameters mainly rely on the following technical approaches: First, the integrated operation method on the welding machine panel, using encoder knobs, physical buttons, and digital tubes or segmented LCD screens to select and adjust parameters. This method typically has a complex menu hierarchy and a long operation path. Second, the independent handheld terminal control method, using a PAD-type programmer connected to the welding machine via wired or wireless connection. This method offers relatively rich parameter setting functions and a graphical interface, requiring the welder to alternate between welding operations and terminal operation. Third, the simplified display and adjustment method at the welding torch end, such as integrating a small display screen and a few adjustment buttons or encoders on the welding torch housing, can display and adjust some commonly used welding parameters. These technical approaches have achieved certain results in engineering applications, laying the foundation for the digital management of welding parameters.

[0004] However, existing technical solutions are still insufficient when dealing with complex working conditions in heavy equipment manufacturing sites, such as frequent switching between multiple processes, welders needing to wear heavy protective gloves, and exposure to strong arc light interference. On the one hand, both welding machine panel operation and independent handheld terminal operation require welders to keep their attention and at least one hand off the welding torch, causing intermittent interruptions between welding operations and parameter settings. This is especially problematic in scenarios requiring repeated trial and error and parameter-by-parameter adjustments, significantly impacting operational continuity. On the other hand, existing integrated welding torch solutions still have room for improvement in the comprehensiveness of parameter setting functions. Their interaction mainly relies on welders manually adjusting from default or current values. When significant parameter changes are needed, the joystick or encoder has a long travel distance, making the adjustment process less convenient. These factors, to some extent, restrict the overall efficiency and user experience of welding operations. Therefore, a parameter setting and display system capable of full-function parameter programming at the welding torch end and possessing intelligent guidance capabilities is needed. Summary of the Invention

[0005] To address the problems in related technologies, this invention provides a full-function programming and display system for welding torch parameters based on a touch screen and menu navigation, thereby overcoming the aforementioned technical problems in existing related technologies.

[0006] To solve the aforementioned technical problem, the present invention is achieved through the following technical solution:

[0007] In a first aspect, embodiments of the present invention provide a full-function programming and display system for welding torch parameters based on a touch screen and menu navigation, specifically including:

[0008] The touch display module is used to display welding parameter information and receive touch operation input;

[0009] The joystick navigation module provides auxiliary operation input for multi-directional joystick and button operation;

[0010] The menu navigation control unit is used to map touch operations and joystick operations into navigation and selection operations of a multi-level menu system, and adaptively adjust the menu interface according to the input mode.

[0011] The full-function parameter programming unit is used to perform welding parameter editing, verification, intelligent prediction, and offset compensation correction.

[0012] The parameter grouping storage unit is used to group, store, manage, and retrieve welding parameters according to the process plan.

[0013] The display control unit is used to switch the display mode according to the working status of the welding torch and to present welding parameters and status information in real time.

[0014] The communication module is used for data communication and power supply multiplexing with external devices.

[0015] As a preferred embodiment of the full-function programming and display system for welding torch parameters based on touch screen and menu navigation described in this invention, the communication module adopts a two-wire carrier communication and power supply multiplexing architecture to connect with external devices.

[0016] The communication module has a built-in channel quality assessment submodule, which is used to detect the signal-to-noise ratio of the current operating frequency band in the idle time slot, and automatically switch to the backup carrier frequency when the signal-to-noise ratio is lower than a preset threshold, so as to realize adaptive frequency hopping carrier modulation.

[0017] The communication module introduces a signal adaptive compensation mechanism at the receiving end, which inversely calculates the equivalent transmission length of the welding cable by analyzing the received amplitude attenuation of the synchronization preamble sequence, and dynamically updates the weight coefficients of the digital compensation filter to perform pre-equalization processing on the received modulated signal.

[0018] The communication module extracts the DC component at the welding torch end through a low-pass filter to provide power, and extracts the modulated signal through a band-pass filter for demodulation to recover the digital signal.

[0019] As a preferred embodiment of the full-function programming and display system for welding torch parameters based on a touch screen and menu navigation according to the present invention, the menu navigation control unit includes:

[0020] The input signal receiving module is used to receive touch operations and joystick operations and perform debouncing and normalization processing.

[0021] The input pattern recognition module is used to determine the source of the current operation input and maintain the input pattern state. When the joystick is used continuously, it enters the joystick input mode, increases the size of the operation response area of ​​the currently displayed menu item and increases the spacing between adjacent menu items. When the touch screen is used, it switches to the touch input mode and restores the default menu item size and spacing.

[0022] The navigation mapping module is used to convert joystick and touch operations into navigation commands for multi-level menu systems.

[0023] As a preferred embodiment of the full-function programming and display system for welding torch parameters based on touch screen and menu navigation described in this invention, the full-function programming unit for parameters includes a parameter intelligent prediction module and an offset compensation correction module.

[0024] The parameter intelligent prediction module is used to extract the currently confirmed working condition features and related parameters when the menu focus of the multi-level menu system automatically moves to the next parameter to be set, and to load the corresponding prediction model according to the currently selected welding process type to perform forward inference and obtain the basic predicted value of the target parameter.

[0025] The parameter intelligent prediction module queries the corresponding offset value from the stored offset compensation table based on the current operating condition domain identifier, combines the basic prediction value with the offset value, generates the final recommended value of the target parameter, and presents it on the parameter editing interface.

[0026] As a preferred embodiment of the full-function programming and display system for welding torch parameters based on touch screen and menu navigation described in this invention, the parameter intelligent prediction module performs missing value processing before forward inference. For currently unset associated parameters, it fills them with estimated values ​​calculated by piecewise linear interpolation based on a preset parameter anchor point library or by using a fixed non-physical range constant in conjunction with a state channel mask, so that the neural network prediction model has the ability to infer under the condition of missing features.

[0027] As a preferred embodiment of the full-function programming and display system for welding torch parameters based on touch screen and menu navigation described in this invention, the offset compensation correction module maintains an offset compensation table in random access memory and periodically backs up the offset compensation table to flash memory using a batch writing strategy.

[0028] When the number of working condition domain entries stored in the offset compensation table reaches a preset upper limit and a new entry needs to be created, the offset compensation correction module uses a least recently used algorithm based on logical access sequence number to eliminate historical entries, and writes the deleted entry data into flash memory for long-term storage before elimination.

[0029] As a preferred embodiment of the full-function programming and display system for welding torch parameters based on touch screen and menu navigation described in this invention, after the user confirms the parameters and records the actual adopted value, the offset compensation correction module calculates the prediction residual between the actual adopted value and the final recommended value.

[0030] When the absolute value of the predicted residual exceeds the preset correction dead zone threshold, the offset compensation correction module uses the predicted residual to perform exponential smoothing update on the offset value of the current operating domain based on the smoothing coefficient, so that the updated offset value gradually approaches the user's long-term operating habits.

[0031] As a preferred embodiment of the full-function programming and display system for welding torch parameters based on touch screen and menu navigation described in this invention, the dynamic maximum limit of the smoothness coefficient and the offset value are determined offline based on historical benchmark verification set and burned into the microcontroller as static constants.

