Intelligent welding system and method based on multi-information fusion and adaptive control

Abstract

The invention discloses an intelligent welding system and method based on multi-information fusion and self-adaptive control, and relates to the field of automatic welding, and the method comprises the following steps: collecting a molten pool area image, welding current and arc voltage signals in a welding process through a visual sensing module and an arc sensing module to obtain a synchronous sensing data set; a molten pool area image in the synchronous sensing data set is preprocessed, geometrical characteristic parameters of a molten pool are extracted, welding current and arc voltage signals in the synchronous sensing data set are analyzed, arc electric signal characteristics are extracted, and a multi-dimensional characteristic parameter set is obtained; and inputting the multi-dimensional characteristic parameter set into a pre-generated welding process cause and effect graph model for reasoning calculation, and outputting a welding seam tracking deviation and a penetration state index. Feedforward-feedback composite control over the welding process is achieved, and the welding quality, stability and consistency under the complex working condition are improved.

Description

Technical Field

[0001] This invention relates to the field of automated welding, and in particular to an intelligent welding system and method based on multi-information fusion and adaptive control. Background Technology

[0002] In the field of automated welding, multi-sensor information fusion is the mainstream technology for achieving adaptive control and improving welding quality stability. The camera acquires images of the molten pool to extract geometric features, while the sensors collect current and voltage signals to obtain electrical characteristics. Subsequently, the Kalman filter algorithm is used to fuse the above multi-source information to estimate the current weld tracking deviation and penetration state. Based on the estimation results, the welding parameters are adjusted by a fuzzy PID controller with preset fixed rules to achieve closed-loop control.

[0003] While existing technologies have improved process awareness to some extent, they lack the ability to predict the dynamic processes of welding. Welding is a complex process with significant heat accumulation and time-varying characteristics. Existing Kalman filter-based methods are essentially optimization estimates of current and historical states, and cannot effectively predict quality deviations that may be caused by slowly varying disturbances in the future. This results in control actions always lagging behind process changes, making it difficult to completely avoid overshoot, affecting the uniformity and consistency of weld formation. The adaptability of controllers with fixed rules is also limited when facing highly nonlinear and time-varying welding conditions. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent welding method based on multi-information fusion and adaptive control, which solves the problems of control lag and overshoot caused by the lack of forward-looking prediction ability of welding process in the prior art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an intelligent welding method based on multi-information fusion and adaptive control, which includes acquiring images of the molten pool area, welding current and arc voltage signals during the welding process through a visual sensing module and an arc sensing module to obtain a synchronous sensing data set.

[0008] The images of the molten pool region in the synchronous sensing data set are preprocessed and the geometric feature parameters of the molten pool are extracted. The welding current and arc voltage signals in the synchronous sensing data set are analyzed and the arc electrical signal features are extracted to obtain a multi-dimensional feature parameter set.

[0009] The multi-dimensional feature parameter set is input into the pre-generated causal graph model of the welding process for inference calculation, and the weld tracking deviation and penetration state index are output.

[0010] Driven by weld tracking deviation and penetration status indicators, an intelligent decision engine is used to simulate the evolution of welding status and predict potential quality deviations.

[0011] The potential quality deviation, along with the current actual weld tracking deviation and penetration status index, are input into an adaptive fuzzy PID controller to calculate and generate optimized adjustment instructions for welding current, welding speed, and wire feed speed.

[0012] The actuator executes optimization and adjustment commands to adjust the welding torch position and welding parameters.

[0013] As a preferred embodiment of the intelligent welding method based on multi-information fusion and adaptive control described in this invention, the method includes the following steps: acquiring images of the molten pool area, welding current, and arc voltage signals during the welding process through a visual sensing module and an arc sensing module to obtain a synchronous sensing data set.

[0014] The main control unit sends a unified hardware trigger pulse to the signal acquisition circuits of the CCD / CMOS camera in the vision sensing module and the arc sensing module.

[0015] The hardware trigger pulse drives the vision sensing module to acquire a frame of the molten pool area image, and simultaneously drives the arc sensing module to acquire a set of welding current and arc voltage signals aligned with the exposure time of the molten pool area image.

[0016] A frame of molten pool area image and a set of welding current and arc voltage signal values ​​aligned with it are packaged together and a high-precision timestamp is attached. The packaged molten pool area image, welding current and arc voltage signal data packet with timestamp is transmitted to the data buffer area of ​​the main control unit.

[0017] In the data buffer of the main control unit, multiple data packets are arranged in chronological order to form a time-correlated sequence of molten pool area image and a set of synchronous sensing data of welding current and arc voltage signal sequences.

[0018] As a preferred embodiment of the intelligent welding method based on multi-information fusion and adaptive control described in this invention, the method includes: preprocessing the molten pool region image in the synchronous sensing data set and extracting the geometric feature parameters of the molten pool; analyzing the welding current and arc voltage signals in the synchronous sensing data set and extracting arc electrical signal features to obtain a multi-dimensional feature parameter set, comprising the following steps:

[0019] A frame of molten pool region image is read from the synchronous sensing data set. Gaussian filtering is performed on the molten pool region image in sequence to suppress noise, and histogram equalization is performed to enhance contrast, resulting in a preprocessed molten pool region image.