[0032] The smoothing coefficient is determined based on the positive correlation between the root mean square error of the basic prediction in the corresponding working domain and the mean of the target benchmark.

[0033] The dynamic maximum limit is statistically determined based on the larger value of the ratio between the distribution quantile boundary of the historical residual sequence and the root mean square error of the basic prediction, and is used to clamp and limit the updated offset value during real-time operation.

[0034] As a preferred embodiment of the full-function programming and display system for welding torch parameters based on touch screen and menu navigation described in this invention, the display control unit automatically switches the display mode according to the working state of the welding torch. In the standby state, a first display mode containing a complete menu navigation interface is enabled for parameter editing. In the welding state, it switches to a second display mode that only displays the actual values ​​of core welding parameters and the working state. In the second display mode, the menu navigation function is disabled and the data acquisition and update logic of the offset compensation correction module is kept running normally in the background.

[0035] Secondly, embodiments of the present invention provide a method for full-function programming and displaying welding torch parameters based on a touch screen and menu navigation, employing any of the aforementioned full-function programming and display systems for welding torch parameters based on a touch screen and menu navigation, specifically including:

[0036] System initialization: The control communication module establishes a connection with external devices and automatically loads historical process plans;

[0037] Working condition settings: Receive the current welding working condition characteristics input through the multi-mode menu navigation and select the welding process type;

[0038] Intelligent prediction and recommendation: When the menu focus is switched to the parameter to be set, known features are extracted and the corresponding prediction model is used to reason to obtain the basic prediction value. Combined with the offset value of the current working condition domain, the final recommendation value is generated and highlighted in the editing interface.

[0039] Parameter confirmation and update: Receive user confirmation or fine-tuning confirmation instructions for recommended values, and when the residual between the actual adopted value and the recommended value exceeds the dead zone, trigger a closed-loop update of the offset value based on exponential smoothing in the background and store it in the offset compensation table.

[0040] Status switching and locking: After all parameters are configured and the arc is started, the display interface will automatically switch to the simplified information welding mode and lock the menu navigation function to prevent accidental operation.

[0041] The present invention has the following beneficial effects:

[0042] 1. This invention integrates a touch display module and a joystick navigation module onto the welding torch bracket, and combines them with a menu navigation control unit to construct a multi-level menu system. This allows welders to program welding parameters fully by operating the joystick with their thumb or lightly touching the screen while holding the welding torch, replacing the traditional independent PAD control terminal. This eliminates the need to switch between the welding torch and external control equipment, allowing welding operations and parameter settings to be completed continuously at the same workstation and with the same handheld posture, thereby improving the continuity and efficiency of on-site operations.

[0043] 2. This invention incorporates a parameter intelligent prediction module and an offset compensation correction module within the full-function parameter programming unit, which work in synergy with the menu navigation control unit. When a welder completes a parameter setting in the menu system and the focus automatically switches to the next parameter to be set, the change in the menu navigation status triggers the parameter intelligent prediction module to initiate the prediction process. It extracts the currently confirmed working condition features and related parameters, infers a basic predicted value using a lightweight neural network model, and then combines this with the personalized offset compensation value accumulated through exponential smoothing under the current working condition domain to generate a recommended value that highly matches the welder's long-term correction habits, directly presented on the parameter editing interface. The welder only needs to confirm with one click to adopt the recommended value, or make minor adjustments based on the recommended value before confirmation. The confirmed prediction residual is then updated and fed back to the offset compensation table through exponential smoothing, causing the recommended value under subsequent working conditions to gradually converge towards the welder's actual preferences. This system triggers a closed-loop linkage of prediction, basic inference, offset compensation, one-click confirmation, and residual feedback update through menu navigation. This simplifies the operation that originally required repeatedly and significantly pushing the joystick to one-click confirmation in most scenarios, reducing the frequency of large-scale parameter adjustments and improving the convenience and efficiency of parameter settings.

[0044] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, the drawings can be obtained from these drawings without creative effort.

[0046] Figure 1 This invention provides a schematic diagram of a full-function programming and display system for welding torch parameters based on a touch screen and menu navigation.

[0047] Figure 2 This is a schematic diagram of the full-function parameter programming unit provided by the present invention.

[0048] Figure 3The flowchart illustrates a full-function programming and display method for welding torch parameters based on a touchscreen and menu navigation, as provided by this invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] Example 1

[0051] When dealing with welding operations involving multiple product types and processes, the existing welding parameter setting and display systems are not yet perfect. There are still problems such as the inability to program full-function parameters at the welding torch end, requiring reliance on an independent PAD control terminal, which leads to operation interruptions, and the lack of intelligent guidance in the parameter setting process, which requires welders to frequently and significantly adjust the joystick from the default value, thus affecting operational efficiency.

[0052] To solve the above technical problems, such as Figure 1 As shown, Embodiment 1 of the present invention provides a full-function programming and display system for welding torch parameters based on a touch screen and menu navigation, applicable to various welding processes such as gas shielded welding (MIG / MAG), argon arc welding (TIG), and manual arc welding (MMA), specifically including:

[0053] The touch display module 100 is used to display welding parameter information and receive touch operation input;

[0054] The joystick navigation module 200 is used to provide auxiliary operation input for multi-directional joysticks and buttons;

[0055] The menu navigation control unit 300 is used to map touch operation and joystick operation into navigation and selection operation of a multi-level menu system, and adaptively adjust the menu interface according to the input mode.

[0056] The full-function parameter programming unit 400 is used to edit, verify, intelligently predict and offset compensation corrections for all welding parameters, including welding current, welding voltage, wire feed speed and welding speed.

[0057] The parameter grouping storage unit 500 is used to group, store, manage, and quickly recall welding parameters according to the process plan.

[0058] The display control unit 600 is used to switch the display mode according to the working status of the welding torch and to present welding parameters and status information in real time.

[0059] The communication module 700 is used for data communication and power supply multiplexing transmission with the welding power source and wire feeder.

[0060] The aforementioned full-function programming and display system for welding torch parameters achieves dual-mode interaction through a touch screen and joystick integrated on the welding torch bracket. Combined with a parameter intelligent prediction and offset compensation correction mechanism triggered by menu navigation status, it replaces the traditional independent PAD control terminal, enabling welders to complete full-function programming and one-click confirmation quick settings of all parameters at the welding station.

[0061] Furthermore, to better illustrate the technical solution of Embodiment 1 of the present invention, a detailed description is provided of the full-function programming and display system for welding torch parameters based on a touch screen and menu navigation, specifically including the following:

[0062] The touch display module 100 and the joystick navigation module 200 are integrated and mounted on the welding torch holder, arranged adjacent to each other. The touch display module 100 uses a 2.4-inch to 4.3-inch anti-glare color TFT LCD screen and supports capacitive touch. The joystick navigation module 200 includes a five-way joystick (up, down, left, right, press to confirm) and a separate return button. The top of the joystick has anti-slip texture to accommodate welding gloves.

[0063] The menu navigation control unit 300, the parameter full-function programming unit 400, the parameter group storage unit 500, the display control unit 600, and the communication module 700 are integrated on the control circuit board inside the welding torch bracket. It is based on an ARM Cortex-M4 microcontroller and equipped with 512KB Flash and 128KB RAM.

[0064] The communication module 700 connects to the external welding power supply and wire feeder via a communication interface on the welding torch holder. To achieve reliable communication and power supply in welding environments with strong electromagnetic interference, high-frequency inversion, and high current, and under transmission conditions with welding cables of varying lengths, a two-wire carrier communication and power supply multiplexing scheme with adaptive channel compensation is adopted, as detailed below:

[0065] Adaptive frequency-hopping carrier modulation: Adaptive binary frequency shift keying is employed. To address the high-order harmonics and broadband electromagnetic burst noise generated by the base frequency (approximately 20kHz) of the welding inverter, the communication module 700 incorporates a channel quality assessment submodule. This submodule extracts the background noise power spectral density on the communication cable using Fast Fourier Transform during idle time slots. Multiple candidate frequency band combinations are preset for carrier frequency configuration. When the signal-to-noise ratio of the current operating frequency band is detected to be below a preset threshold or subjected to harmonic interference at the same frequency, the module automatically switches to a backup carrier frequency combination with a low-noise background during frame intervals, thereby dynamically avoiding random burst interference within the hundreds of kHz range.

[0066] Adaptive compensation for long cable transmission signals based on physical characteristics: To address the high-frequency attenuation and phase distortion of carrier signals caused by the distributed parameters of welding cables of varying lengths (from several meters to tens of meters), the communication module 700 introduces an adaptive signal compensation mechanism at the receiving end. A transmission attenuation compensation model incorporating the physical characteristics of the cable is constructed, and the transfer function of the digital compensation filter is designed as follows: ;in, The complex transfer function of the compensation filter; The equivalent transmission length of the welding cable, in units of ; For complex frequency variables, The unit is ; The distributed resistance per unit length of the cable, in units of ; The distributed inductance per unit length of the cable, in units of ; The distributed conductivity per unit length of the cable, in units of ; The distributed capacitance per unit length of the cable, in units of .

[0067] In actual operation, the communication module 700 analyzes the received amplitude attenuation of the synchronization preamble sequence to invert the equivalent transmission length under the current operating conditions in real time. It also dynamically updates the weight coefficients of the compensation filter and performs digital domain pre-equalization processing on the received frequency shift keying modulation signal.

[0068] Power Supply Superposition and Separation: At the welding machine end, the 24V DC power supply is superimposed on the FM carrier signal and transmitted over the same twisted pair cable via a bias network. At the welding gun end, the DC component is extracted through a multi-stage LC low-pass filter and then regulated to power each module. The carrier signal is extracted through a parameter-adjustable bandpass filter, and after adaptive compensation filtering and amplitude limiting amplification, it is sent to the microcontroller for frequency discrimination and demodulation to recover the digital signal. The physical layer uses differential signal transmission, with a common-mode choke to suppress common-mode interference, and a transient voltage suppressor for surge protection.

[0069] Communication Protocol: A custom lightweight frame protocol is adopted, compatible with the data organization structure of the CANopen object dictionary. The data frame format includes: start identifier, target device address, command code, data length, data segment, and CRC-16 check field. The welding power supply and wire feeder are distinguished by the device address, realizing half-duplex communication. The communication process has timeout retransmission and error counting mechanisms to ensure data integrity under impulse noise interference.

[0070] Furthermore, the menu navigation control unit 300 includes an input signal receiving module 310, an input pattern recognition module 320, and a navigation mapping module 330.

[0071] The input signal receiving module 310 receives touch signals from the touch display module 100 and joystick signals from the joystick navigation module 200, and performs debouncing and normalization processing. The touch signals include click, swipe, and long press; the joystick signals include directional movement and button press.

[0072] The input pattern recognition module 320 determines the source of the current input and maintains the input pattern state. When the joystick is used continuously, the system enters the joystick input mode; when the touchscreen is used, the system switches to the touch input mode and adjusts the display state of the menu system according to the current input mode: when in joystick input mode, the size of the operation response area of ​​the currently displayed menu item is increased, and the spacing between adjacent menu items is increased; when in touch input mode, the default menu item size and spacing are restored.

[0073] The navigation mapping module 330 converts joystick signals and touch signals into navigation commands for the menu system. The joystick's up, down, left, and right movements are mapped to the menu focus moving up, down, left, and right, respectively, while the confirmation button is mapped to confirming menu item selections. In the touch signals, clicking a menu item is directly mapped to confirmation, and swiping left or right is mapped to increasing / decreasing values ​​or switching pages.

[0074] The menu system has a multi-level hierarchical structure, including a main menu level, parameter setting level, parameter editing level, and process scheme management level. The main menu level includes welding process selection, parameter settings, process scheme management, and system settings. The parameter setting level is functionally divided into subgroups for basic parameters, wire feeding parameters, pulse parameters, and process control parameters. The parameter editing level provides interfaces for increasing / decreasing values ​​and switching modes for selected parameters. The process scheme management level provides operations for viewing, selecting, creating, and managing schemes.

[0075] Furthermore, such as Figure 2 As shown, the full-function parameter programming unit 400 includes a parameter classification module 410, a parameter editing module 420, a parameter validity verification module 430, a parameter intelligent prediction module 440, and an offset compensation correction module 450.

[0076] The parameter classification module 410 divides all programmable welding parameters into basic parameter groups, wire feeding parameter groups, pulse parameter groups, and process control parameter groups according to their function types.

[0077] The parameter editing module 420 responds to parameter editing commands received via the menu system, allowing for numerical increases or decreases, mode switching, or option selection of the selected welding parameters. Numerical adjustment supports both step and continuous acceleration modes.

[0078] After the parameters are edited, the parameter validity verification module 430 performs verification according to preset parameter validity rules, including parameter value range verification, parameter constraint relationship verification, and process rationality verification. If the verification fails, the reason for failure and correction suggestions are displayed on the touch display module 100.

[0079] The intelligent parameter prediction module 440 embeds multiple offline pre-trained neural network prediction models and, in conjunction with an offset compensation table isolated by operating domain, achieves parameter prediction. The specific implementation steps are as follows:

[0080] 441. The intelligent parameter prediction module embeds multiple offline pre-trained neural network prediction models and, in conjunction with offset compensation tables isolated by working condition domain, achieves parameter prediction. Considering the differences in parameter systems for various welding processes such as MIG / MAG, TIG, and MMA, corresponding basic prediction models are trained and deployed independently for each welding process. The specific implementation steps for offline model training and online deployment are as follows:

[0081] An independent feature mapping system is constructed for each welding process. Taking the MIG model as an example, the input layer contains a total of 15 feature nodes. The base material uses 4-dimensional unique thermal encoding, corresponding to carbon steel, stainless steel, aluminum alloy, and copper alloy; the shielding gas type uses 4-dimensional unique thermal encoding, corresponding to 100% carbon dioxide, argon-rich mixture, pure argon, and others; plate thickness and welding wire diameter are both treated as single-dimensional continuous variables, with ranges set to 0.5 to 50.0 mm and 0.8 to 1.6 mm respectively, and pre-normalized to the 0-1 range; the remaining 5 dimensions serve as associated parameter channels, receiving currently confirmed welding values ​​such as welding current and welding voltage. If not set, missing value handling mechanisms are used to fill in the features. The model's output layer adopts a multi-task learning architecture, setting a fixed set of target parameters for different welding processes. The output layer of the MIG model is fixed at four neurons, corresponding to welding voltage, welding current, wire feed speed, and inductance, respectively. Similarly, the output layer of the TIG model is also fixed at four neurons, corresponding to base current, peak current, pulse frequency, and duty cycle. During forward inference at the welding torch end, the output layer synchronously outputs the predicted values ​​of all core parameters for that process. The system, based on the current focus position in the menu navigation, selectively extracts the values ​​of the corresponding output nodes as the base predicted values ​​for the target parameters.