[0020] The Canny edge detection algorithm is applied to the preprocessed molten pool region image to identify the molten pool contour, and the molten pool width is calculated based on the contour. The tail angle of the molten pool is calculated, which constitutes the geometric feature parameters of the molten pool.

[0021] Welding current and arc voltage signal sequences that are time-aligned with the image of the molten pool region are read from the synchronous sensing data set. The welding current and arc voltage signal sequences are low-pass filtered to remove high-frequency interference. The standard deviation of the welding current signal within a short time window is calculated, and the average value of the arc voltage signal within the same time window is calculated to form the arc electrical signal characteristics.

[0022] The geometric feature parameters of the molten pool and the electrical signal features of the electric arc are combined to form a multi-dimensional feature parameter set.

[0023] As a preferred embodiment of the intelligent welding method based on multi-information fusion and adaptive control described in this invention, the method includes the following steps: inputting a multi-dimensional feature parameter set into a pre-generated causal graph model of the welding process for inference calculation, and outputting weld tracking deviation and penetration state indicators.

[0024] By collecting historical welding data and analyzing it using a causal discovery algorithm, a causal graph model of the welding process is constructed and loaded into the memory of the main control unit.

[0025] The multi-dimensional feature parameter set is input into the loaded welding process causal graph model, and causal inference is performed on the multi-dimensional feature parameter set based on the internally defined causal relationship network.

[0026] The causal reasoning results quantify the lateral offset between the welding torch posture and the weld centerline, and calculate the weld tracking deviation.

[0027] The causal reasoning results are also used to comprehensively evaluate the degree of penetration, and a normalized penetration state index is calculated.

[0028] As a preferred embodiment of the intelligent welding method based on multi-information fusion and adaptive control described in this invention, the method includes the following steps: A smart decision engine is driven by weld tracking deviation and penetration state indicators to simulate the evolution of welding states and predict potential quality deviations.

[0029] By training a reinforcement learning agent on historical welding datasets, the evolution law of welding state and the optimal control strategy are learned, and the trained reinforcement learning agent is encapsulated as an intelligent decision engine.

[0030] Weld tracking deviation and penetration status indicators are input into the intelligent decision engine, which calls the internally integrated dynamic prediction model for the welding process.

[0031] The dynamic prediction model for the welding process uses weld tracking deviation and penetration state index as initial conditions to simulate the evolution of the welding state within a future time window;

[0032] Based on the results of the welding state evolution process, the predicted values ​​of potential weld width deviation and insufficient weld depth quality deviation are calculated;

[0033] As a preferred embodiment of the intelligent welding method based on multi-information fusion and adaptive control described in this invention, the method includes the following steps: inputting potential quality deviations, current actual weld tracking deviations, and penetration state indicators into an adaptive fuzzy PID controller to calculate and generate optimized adjustment instructions for welding current, welding speed, and wire feed speed.

[0034] The potential quality deviations, along with the current actual weld tracking deviations and penetration status indicators, are input into the input interface of the adaptive fuzzy PID controller.

[0035] The precise values ​​of weld tracking deviation, penetration status index, and potential quality deviation are converted into fuzzy linguistic variables through a fuzzy interface.

[0036] The logical relationship between the typical states of input variables such as weld tracking deviation and penetration status indicators and their corresponding ideal control actions is summarized and categorized into conditional rules in the form of IF...THEN..., forming a fuzzy rule base;

[0037] Inference is performed on fuzzy linguistic variables based on a fuzzy rule base to generate fuzzy control outputs. The fuzzy control outputs are then converted into precise adjustment values ​​through a defuzzification interface.

[0038] The precise adjustment values ​​are mapped to the adjustment values ​​of welding current, welding speed, and wire feed speed, respectively. The adaptive fuzzy PID controller outputs optimized adjustment commands for the adjustment values ​​of welding current, welding speed, and wire feed speed.

[0039] As a preferred embodiment of the intelligent welding method based on multi-information fusion and adaptive control described in this invention, the following steps are included: The welding torch posture and welding parameters are adjusted by executing optimization adjustment commands through an actuator:

[0040] The actuator receives a set of optimized adjustment instructions from the adaptive fuzzy PID controller. The actuator parses the set of optimized adjustment instructions and extracts the welding current adjustment, welding speed adjustment, and wire feed speed adjustment.

[0041] The actuator sends the welding current adjustment amount to the welding power source, which adjusts the output current according to the welding current adjustment amount. The actuator sends the welding speed adjustment amount to the robot motion controller, which adjusts the moving speed of the welding torch according to the welding speed adjustment amount.

[0042] The actuator sends the wire feeding speed adjustment amount to the wire feeder controller, which then adjusts the speed of the wire feeding motor according to the adjustment amount.

[0043] Secondly, the present invention provides an intelligent welding system based on multi-information fusion and adaptive control, including a data acquisition module, which acquires images of the molten pool area, welding current and arc voltage signals during the welding process through a visual sensing module and an arc sensing module, and obtains a set of synchronous sensing data.

[0044] The data processing module preprocesses the molten pool area image in the synchronous sensing data set and extracts the geometric feature parameters of the molten pool. It also analyzes the welding current and arc voltage signals in the synchronous sensing data set to extract arc electrical signal features and obtain a multi-dimensional feature parameter set.

[0045] The prediction module inputs a multi-dimensional feature parameter set into a pre-generated causal graph model of the welding process for inference calculation, and outputs weld tracking deviation and penetration status indicators.

[0046] The deviation module, driven by weld tracking deviation and penetration status indicators, simulates the evolution of welding status and predicts potential quality deviations.

[0047] The adjustment module inputs potential quality deviations, current actual weld tracking deviations, and penetration status indicators into an adaptive fuzzy PID controller to calculate and generate optimized adjustment instructions for welding current, welding speed, and wire feed speed.

[0048] The adjustment module executes optimization and adjustment commands through the actuator to adjust the welding torch position and welding parameters.

[0049] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the intelligent welding method based on multi-information fusion and adaptive control as described in the first aspect of the present invention.

[0050] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent welding method based on multi-information fusion and adaptive control as described in the first aspect of the present invention.

[0051] The beneficial effects of this invention are as follows: by using causal analysis rather than simple correlation, it outputs more accurate and reliable weld tracking deviation and penetration state indicators, realizing in-depth understanding and anti-interference diagnosis of the welding state; subsequently, these indicators are used to drive an intelligent decision engine, which simulates the future evolution of the welding state through its internal dynamic prediction model, and predicts potential quality deviations in a forward-looking manner. The predicted deviations and the current actual deviations are input together into an adaptive fuzzy PID controller to generate coordinated optimization adjustment instructions for welding current, welding speed and wire feed speed, which are then precisely executed by the actuator. This achieves feedforward-feedback composite control of the welding process, improving the welding quality, stability and consistency under complex working conditions. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart of an intelligent welding method based on multi-information fusion and adaptive control.

[0054] Figure 2 This is a schematic diagram of an intelligent welding system based on multi-information fusion and adaptive control.

[0055] Figure 3 This is a flowchart of the multi-dimensional feature parameter extraction process.

[0056] Figure 4 This is a timing diagram for synchronous data acquisition from multiple sensors. Detailed Implementation

[0057] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0058] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0059] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0060] Reference Figures 1-4 This is one embodiment of the present invention, which provides an intelligent welding method based on multi-information fusion and adaptive control, including the following steps:

[0061] S1. Acquire images of the molten pool area, welding current and arc voltage signals during the welding process through the vision sensing module and the arc sensing module to obtain a set of synchronous sensing data.

[0062] S1.1 The main control unit sends a unified hardware trigger pulse to the signal acquisition circuits of the CCD / CMOS camera in the vision sensing module and the arc sensing module.

[0063] Furthermore, the generation and distribution timing of pulses are precisely controlled by programmable logic devices to ensure that the synchronization accuracy of the pulse leading edge meets the requirements of multi-sensor collaborative sampling. The pulse signal is conditioned by isolation circuits and driving circuits to match the electrical interface specifications of the CCD / CMOS camera of the vision sensing module and the signal acquisition circuit of the arc sensing module.

[0064] S1.2 The hardware trigger pulse drives the vision sensing module to acquire a frame of the molten pool area image, and simultaneously drives the arc sensing module to acquire a set of welding current and arc voltage signals aligned with the exposure time of the molten pool area image.

[0065] Furthermore, for the vision sensing module, the trigger pulse directly controls the start time of exposure of the CCD / CMOS camera, ensuring that the acquired frame of the molten pool area image is the light field information at the moment the pulse arrives. For the arc sensing module, the trigger pulse serves as the start signal of the signal acquisition circuit, controlling the analog-to-digital converter to start continuously sampling the welding current and arc voltage signals, and ensuring that the time of the first sampling point is aligned with the middle time of the camera exposure.

[0066] S1.3 Pack a frame of molten pool area image and a set of welding current and arc voltage signal values ​​that are time-aligned with it into a package, and attach a high-precision timestamp. Transmit the packaged molten pool area image, welding current and arc voltage signal data package with timestamp to the data buffer area of ​​the main control unit.

[0067] Furthermore, the specific implementation of data packaging and timestamp appending relies on an efficient data encapsulation protocol. Within each acquisition cycle, a pixel data block representing a frame of molten pool area image at the same physical moment and a corresponding array of sampled values ​​of welding current and arc voltage signals are included. The header of the data packet is embedded with a high-resolution timestamp provided by a high-precision clock source.

[0068] S1.4 In the data buffer area of ​​the main control unit, multiple consecutive data packets are arranged in chronological order to form a time-correlated sequence of molten pool area image and a set of synchronous sensing data of welding current and arc voltage signal sequences.

[0069] Furthermore, by implementing a first-in-first-out queue management mechanism in the data buffer of the main control unit, continuously arriving data packets with timestamps are arranged in the buffer according to the order of their timestamps, forming a time-sequential data stream. This data stream includes a frame-sequential sequence of molten pool area images and a time-sequential sequence of welding current and arc voltage signals. Each frame in the image sequence has a strict correspondence with a certain segment of data in the signal sequence in terms of timestamps. The deep management of the buffer and the data overflow handling strategy ensure that no data packets are lost under high-speed data streams. The purpose of this step is to organize discrete data packets into a structured, intrinsically related set of synchronous sensing data in the time dimension.

[0070] S2. Preprocess the molten pool area image in the synchronous sensing data set and extract the geometric feature parameters of the molten pool. Analyze the welding current and arc voltage signals in the synchronous sensing data set to extract the arc electrical signal features and obtain a multi-dimensional feature parameter set.