[0082] On the offline server side, standard welding process specification libraries, automated special machine operation logs, and expert welder operation records are collected through an industrial IoT gateway. After outlier removal and smoothing filtering, a supervised learning dataset of 50,000 samples is constructed. A random stratified sampling algorithm is used to divide the dataset into non-overlapping training, validation, and test sets at a ratio of 70%, 15%, and 15%, respectively. The model topology uses two fully connected hidden layers, each containing 32 neurons, and applies the Corrected Linear Unit (ReLU) as the non-linear activation function. The model training process uses the Adam optimizer, with a batch size of 128 and a maximum training epoch of 200. The initial learning rate is set to 0.1%, and a cosine annealing learning rate decay strategy is used to ensure smooth convergence in the later stages of training. The mean squared error loss function is used, and an L2 regularization term is added to the weights of the fully connected layers to suppress model overfitting.

[0083] After the single-floating-point precision (FP32) model has been trained and reached the validation set convergence criterion, post-trained static quantization (PTQ) is applied to convert it into an 8-bit integer quantization (INT8) model. During quantization calibration, forward inference is performed using test set data to statistically analyze the histogram of the activation values ​​of each hidden layer. The Kullback-Leibler divergence algorithm is then used to find the optimal cutoff threshold and determine the scaling factor between floating-point and integer types. Through this quantization process, the data bit width of the model weights is reduced, and the single-model volume compression rate reaches approximately 75%, ensuring that the solidified volume of a single process model is stably controlled within 20KB. Before code deployment, quantization error evaluation is performed, comparing the absolute deviation of the quantized model's output values ​​on the test set with that of the native floating-point model.

[0084] Tests show that the truncation error introduced by quantization is less than 0.1 volts in the welding voltage dimension and less than 1 amp in the welding current dimension, which is well within the parameter disturbance tolerance range allowed by the welding process. Finally, the compressed INT8 model weights, along with the scaling factors of each layer, are burned into the microcontroller's read-only memory in the form of a one-dimensional constant array in C language. During online real-time inference, efficient output is achieved through shift and fixed-point multiplication-addition operations.

[0085] 442. To ensure the model can still perform forward inference correctly when the input feature vector contains missing values, a built-in missing value handling mechanism is implemented, which specifically includes:

[0086] Basic Anchor Point Interpolation Filling Scheme: For routine operating conditions, the microcontroller's flash memory is categorized by base material and welding process, pre-setting and storing only the parameter anchor point data of core key variables under typical boundary conditions. For example, for gas metal arc welding of stainless steel, only the recommended welding current and voltage reference values ​​corresponding to four nodes with plate thicknesses of 1mm, 3mm, 6mm, and 10mm are stored. When there are undefined associated parameters in actual operation, the system extracts the currently known core operating condition variables, retrieves two adjacent physical nodes from the pre-set parameter anchor point library, and calculates the estimated value of the target missing parameter through piecewise linear interpolation, which is then used as the default value for filling. By replacing the exhaustive table construction of multi-dimensional operating conditions with real-time dynamic calculation, the storage complexity is reduced from the combination order of feature dimensions to the constant order of node dimensions, adapting to the storage resource limitations of the microcontroller.

[0087] Dual-channel masked training and inference scheme: To address the problem of feature-deficient inference in the absence of prior knowledge, the input layer topology and offline training logic of the neural network are modified.

[0088] Input layer structure expansion: The model's input nodes are split into two channels, with each original input feature corresponding to a numerical channel and a state channel. When a parameter is set, its numerical channel receives the normalized parameter value, and the state channel is set to a valid Boolean value; when the parameter is not set, the numerical channel receives a fixed non-physical range constant, and the state channel is set to a missing Boolean value. This structure avoids directly mixing missing identifiers into the physical continuous variable space, which could distort the data distribution.

[0089] Masking Pattern and Probability Distribution Construction: When constructing the training set offline, masking data augmentation is performed on each welding process specification sample with complete parameters. For each associated parameter in the input feature vector, a random mask sequence is generated using an independent Bernoulli distribution. The single-dimensional mask trigger probability is dynamically set to a random value within the range of 15% to 40%, thereby generating augmented training batches containing combinations of multi-dimensional features with different proportions of missing features.

[0090] Hidden Layer Physical Coupling Inference: During the network's backpropagation training, regardless of the feature mask combination of the input layer, the loss function continuously calculates the error between the output layer and the complete feature labels and updates the gradient. This calculation process forces the hidden layer neurons to discover and fit the inherent physical relationships and process constraints between various welding parameters. When performing real-time forward inference on the microcontroller, the model identifies the missing dimensions of the input vector through the state channel, utilizes the fixed feature association weights in the hidden layer, and infers reasonable prediction target values ​​from the feature responses of known parameter dimensions.

[0091] 443. The system predefines operating condition domains, which are generated by combining material, plate thickness range, and welding wire diameter to form an operating condition domain identifier. For each operating condition domain, an offset compensation table is maintained in the MCU RAM, recording the offset values ​​corresponding to each predictable parameter within that domain. (Initial values ​​are all 0).

[0092] To balance power-loss integrity and Flash write lifespan, a fixed area in the Flash memory is allocated as a non-volatile backup of the offset table. A RAM caching and batch write strategy is employed: during normal operation, offset value updates only modify entries in RAM; when the cumulative number of offset update operations reaches a preset threshold (e.g., 10 times), or when the system receives a shutdown command and enters standby / hibernation mode, the offset compensation table in RAM is written to the Flash memory. A balanced write / erase algorithm is used to distribute write / erase operations across different physical sectors of the Flash memory, extending its lifespan.

[0093] 444. To prevent the storage resources from being exhausted due to the infinite growth of entries in the operating domain offset compensation table over time, the system sets a storage limit and a least recently used (LRU) eviction mechanism based on logical sequence numbers, specifically including:

[0094] The system allocates a static contiguous block of memory (RAM) to maintain the offset compensation table. This table has a maximum configuration of 64 entries. Each entry consists of a structure containing: a 32-bit condition domain identifier, an array of offset values ​​for multi-dimensional associated parameters, a status check bit, and a 32-bit unsigned integer logical access sequence number. After byte alignment, the space occupied by a single entry is controlled to within 32 bytes. The 64 entries occupy a maximum of 2KB of RAM, approximately 1.5% of the microcontroller's total 128KB capacity, thus not compressing the system's core operating resources.