[0071] S2.1 Read a frame of molten pool region image from the synchronous sensing data set, perform Gaussian filtering on the molten pool region image in sequence to suppress noise, and perform histogram equalization to enhance contrast, to obtain the preprocessed molten pool region image.

[0072] Furthermore, a weighted average of each pixel's neighborhood is applied using a weighted template generated by a Gaussian function. This effectively suppresses high-frequency random noise caused by arc flicker and inherent sensor noise, smoothing out noise while preserving molten pool edge information. The histogram equalization algorithm performs non-linear stretching of the image's grayscale distribution, redistributing pixel intensity values ​​to make the grayscale histogram distribution more uniform. This enhances the contrast between the molten pool region and the background, making the previously blurred molten pool boundary clearly discernible. The purpose of this combined processing method is to improve image quality, ultimately yielding a preprocessed image of the molten pool region.

[0073] S2.2 Apply the Canny edge detection algorithm to the preprocessed molten pool region image to identify the molten pool contour, calculate the molten pool width based on the contour, and calculate the tail angle of the molten pool, i.e., the rear angle, to form the geometric feature parameters of the molten pool.

[0074] Furthermore, by traversing the coordinates of each row of the image, the difference between the maximum and minimum x-coordinates of the contour points in that row is calculated as the width of the molten pool at the y-row. The maximum width of all rows is then calculated to obtain the molten pool width. When calculating the tail angle of the molten pool, the least-squares line fitting is performed on the contour point sets on the left and right sides of the tail to obtain the slopes of the fitted lines on the left and right sides. Then, the arctangent function is used to calculate the angle between the two lines. These geometric feature parameters together constitute the geometric feature parameters of the molten pool.

[0075] The expression for melt width is:

[0076] ;

[0077] in, For the width of the melt, For the molten pool in The width of the row, These are the row coordinates of the image.

[0078] The geometric feature parameter expression is:

[0079] ;

[0080] in, The angle at the rear of the molten pool. The slope of the straight line fitted to the left contour. The slope of the straight line fitted to the right contour.

[0081] S2.3 Read the welding current and arc voltage signal sequences that are time-aligned with the image of the molten pool area from the synchronous sensing data set. Perform low-pass filtering on the welding current and arc voltage signal sequences to remove high-frequency interference. Calculate the standard deviation of the welding current signal within a short time window and the average value of the arc voltage signal within the same time window to form the arc electrical signal characteristics.

[0082] Furthermore, a low-pass filter is used to filter out high-frequency electromagnetic interference. When calculating the standard deviation of the welding current signal within a short time window, the arithmetic mean of all welding current samples within the window is first calculated. Then, the square root of the sum of squares of the differences between each sample and the average value is calculated. The average value of the arc voltage signal within the same time window is calculated by directly taking the arithmetic mean of the sample value sequence. The standard deviation of the welding current characterizes the stability of the arc, and the average value of the arc voltage reflects the arc length. These statistics together constitute the characteristics of the arc electrical signal.

[0083] The expression for the average value of the welding current sampling sequence is:

[0084] ;

[0085] in, This represents the average welding current within the time window. This represents the total number of sampling points within a short time window. For sampling points, For the first time window The sampled value of the welding current,

[0086] ;

[0087] in, The standard deviation of the welding current;

[0088] The average value of the arc voltage is expressed as:

[0089] ;

[0090] in, This represents the average value of the arc voltage within the time window. For the first time window The sampled values ​​of the arc voltage.

[0091] S2.4. Combine the geometric characteristic parameters of the molten pool with the electrical signal characteristics of the arc to form a multi-dimensional characteristic parameter set.

[0092] Furthermore, the operation of merging the geometric feature parameters of the molten pool and the electrical signal features of the arc to form a multi-dimensional feature parameter set is essentially a splicing and fusion of feature vectors from different physical domains. Visual features such as weld width and back angle of the molten pool, and electrical features such as the standard deviation of welding current and the average value of arc voltage, are organized into a unified multi-dimensional vector. The formation of the multi-dimensional feature parameter set allows subsequent processing to utilize both visual morphological information and dynamic characteristics of electrical signals simultaneously.

[0093] S3. Input the multi-dimensional feature parameter set into the pre-generated welding process cause-effect graph model for inference calculation, and output the weld tracking deviation and penetration status index.

[0094] S3.1 By collecting historical welding data, analyzing it using a causal discovery algorithm, constructing a causal graph model of the welding process, and loading the causal graph model of the welding process into the memory of the main control unit.

[0095] Furthermore, causal discovery algorithms are used to automate the analysis of historical datasets, aiming to identify and determine the causal relationship network between welding parameters, sensing features, and quality results, rather than simple statistical correlations, thereby constructing a causal graph model of the welding process.

[0096] It should be noted that the causal graph model of the welding process is persistently stored in non-volatile memory and loaded into the memory of the main control unit at the start of the welding task for high-speed access. This provides a pre-built computational framework of the inherent causal logic of the welding process for real-time causal inference, and places the complex construction and learning process of the causal graph model of the welding process offline, ensuring the computational efficiency of the real-time control loop and enabling the causal graph model of the welding process to reside in memory for later use.

[0097] S3.2 Input the multi-dimensional feature parameter set into the loaded welding process causal graph model, and perform causal reasoning on the multi-dimensional feature parameter set based on the internally defined causal relationship network.

[0098] Furthermore, calculations are carried out within the theoretical framework of graph models. After the multi-dimensional feature parameter set is input into the causal graph model of the welding process as observational evidence, the inference process is carried out based on the causal relationship network represented by the directed edges between the node variables defined in the model. The inference calculation involves probability propagation or intervention effect assessment based on do-calculus. By considering the conditional independence between variables and the specific conditions set by the intervention operation, the posterior distribution or expected value of other variables under the given observational evidence is calculated.

[0099] It should be noted that using causal logic rather than simple correlation to infer state quantities that cannot be directly measured can more accurately estimate the true process state, effectively avoid misjudgments caused by confounding factors, and ultimately complete an in-depth interpretation of multi-dimensional feature parameter sets based on causal relationship networks.

[0100] S3.3. The causal reasoning results quantify the lateral offset between the welding torch posture and the weld centerline, and calculate the weld tracking deviation.

[0101] Furthermore, given the current observational evidence, i.e., the multi-dimensional feature parameter set, and after applying intervention operations to fix certain controllable variables in the model, the expected value of the potential outcome variable, i.e., the true lateral offset, is calculated.

[0102] It should be noted that by transforming the mathematical results of causal reasoning into a weld seam tracking deviation with clear physical meaning, and by isolating confounding effects through causal intervention, the estimated deviation is closer to the actual offset caused by the single cause of the welding torch pose, and a more reliable weld seam tracking deviation value is obtained for control.

[0103] The expression for weld seam tracking deviation is:

[0104] ;

[0105] in, For weld seam tracking deviation, For expectation operator, This is the lateral offset. For intervention operations, For observational evidence;

[0106] S3.4 The causal reasoning results are also used to comprehensively evaluate the degree of penetration and calculate a normalized penetration state index.

[0107] Furthermore, the calculation of weld tracking deviation shares the same causal reasoning framework but focuses on different target variables. The calculation expression for the penetration state index is as follows: This means that, under the same intervention conditions and observational evidence, the expected value of another potential outcome variable, namely the true melt penetration state, is calculated. The expected value is then mapped to a preset normalization range, such as between zero and one, thereby generating a melt penetration state index. This yields a standardized, dimensionless quantitative index that comprehensively reflects the melt penetration quality. This index integrates the causal contributions of multi-dimensional information, providing a more robust melt penetration state assessment than a single sensor signal, and outputting a normalized melt penetration state index that is easy for the controller to understand and use.

[0108] The expression for the penetration state index is:

[0109] ;

[0110] in, As an indicator of the melt penetration state, It is in a fully melted state;

[0111] S4. Based on weld tracking deviation and penetration status indicators, an intelligent decision engine is driven to simulate the evolution of welding status and predict potential quality deviations.

[0112] S4.1. Train a reinforcement learning agent on historical welding datasets to learn the evolution law of welding state and the optimal control strategy, and encapsulate the trained reinforcement learning agent into an intelligent decision engine.

[0113] Furthermore, a Markov decision process framework is adopted, modeling the evolution of welding state as state transitions, using control actions as decisions, and welding quality indicators as reward signals. The reinforcement learning agent gradually approaches the optimal control policy function through trial and error learning in interaction with the environment, and can map the state space to the action space. The trained reinforcement learning agent is encapsulated as a software component with a standard interface, namely the intelligent decision engine, which contains the learned state evolution laws and control strategies. This gives the decision engine the ability to autonomously learn complex nonlinear process dynamics from data, avoiding the dependence of traditional methods on precise physical models, adapting to changing working conditions, and forming an intelligent decision with empirical knowledge.

[0114] S4.2 Input the weld tracking deviation and penetration status indicators into the intelligent decision engine and call the internally integrated dynamic prediction model of the welding process.

[0115] Furthermore, after the weld tracking deviation and penetration status indicators are input into the intelligent decision engine, the specific implementation of the dynamic prediction model of the welding process called by the engine follows the principle of modular design. The intelligent decision engine provides a standardized data input interface to receive the weld tracking deviation and penetration status indicators calculated by the previous step. The engine maintains an instance of the dynamic prediction model of the welding process as a functional component of the engine. After receiving new status indicators, the input interface triggers the internal scheduling logic to transmit the data to the calculation unit of the dynamic prediction model of the welding process.

[0116] S4.3 The dynamic prediction model for the welding process uses weld tracking deviation and penetration state index as initial conditions to simulate the evolution of the welding state within a future time window.

[0117] Furthermore, weld tracking deviation and penetration state indicators together constitute a quantitative description of the current welding state, namely the initial state vector of the welding process dynamic prediction model. Based on its embedded dynamic laws, molten pool fluid dynamics relationship, or data-driven state transition function, the welding process dynamic prediction model extrapolates the state change trajectory over a future period of time by numerical integration. In the extrapolation process, it is usually assumed that the control input sequence remains unchanged or changes according to a certain known law within the prediction time window, predicting the dynamic behavior of the welding process over time, extending the static state assessment to dynamic trend prediction, providing a basis for forward-looking control, and obtaining the continuous evolution trajectory of the welding state within the future time window.

[0118] S4.4 Based on the results of the welding state evolution process, calculate the predicted values ​​of potential weld width deviation and insufficient weld depth quality deviation.