[0095] To avoid introducing a hardware real-time clock (RTC) module that relies on an independent power supply, the system uses a monotonically increasing logical access sequence number instead of an absolute timestamp. The microcontroller maintains a 32-bit global access counter in memory. During the operation of the welding torch, each time any entry in the offset compensation table is read by the parameter intelligent prediction module or updated by the offset compensation correction module, the global access counter increments by 1, and the incremented count value is directly overwritten into the logical access sequence number field of the target entry.

[0096] When the system backs up the offset compensation table from memory to flash memory according to the preset batch write strategy, the 32-bit logical access sequence number built into each table entry is synchronously written to the non-volatile memory sector as part of the data payload. During the initialization phase of system power failure and restart, the microcontroller loads the most recent table entry data stored in flash memory into memory. Subsequently, the system performs a sequence scan, traversing all the logical access sequence numbers of the loaded table entries to retrieve the current maximum value. The system increments the obtained maximum sequence number by 1 and assigns it to the global access counter in memory to restore the initial state. This mechanism ensures the global monotonically increasing continuity of the access sequence across power failure cycles without increasing hardware costs.

[0097] When the number of entries stored in memory reaches the preset limit of 64, and the system receives a new, unrecorded condition field parameter and requests space allocation, the microcontroller initiates a linear traversal search. The system compares the logical access sequence numbers of all current entries, locates the entry with the smallest sequence number, and identifies it as a cold data entry with the lowest historical access frequency. The system then transfers the entire data structure of this entry to be phased out to the long-term archive area of ​​flash memory, releases its node pointer in memory, overwrites and allocates it to the newly requested condition field, and assigns the new entry the new value of the current global access counter.

[0098] As an alternative, a first-in, first-out (FIFO) elimination strategy can be adopted, which prioritizes deleting the earliest created operating condition field record. This is simpler to implement and consumes less RAM and CPU resources.

[0099] 445. The parameter intelligent prediction module 440 triggers prediction by monitoring the menu navigation status, and includes two triggering methods:

[0100] Parameter switching trigger: This is triggered when a welder completes the editing and confirmation of a parameter in the parameter setting process, and the menu focus automatically moves to the next parameter to be set. The system extracts the currently confirmed working condition features and parameter values, processes missing values, and combines them into the input feature vector of the corresponding process model. It then performs forward inference of the model to obtain the basic predicted value of the target parameter. .

[0101] Global pre-fill trigger: When a welder creates a new process plan or calls an existing plan, the system performs batch basic predictions on the parameters that have not yet been set in the plan.

[0102] Obtain the base forecast value Then, the system retrieves the offset value of this parameter from the RAM offset compensation table based on the current operating condition domain ID. (If there is no record, the value is 0). The final recommendation value is calculated as follows: The final recommended value is quickly clamped to the allowable range by the parameter validity verification module 430 and then displayed on the parameter editing interface by the display control unit 600: the initial value of the numerical input box is set to... The value is highlighted and accompanied by a message: "Suggested value, press the OK button to adopt / press the direction keys to fine-tune." Welders can directly press the OK button to adopt the value, or make minor adjustments using the joystick and then confirm.

[0103] 446. After confirming the parameters, the system records the actual values ​​used. Calculate the predicted residuals: The predicted residual Indicates the current offset value The amount of correction that was not fully absorbed.

[0104] To avoid invalid updates caused by minor adjustments, a correction dead zone threshold is set. (For example, welding current dead zone ±1A, wire feed speed dead zone ±0.1m / min). If Exceeding the preset dead zone threshold Then the offset value of the current operating condition domain Perform an exponential smoothing update based on the predicted residuals: ;

[0105] The above update logic is also equivalent to: ;in, The historical cumulative offset compensation value corresponding to the current operating condition domain before the update has the same unit of measurement as the physical unit of the target welding parameter. For example, the unit is volts for welding voltage parameters. The range of values ​​is limited by the static maximum amplitude. The unit for the wire feeding speed parameter is meters per minute (m / min). The range of values ​​is ; The latest offset compensation value corresponding to the current working condition domain after the update has its unit of measurement consistent with the physical unit of the target welding parameters; This refers to the predicted residual between the actual value used in the current operation step and the final recommended value of the system. Its unit of measurement is consistent with the physical unit of the target welding parameter. For example, the unit of the welding voltage residual is... The unit of wire feed speed residual is The range of values ​​is determined by the operator's maximum adjustment span in a single operation; The smoothing coefficient is used to control the weighting of the residual absorbed in the current operation step. It is a dimensionless constant, and its value ranges from 1 to 10. between.

[0106] The above mechanism ensures that the offset value gradually approaches the user's long-term operating habits, and will not be affected by the welder accepting the recommended value. This can lead to unnecessary cumulative drift.

[0107] To ensure update stability, offset value Maximum amplitude is set Furthermore, within the same operating range, the system counts the cumulative number of effective corrections for that range; when the cumulative number of effective corrections reaches a preset threshold (e.g., 3 times), the corresponding offset value is used to participate in subsequent predictions; when it is below the threshold, the offset value is only accumulated but does not participate in predictions. Calculation, using only Recommended value.

[0108] 447. No model weight training is performed at the welding torch end. The system periodically uploads log information such as the operating condition characteristics, basic predicted values, actual adopted values, and offset update amounts for each parameter correction to the storage device connected to the welding power supply end via the communication module 700, or to the cloud server via the welding machine networking module. After collecting operating data from multiple welding machines in the cloud, each process model can be retrained on the server side, and the model parameters stored in the welding torch end MCU can be updated via firmware upgrade.

[0109] Furthermore, the parameter grouping storage unit 500 includes a process scheme creation module 510, a parameter association storage module 520, a quick recall module 530, and a scheme management module 540.

[0110] The process scheme creation module 510 is used to create process scheme records that include information such as identifier, name, applicable conditions, and time.

[0111] The parameter association storage module 520 associates and stores the currently programmed welding parameter combination with the process scheme identifier.

[0112] The quick call module 530 responds to selection operations, reads the corresponding welding parameter combination, loads it into the programming unit, and automatically sends it to external devices. During the call, it can automatically perform pre-filling processing of basic prediction plus offset compensation.

[0113] The solution management module 540 provides solutions for viewing, renaming, copying, deleting, and exporting.

[0114] Furthermore, the display control unit 600 includes a welding status display module 610, a process scheme display module 620, and a display mode switching module 630.

[0115] The welding status display module 610 displays the actual welding current, welding voltage, and wire feed speed fed back by the communication module 700 in real time during the welding process, and compares them with preset target values ​​side by side or in the form of a deviation indicator bar. When the actual value deviates from the target value by more than a preset threshold, the value is visually alerted by changing color or flashing.

[0116] The process solution display module 620 continuously displays the currently active process solution identifier and solution name on the touch display module 100.