[0119] It should be noted that the state evolution trajectory output by the dynamic prediction model of the welding process contains a series of state vectors for future moments. For each predicted state at a future moment, it is converted into the corresponding quality parameter value through a predefined quality mapping function. The specific combination of state variables is mapped into the predicted weld width and predicted penetration state index. These predicted values ​​are compared with the preset target values, and the difference is calculated to obtain the predicted weld width deviation and the predicted penetration state deviation.

[0120] The predicted values ​​for width deviation and insufficient melting depth quality deviation are expressed as follows:

[0121] ;

[0122] in, For the predicted melt width deviation, For the predicted melt penetration deviation, For dynamic prediction models of the welding process, This is the current welding status. To predict the time, In the time interval The control input sequence of the internal hypothesis.

[0123] S5. Input the potential quality deviation, the current actual weld tracking deviation, and the penetration status index into the adaptive fuzzy PID controller to calculate and generate optimized adjustment instructions for welding current, welding speed, and wire feed speed.

[0124] S5.1 Input the potential quality deviation and the current actual weld tracking deviation and penetration status index into the input interface of the adaptive fuzzy PID controller.

[0125] Furthermore, when inputting the potential quality deviations along with the current actual weld tracking deviations and penetration status indicators into the input interface of the adaptive fuzzy PID controller, a multi-channel data receiving mechanism needs to be established. The input interface adopts a parallel buffer queue structure to receive real-time weld tracking deviations and penetration status indicators from the causal inference module, as well as potential quality deviation data from the prediction module.

[0126] It should be noted that each input channel is equipped with data validity verification logic to ensure the physical rationality of the input value and the legality of its numerical range. The data synchronization mechanism ensures that input values ​​from different sources are aligned in timestamps, avoiding control decision distortion due to differences in data arrival times. This provides the controller with comprehensive and synchronized status information, integrating real-time status and future trends, laying the foundation for generating optimized instructions that coordinate multiple objectives, and completing the reliable reception and temporary storage of multi-source input data.

[0127] S5.2 Convert the precise values ​​of weld tracking deviation, penetration status index and potential quality deviation into fuzzy linguistic variables through the fuzzification interface.

[0128] Furthermore, the process of converting precise values ​​into fuzzy linguistic variables through the fuzzification interface is essentially mapping precise numerical inputs to fuzzy sets of natural language descriptions. The fuzzification interface defines a set of fuzzy sets and their membership functions for each input variable, such as weld seam tracking deviation, converting a single numerical value into a membership vector of multiple fuzzy linguistic variables. This conversion from precise to fuzzy quantities enables the controller to handle uncertainties and imprecise expressions, creating a premise for language rule-based reasoning and realizing the mapping of input variables from the numerical domain to the linguistic domain.

[0129] S5.3. Summarize and generalize the logical relationships between typical states of input variables such as weld tracking deviation and penetration status indicators and their corresponding ideal control actions into conditional rules of the form IF...THEN... to form a fuzzy rule base.

[0130] Furthermore, the construction of the fuzzy rule base is a formalized expression process based on welding domain knowledge and historical operational experience. The rule base consists of multiple IF-THEN conditional rules. The IF part defines specific combinations of input fuzzy linguistic variables, and the THEN part specifies the corresponding output fuzzy linguistic variables. The rule base needs to cover all possible combinations of input states and reflect the welding process experts' understanding of control strategies for complex working conditions. The rule-based knowledge representation method enables the controller to simulate the decision-making thinking of human experts, transforming implicit experiential knowledge into explicit executable logic, forming a complete set of fuzzy control rules.

[0131] S5.4. Based on the fuzzy rule base, reason about fuzzy linguistic variables to generate fuzzy control output, and convert the fuzzy control output into precise adjustment values ​​through the defuzzification interface.

[0132] Furthermore, the inference process based on the fuzzy rule base adopts the Mamdani-type fuzzy inference method. Fuzzy control output is obtained through synthesis operation. The inference engine evaluates the antecedent matching degree of all fuzzy rules in parallel, realizes the implication relationship between the antecedent and consequent of the rule through minimum operation, and then aggregates the fuzzy set of consequents of all activated rules through maximum operation. The defuzzification interface adopts algorithms such as the centroid method or the maximum membership method to convert the aggregated fuzzy output set into accurate adjustment value. The inference mechanism can handle the complex situation of multiple rules being activated at the same time, realize nonlinear mapping while maintaining the smoothness of control, and finally obtain accurate control adjustment value.

[0133] S5.5 The precise adjustment values ​​are mapped to the adjustment values ​​of welding current, welding speed and wire feed speed respectively. The adaptive fuzzy PID controller outputs optimized adjustment instructions for the adjustment values ​​of welding current, welding speed and wire feed speed.

[0134] Furthermore, a scaling transformation relationship is established from control quantities to execution parameters. The mapping module, based on the characteristic parameters of the welding power supply, robot motion controller, and wire feeder, proportionally converts the unified control quantity adjustment value into the corresponding physical quantity adjustment range for each actuator. The mapping relationship must consider the dynamic response characteristics and operating range limitations of each actuator to ensure that the generated adjustment commands are executable on the actual equipment. The channel-specific command mapping realizes the transformation from control strategy to specific equipment actions, ensuring the synchronization and consistency of multi-actuator collaborative work, and outputting a complete set of optimized adjustment commands.

[0135] S6. The actuator executes optimization and adjustment commands to adjust the welding torch position and welding parameters.

[0136] S6.1 The actuator receives the set of optimization adjustment instructions from the adaptive fuzzy PID controller. The actuator parses the set of optimization adjustment instructions and extracts the welding current adjustment amount, welding speed adjustment amount, and wire feed speed adjustment amount.

[0137] Furthermore, after receiving the set of optimization adjustment instructions from the adaptive fuzzy PID controller, the actuator uses an instruction decoding and data extraction mechanism. The set of optimization adjustment instructions is usually encapsulated in a structured data format. The actuator first parses the data packets through the communication protocol to verify the integrity and validity of the instructions.

[0138] It should be noted that, based on the preset instruction structure definition, the parsing logic identifies and separates three independent parameter adjustment values ​​from the set: welding current adjustment, welding speed adjustment, and wire feed speed adjustment. Each adjustment value is assigned a specific identifier and value to ensure that it can be accurately distributed to the corresponding execution unit. This deconstructs the composite instruction into independent equipment control commands, ensuring the synchronization and consistency of multi-parameter adjustment instructions, laying the foundation for subsequent accurate execution, and ultimately completing the accurate extraction from the composite instruction to the independent adjustment values.

[0139] S6.2 The actuator sends the welding current adjustment amount to the welding power source, which adjusts the output current according to the welding current adjustment amount. The actuator sends the welding speed adjustment amount to the robot motion controller, which adjusts the moving speed of the welding torch according to the welding speed adjustment amount.

[0140] Furthermore, the actuator distributes the analyzed welding current and welding speed adjustments to the corresponding equipment controllers. The welding current adjustment is sent to the welding power source via analog signals or a digital communication bus. The control loop inside the welding power source corrects the set value of the output current in real time based on the received adjustment, thereby changing the arc heat input. The welding speed adjustment is sent to the robot motion controller via a motion control network. The robot motion controller replans the movement trajectory and speed of the welding torch based on this adjustment and adjusts the motion parameters of the servo motor. This achieves coordinated control of the welding energy input and the welding torch movement speed, ensuring the matching of heat input and welding speed, avoiding weld formation defects caused by parameter mismatch, and completing the precise adjustment of welding current and welding torch movement speed.

[0141] S6.3 The actuator sends the wire feeding speed adjustment amount to the wire feeder controller, and the wire feeder controller adjusts the speed of the wire feeding motor according to the wire feeding speed adjustment amount.

[0142] Furthermore, the actuator sends the wire feeding speed adjustment to the wire feeder controller to adjust the wire feeding motor speed. The wire feeding speed adjustment is transmitted to the wire feeder controller through a dedicated interface. The controller changes the drive signal of the wire feeding servo motor according to the adjustment, thereby precisely controlling the wire feeding speed. The adjustment of the wire feeding speed directly changes the deposition rate of the filler metal, which needs to be dynamically coordinated with the changes in welding current and welding speed to maintain a stable arc process and weld formation, and precisely control the amount of deposited metal in the weld. By maintaining the matching of the wire feeding speed with other parameters, the stability of the welding process and the uniformity of weld quality are ensured, ultimately achieving precise control of the wire feeding speed.

[0143] This embodiment also provides an intelligent welding system based on multi-information fusion and adaptive control, including: a data acquisition module, which acquires images of the molten pool area, welding current and arc voltage signals during the welding process through a visual sensing module and an arc sensing module, and obtains a set of synchronous sensing data;

[0144] The data processing module preprocesses the molten pool area image in the synchronous sensing data set and extracts the geometric feature parameters of the molten pool. It also analyzes the welding current and arc voltage signals in the synchronous sensing data set to extract arc electrical signal features and obtain a multi-dimensional feature parameter set.

[0145] The prediction module inputs a multi-dimensional feature parameter set into a pre-generated causal graph model of the welding process for inference calculation, and outputs weld tracking deviation and penetration status indicators.

[0146] The deviation module, driven by weld tracking deviation and penetration status indicators, simulates the evolution of welding status and predicts potential quality deviations.

[0147] The adjustment module inputs potential quality deviations, current actual weld tracking deviations, and penetration status indicators into an adaptive fuzzy PID controller to calculate and generate optimized adjustment instructions for welding current, welding speed, and wire feed speed.

[0148] The adjustment module executes optimization and adjustment commands through the actuator to adjust the welding torch position and welding parameters.

[0149] This embodiment also provides a computer device applicable to the intelligent welding method based on multi-information fusion and adaptive control, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the intelligent welding method based on multi-information fusion and adaptive control as proposed in the above embodiment.

[0150] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0151] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the intelligent welding method based on multi-information fusion and adaptive control as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0152] In summary, this invention outputs more accurate and reliable weld tracking deviation and penetration status indicators through causal analysis rather than simple correlation, achieving in-depth understanding and anti-interference diagnosis of the welding state. Subsequently, these indicators drive an intelligent decision engine, which simulates the future evolution of the welding state through its internal dynamic prediction model, proactively predicting potential quality deviations. The predicted deviations and the current actual deviations are input together into an adaptive fuzzy PID controller to generate coordinated optimization adjustment instructions for welding current, welding speed, and wire feed speed, which are then precisely executed by the actuator. This achieves feedforward-feedback composite control of the welding process, improving welding quality, stability, and consistency under complex working conditions.