[0117] The display mode switching module 630 automatically switches the display mode according to the working status of the welding torch:

[0118] Standby mode (first display mode): Displays the main operation interface with a complete menu navigation interface, providing full parameter editing, scheme management and intelligent prediction functions.

[0119] Welding Status (Second Display Mode): Displays a simplified information interface, showing only the actual values ​​of core welding parameters, process scheme identifiers, and welding status indicators. Menu navigation is disabled to prevent accidental operation; only the function of long-pressing the joystick confirmation button to enter parameter fine-tuning mode is retained. During welding status, the data acquisition and update logic of the offset compensation correction module 450 operates normally in the background without interfering with the display interface.

[0120] To better illustrate the full-function programming and display system for welding torch parameters in this invention, such as... Figure 3 As shown, a method for full-function programming and displaying welding torch parameters based on touch screen and menu navigation is also provided. Taking a large structural component welding workshop in a heavy equipment manufacturing base as an example, a closed-loop parameter setting framework is formed through a dual-mode menu navigation system (touch and joystick at the welding torch end) combined with an intelligent prediction and offset compensation correction mechanism with isolated working conditions. This framework consists of menu navigation-triggered prediction → basic model inference → personalized offset compensation → one-click confirmation setting → smooth update of residual index. This allows welders to complete full-function parameter programming without leaving their workstations, and in most scenarios, only one-click confirmation is required. The specific steps include:

[0121] S1. The system is powered on and initialized. The communication module 700 establishes a connection with the welding power supply and wire feeder through two-wire BFSK carrier communication. The touch screen displays the standby main interface and automatically loads the last used process scheme.

[0122] S2. The welder selects "New Process Scheme" via the joystick, sets the working conditions as follows: base material stainless steel, plate thickness 3mm, welding wire diameter 1.0mm, and selects MIG welding process.

[0123] S3. Enter the parameter setting process. The system automatically loads the MIG prediction model. The welder first sets the welding current to 120A and confirms. The menu focus automatically switches to the "Welding Voltage" parameter setting option.

[0124] S4. The intelligent parameter prediction module 440 detects a parameter switching trigger, extracts the currently confirmed features (stainless steel, 3mm, 1.0mm, current 120A), fills in the undefined associated parameters with the global default values ​​for this operating condition domain, and sends them to the MIG model for inference to obtain the basic predicted values. =18.2V. Checking the RAM revealed the current voltage offset. The recommended value is calculated as +0.4V. =18.6V, and highlighted in the editing interface. The welder deemed it suitable and directly pressed the confirmation button to adopt it.

[0125] S5. The menu focus automatically moves to the "Wire Feeding Speed" parameter setting option. The model's basic prediction value is 5.2 m / min, and the current offset is... The initial speed was 0 m / min, with a recommended value of 5.2 m / min. The welder, based on experience, lightly adjusted the rocker arm to 5.5 m / min and confirmed the result. Predicted residuals. =+0.3m / min, exceeding the preset dead zone threshold, perform offset update: Write the update result to the RAM offset table and update the timestamp.

[0126] S6. Once all parameters are set and pass validity verification, they are automatically saved to the new solution. The welder activates the welding torch, and the screen automatically switches to the simplified welding mode interface, displaying only the actual values ​​of core parameters and the solution identifier. The menu navigation function is locked.

[0127] S7. During the welding process, the background collects correction logs. The offset value is gradually converged to the welder's long-term correction habits through exponential smoothing updates. The recommended value is automatically adjusted under the same working conditions, and the proportion of confirmed and usable values ​​is gradually increased.

[0128] The aforementioned method utilizes a dual-mode menu navigation system (touch control and joystick) combined with an intelligent prediction and offset compensation correction mechanism based on working condition domain isolation. This forms a closed-loop parameter setting process: menu navigation-triggered prediction → basic model inference → personalized offset compensation → one-click confirmation setting → residual exponential smooth update. This allows welders to complete full-function programming of all welding parameters at the welding station, replacing traditional independent PAD control terminals. Simultaneously, intelligent prediction based on automatic triggering of menu navigation status directly positions the parameters to be set near the welder's expected value, reducing the frequency of large-range joystick adjustments. In most scenarios, only one-click confirmation is required, improving parameter setting efficiency. The method employs an offline-trained multi-process lightweight model and a decoupled architecture for welding gun-end offset compensation, allowing the welding gun to perform only inference and simple arithmetic updates without online training, ensuring high feasibility under MCU resource conditions. The use of an exponential smooth update mechanism based on predicted residuals and a working condition domain isolation strategy ensures that the offset converges stably to the welder's long-term correction habits and that different material working conditions do not interfere with each other, balancing personalized adaptation and system stability.

[0129] As an optional embodiment, the joystick navigation module 200 includes digital quantities (direction movement, button pressing, and return button) and analog quantities (if the joystick adopts a dual-axis potentiometer structure).

[0130] One approach is to use a periodic software counter debouncing algorithm triggered by a microcontroller timer to debouncing digital signals. The principle is as follows: A timer interrupt with a period of 5ms is set, and the GPIO pin levels of the joystick and buttons are polled in the interrupt service routine (ISR). An 8-bit ring shift register or continuous counter is maintained for each pin. Only when the pin level remains consistent for four consecutive samplings (i.e., 20ms) is the state of the key value confirmed to have flipped (pressed or released). This method can filter out high-frequency spike pulses generated at the moment of mechanical contact closure and opening.

[0131] The analog / digital signal normalization can be achieved using state machine encoding (for switching signals) and linear normalization (for potentiometer axial ADC values), as follows:

[0132] Switch-type joystick: The debouncing states of the six independent GPIOs (up, down, left, right, confirm, and return) are normalized and encoded into an 8-bit standard key-value byte. (For example: the upward key is encoded as 0x01, and the confirmation key is encoded as 0x05), and the subsequent modules are uniformly delivered for identification.

[0133] Potentiometer-type joystick: If the joystick output is a dual-axis ADC voltage value, then... The native 12-bit ADC sample value ( ), normalized to through the mapping formula The interval. Let the range of the intermediate dead zone be... After exceeding the dead zone, the X-axis component The normalization formula is: ;in The current sampling voltage, Center bias voltage, This is the maximum effective span on one side, thereby eliminating resistance errors between different hardware batches.

[0134] As an optional embodiment, the capacitive touch signal output by the touch display module 100 is easily affected by welding arc light and high-frequency pulse interference, resulting in coordinate jumps and multi-point mis-touch.

[0135] Touch signal dejitter can be achieved by using a moving average filter in conjunction with timing dead-time limiting, and the principle is as follows:

[0136] Spatial debouncing: When the touch chip (IC) triggers an interrupt, the MCU continuously reads... Groups (e.g.) Current coordinates Calculate its standard deviation. If the Euclidean distance between two sets of continuous coordinates exceeds the set abrupt change threshold... If the signal is a flying point caused by electromagnetic noise, it will be discarded directly.

[0137] Smoothing filtering: A moving weighted average is applied to the valid coordinates that have passed spatial verification to calculate the smoothed coordinates of the current frame. ;in For the weighting coefficients, satisfying The closer the sampling point is to the current sampling point, the greater its weight, which is used to eliminate thermal noise on the screen surface.

[0138] Timing debouncing: For click events, a timing window of 50ms is set between touch press and touch release. Touch signals shorter than this time are treated as blank noise and not responded to.

[0139] One approach is to use bilinear coordinate mapping to normalize touch signals, based on the following principle:

[0140] The native resolution of the capacitive touch sensing layer (e.g.) ) is usually related to the display resolution of the color TFT LCD screen (e.g. or The parameters are inconsistent, and the direction may be mirrored or reversed. The module uses preset calibration matrix parameters. The original coordinates of the filtered touch pixels Normalized to absolute pixel coordinates of the LCD screen :

[0141] ;

[0142] ;

[0143] After this normalization process, regardless of the size or resolution of the touchscreen used at the bottom layer, the navigation mapping module 330 receives standard screen pixel coordinates, thus achieving hardware and software decoupling.

[0144] As an optional embodiment, in step S446, the smoothing coefficient The method for determining it is as follows:

[0145] For any predefined operating condition domain (A unique identifier field generated by a combination of specific material, plate thickness range, and welding wire diameter) Historical expert-level welding procedure specification (WPS) data and actual adoption data of experienced welders corresponding to this working condition field are extracted from the offline server database to construct a system containing... Static benchmark validation set of group samples ;in For the input feature vector, The actual values ​​used for the corresponding target parameters. The benchmark validation set. Input feature vector Each value is input into a pre-trained and quantized compressed process model (such as a MIG model), and forward inference is performed to obtain the corresponding sequence of basic predicted values. Subsequently, the original prediction residuals between the base prediction values ​​and the actual adopted values ​​under this operating condition domain are calculated. : ; Utilizing native prediction residual sequences Calculate the root mean square error of the basic prediction in this operating domain without any offset compensation. : Using the root mean square error of the basic prediction Target baseline mean of parameters corresponding to this operating condition domain The ratio between them, i.e., the coefficient of variation, is calculated. ;in, . Normalization fixed as , Normalization fixed as .

[0146] Furthermore, in this embodiment, the maximum amplitude is... Used to prevent offset compensation value from the welding torch tip due to misoperation or abnormal operating conditions. Unrestricted drift. Boundary statistics can be calculated based on the distribution characteristics of the historical baseline residual sequence. The residual sequence obtained above... Sort by absolute value and calculate its Quantiles, denoted as That is, ensure The historical expert corrections all fall within this boundary. To ensure the robustness of the algorithm, this system sets the dynamic maximum amplitude of this operating domain. Defined as: ;

[0147] During real-time operation at the welding torch end, once the offset compensation correction module 450 calculates... Immediately apply the following clamping restrictions:

[0148] ;

[0149] The unique identifier and corresponding smoothing coefficient for each working condition domain obtained from the offline calculations in the above steps are used as the basis for further calculations. and dynamic maximum amplitude limit The structure is arranged as a statically addressable structure table and burned into the Flash memory of the welding torch-end MCU along with the process model as a read-only constant array (const form). When the welder switches to a certain working condition domain during actual operation, the parameter intelligent prediction module 440 and the offset compensation correction module 450 directly read the corresponding value from the Flash memory based on the working condition domain ID. and The static constant is used in the background exponential smoothing and clamping formulas.

[0150] As an optional embodiment, in step S446, to address the lack of adaptability of the fixed correction dead zone threshold under different operating conditions such as low current in thin plates and high current in medium-thick plates, the correction dead zone setting mechanism within the full-function parameter programming unit specifically includes:

[0151] Before performing exponential smoothing updates, the offset compensation correction module dynamically calculates the correction dead zone threshold for the target parameter based on the baseline prediction value output by the current parameter intelligent prediction module. For arbitrary continuous physical parameters, such as welding current, welding voltage, and wire feed speed, a calculation model combining proportional constraints and absolute limiting is used to dynamically calculate the correction dead zone threshold. ;in, Dynamically adjust the dead zone threshold for the current target parameters; These are the base predicted values ​​for the current target parameters; This is the proportional adjustment coefficient; The minimum dead zone threshold is the basic parameter of the target. The upper limit threshold of the maximum dead zone for the target parameter.

[0152] For example, taking welding current as an example, the threshold configuration for this parameter preset in the microcontroller's flash memory is as follows: Basic minimum dead zone threshold. Maximum dead zone upper limit threshold Proportional adjustment coefficient When dealing with thin plate conditions, if the current base prediction value... The product term is calculated to be After clamping using the above formula, the actual output dynamically corrects the dead zone threshold. When dealing with medium-thick plate conditions, if the current foundation prediction value... The product term is calculated to be This value falls within the upper and lower limits, and the actual output dynamically corrects the dead zone threshold. .

[0153] In actual operation, the dynamic correction dead zone threshold of the current target parameters is calculated. Then, the offset compensation correction module extracts the absolute value of the prediction residual between the actual adopted value and the final recommended value. When this absolute value is greater than the dynamic correction dead zone threshold... When this occurs, the system triggers an exponential smoothing closed-loop update operation for the offset value of a specific parameter within the operating range, thereby achieving adaptive anti-interference filtering under the full range of operating conditions.

[0154] As an optional embodiment, in step S446, the smoothing coefficient (For example Under continuous large-scale correction conditions, the offset convergence path can easily become too long and the parameter tuning efficiency can be low. This embodiment introduces a dynamic gain operator based on residual variability to perform nonlinear correction on the traditional exponential smoothing formula. To accurately capture the welder's true intention for continuous large-scale adjustments and eliminate random parameter tuning noise, the system maintains a length of [missing information - likely a variable] in memory. ( Set as 3) Historical residual sliding window sequence The system constructs a dynamic gain operator in real time by calculating the unidirectional symbol accumulation characteristics and fluctuation intensity of the sequence. The adaptive update formula is designed as follows:

[0155] ;

[0156] ;

[0157] in, For dynamic gain operators; The nonlinear sensitivity adjustment coefficient can be preset to [value]. ; To prevent tiny positive numbers with a denominator of zero.

[0158] If the welder makes minor adjustments oscillating around the recommended value, the historical residuals will alternate between positive and negative signs. At this point, the numerator term... ,lead to The formula adaptively degenerates into a conventional fixed-step exponential smoothing update, ensuring that the system does not experience policy oscillations due to occasional random disturbances and maintaining system stability. If there is a significant deviation between the system's recommended value and the welder's long-term preferences, leading to continuous... Then, make significant parameter adjustments in the same direction (i.e., all) The signs are the same and both are much larger than the dead zone threshold. At this point, the sum of the absolute values ​​of the residuals equals the absolute value of the sum, and the ratio of the preceding terms approaches 1 / 2. Meanwhile, due to single residual The ratio of the squares of the latter term increases exponentially, through Saturation clamping constraint of function, dynamic gain operator It will converge rapidly to its theoretical maximum. In this state, the equivalent smoothing coefficient changes from the basic... (like Instant adaptive scaling to (like This doubles the rate at which a single-step update absorbs the current residual, and the offset value... This allows it to overcome the asymptotic limitations of conventional exponential smoothing, achieving rapid leaps and convergence to new target operating domains. The mechanism balances filter stability during fine-tuning with dynamic response efficiency in scenarios involving large parameter adjustments.

[0159] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0160] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A full-function programming and display system for welding torch parameters based on a touch screen and menu navigation, characterized in that, include: The touch display module is used to display welding parameter information and receive touch operation input; The joystick navigation module provides auxiliary operation input for multi-directional joystick and button operation; The menu navigation control unit is used to map touch operations and joystick operations into navigation and selection operations of a multi-level menu system, and adaptively adjust the menu interface according to the input mode. The full-function parameter programming unit is used to perform welding parameter editing, verification, intelligent prediction, and offset compensation correction. The parameter grouping storage unit is used to group, store, manage, and retrieve welding parameters according to the process plan. The display control unit is used to switch the display mode according to the working status of the welding torch and to present welding parameters and status information in real time. The communication module is used for data communication and power supply multiplexing with external devices.

2. The full-function programming and display system for welding torch parameters based on touch screen and menu navigation according to claim 1, characterized in that: The communication module uses a two-wire carrier communication and power supply multiplexing architecture to connect to external devices. The communication module has a built-in channel quality assessment submodule, which is used to detect the signal-to-noise ratio of the current operating frequency band in the idle time slot, and automatically switch to the backup carrier frequency when the signal-to-noise ratio is lower than a preset threshold, so as to realize adaptive frequency hopping carrier modulation. The communication module introduces a signal adaptive compensation mechanism at the receiving end, which inversely calculates the equivalent transmission length of the welding cable by analyzing the received amplitude attenuation of the synchronization preamble sequence, and dynamically updates the weight coefficients of the digital compensation filter to perform pre-equalization processing on the received modulated signal. The communication module extracts the DC component at the welding torch end through a low-pass filter to provide power, and extracts the modulated signal through a band-pass filter for demodulation to recover the digital signal.

3. The full-function programming and display system for welding torch parameters based on a touch screen and menu navigation as described in claim 1, characterized in that, The menu navigation control unit includes: The input signal receiving module is used to receive touch operations and joystick operations and perform debouncing and normalization processing. The input pattern recognition module is used to determine the source of the current operation input and maintain the input pattern state. When the joystick is used continuously, it enters the joystick input mode, increases the size of the operation response area of ​​the currently displayed menu item and increases the spacing between adjacent menu items. When the touch screen is used, it switches to the touch input mode and restores the default menu item size and spacing. The navigation mapping module is used to convert joystick and touch operations into navigation commands for multi-level menu systems.

4. The full-function programming and display system for welding torch parameters based on touch screen and menu navigation according to claim 1, characterized in that: The full-function parameter programming unit includes a parameter intelligent prediction module and an offset compensation and correction module; The parameter intelligent prediction module is used to extract the currently confirmed working condition features and related parameters when the menu focus of the multi-level menu system automatically moves to the next parameter to be set, and to load the corresponding prediction model according to the currently selected welding process type to perform forward inference and obtain the basic predicted value of the target parameter. The parameter intelligent prediction module queries the corresponding offset value from the stored offset compensation table based on the current operating condition domain identifier, combines the basic prediction value with the offset value, generates the final recommended value of the target parameter, and presents it on the parameter editing interface.

5. The full-function programming and display system for welding torch parameters based on touch screen and menu navigation according to claim 4, characterized in that: The parameter intelligent prediction module performs missing value processing before forward inference. For currently undefined associated parameters, it fills them with estimated values ​​calculated by piecewise linear interpolation based on a preset parameter anchor point library or by using fixed non-physical range constants in conjunction with state channel masks, so that the neural network prediction model has the ability to infer under the condition of missing features.

6. The full-function programming and display system for welding torch parameters based on touch screen and menu navigation according to claim 4, characterized in that: The offset compensation correction module maintains an offset compensation table in random access memory and periodically backs up the offset compensation table to flash memory using a batch write strategy. When the number of working condition domain entries stored in the offset compensation table reaches a preset upper limit and a new entry needs to be created, the offset compensation correction module uses a least recently used algorithm based on logical access sequence number to eliminate historical entries, and writes the deleted entry data into flash memory for long-term storage before elimination.

7. The full-function programming and display system for welding torch parameters based on touch screen and menu navigation according to claim 4, characterized in that: After the user confirms the parameters and records the actual adopted value, the offset compensation correction module calculates the prediction residual between the actual adopted value and the final recommended value; When the absolute value of the predicted residual exceeds the preset correction dead zone threshold, the offset compensation correction module uses the predicted residual to perform exponential smoothing update on the offset value of the current operating domain based on the smoothing coefficient, so that the updated offset value gradually approaches the user's long-term operating habits.

8. The full-function programming and display system for welding torch parameters based on touch screen and menu navigation according to claim 7, characterized in that: The dynamic maximum limit of the smoothing coefficient and the offset value are both determined offline based on historical benchmark verification sets and burned into the microcontroller as static constants; The smoothing coefficient is determined based on the positive correlation between the root mean square error of the basic prediction in the corresponding working domain and the mean of the target benchmark. The dynamic maximum limit is statistically determined based on the larger value of the ratio between the distribution quantile boundary of the historical residual sequence and the root mean square error of the basic prediction, and is used to clamp and limit the updated offset value during real-time operation.

9. The full-function programming and display system for welding torch parameters based on touch screen and menu navigation according to claim 1, characterized in that: The display control unit automatically switches the display mode according to the working status of the welding torch. In standby mode, it enables the first display mode, which includes a complete menu navigation interface, for parameter editing. In welding mode, it switches to the second display mode, which only displays the actual values ​​of core welding parameters and the working status. In the second display mode, the menu navigation function is disabled and the data acquisition and update logic of the offset compensation correction module is kept running normally in the background.

10. A method for full-function programming and displaying welding torch parameters based on a touch screen and menu navigation, employing the full-function programming and display system for welding torch parameters based on a touch screen and menu navigation as described in any one of claims 1 to 9, characterized in that, include: System initialization: The control communication module establishes a connection with external devices and automatically loads historical process plans; Working condition settings: Receive the current welding working condition characteristics input through the multi-mode menu navigation and select the welding process type; Intelligent prediction and recommendation: When the menu focus is switched to the parameter to be set, known features are extracted and the corresponding prediction model is used to reason to obtain the basic prediction value. Combined with the offset value of the current working condition domain, the final recommendation value is generated and highlighted in the editing interface. Parameter confirmation and update: Receive user confirmation or fine-tuning confirmation instructions for recommended values, and when the residual between the actual adopted value and the recommended value exceeds the dead zone, trigger a closed-loop update of the offset value based on exponential smoothing in the background and store it in the offset compensation table. Status switching and locking: After all parameters are configured and the arc is started, the display interface will automatically switch to the simplified information welding mode and lock the menu navigation function to prevent accidental operation.

Images