[0153] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An intelligent welding method based on multi-information fusion and adaptive control, characterized in that: This includes acquiring images of the molten pool area, welding current, and arc voltage signals during the welding process through a vision sensing module and an arc sensing module, to obtain a set of synchronous sensing data. The images of the molten pool region in the synchronous sensing data set are preprocessed and the geometric feature parameters of the molten pool are extracted. The welding current and arc voltage signals in the synchronous sensing data set are analyzed and the arc electrical signal features are extracted to obtain a multi-dimensional feature parameter set. The multi-dimensional feature parameter set is input into the pre-generated causal graph model of the welding process for inference calculation, and the weld tracking deviation and penetration state index are output. Driven by weld tracking deviation and penetration status indicators, an intelligent decision engine is used to simulate the evolution of welding status and predict potential quality deviations. The potential quality deviation, along with the current actual weld tracking deviation and penetration status index, are input into an adaptive fuzzy PID controller to calculate and generate optimized adjustment instructions for welding current, welding speed, and wire feed speed. The actuator executes optimization and adjustment commands to adjust the welding torch position and welding parameters.

2. The intelligent welding method based on multi-information fusion and adaptive control as described in claim 1, characterized in that: The welding process involves acquiring images of the molten pool area, welding current, and arc voltage signals through a visual sensing module and an arc sensing module to obtain a synchronous sensing data set, including the following steps: The main control unit sends a unified hardware trigger pulse to the signal acquisition circuits of the CCD / CMOS camera in the vision sensing module and the arc sensing module. The hardware trigger pulse drives the vision sensing module to acquire a frame of the molten pool area image, and simultaneously drives the arc sensing module to acquire a set of welding current and arc voltage signals aligned with the exposure time of the molten pool area image. A frame of molten pool area image and a set of welding current and arc voltage signal values ​​aligned with it are packaged together and a high-precision timestamp is attached. The packaged molten pool area image, welding current and arc voltage signal data packet with timestamp is transmitted to the data buffer area of ​​the main control unit. In the data buffer of the main control unit, multiple data packets are arranged in chronological order to form a time-correlated sequence of molten pool area image and a set of synchronous sensing data of welding current and arc voltage signal sequences.

3. The intelligent welding method based on multi-information fusion and adaptive control as described in claim 2, characterized in that: The process involves preprocessing the molten pool region image from the synchronous sensing dataset and extracting its geometric feature parameters. It also involves analyzing the welding current and arc voltage signals from the synchronous sensing dataset to extract arc electrical signal features, thereby obtaining a multi-dimensional feature parameter set. This includes the following steps: A frame of molten pool region image is read from the synchronous sensing data set. Gaussian filtering is performed on the molten pool region image in sequence to suppress noise, and histogram equalization is performed to enhance contrast, resulting in a preprocessed molten pool region image. The Canny edge detection algorithm is applied to the preprocessed molten pool region image to identify the molten pool contour, and the molten pool width is calculated based on the contour. The tail angle of the molten pool is calculated, which constitutes the geometric features of the molten pool. Welding current and arc voltage signal sequences that are time-aligned with the image of the molten pool region are read from the synchronous sensing data set. The welding current and arc voltage signal sequences are low-pass filtered to remove high-frequency interference. The standard deviation of the welding current signal within a short time window is calculated, and the average value of the arc voltage signal within the same time window is calculated to form the arc electrical signal characteristics. The geometric feature parameters of the molten pool and the electrical signal features of the electric arc are combined to form a multi-dimensional feature parameter set.

4. The intelligent welding method based on multi-information fusion and adaptive control as described in claim 3, characterized in that: The multi-dimensional feature parameter set is input into a pre-generated causal graph model of the welding process for inference calculation, and the output weld tracking deviation and penetration state index are included in the following steps: By collecting historical welding data and analyzing it using a causal discovery algorithm, a causal graph model of the welding process is constructed and loaded into the memory of the main control unit. The multi-dimensional feature parameter set is input into the loaded welding process causal graph model, and causal inference is performed on the multi-dimensional feature parameter set based on the internally defined causal relationship network. The causal reasoning results quantify the lateral offset between the welding torch posture and the weld centerline, and calculate the weld tracking deviation. The causal reasoning results are also used to comprehensively evaluate the degree of penetration, and a normalized penetration state index is calculated.

5. The intelligent welding method based on multi-information fusion and adaptive control as described in claim 4, characterized in that: Driven by weld tracking deviation and penetration status indicators, an intelligent decision engine simulates the evolution of welding conditions and predicts potential quality deviations, including the following steps: By training a reinforcement learning agent on historical welding datasets, the evolution law of welding state and the optimal control strategy are learned, and the trained reinforcement learning agent is encapsulated as an intelligent decision engine. Weld tracking deviation and penetration status indicators are input into the intelligent decision engine, which calls the internally integrated dynamic prediction model for the welding process. The dynamic prediction model for the welding process uses weld tracking deviation and penetration state index as initial conditions to simulate the evolution of the welding state within a future time window; Based on the results of the welding state evolution process, the predicted values ​​of potential weld width deviation and insufficient weld depth quality deviation are calculated.

6. The intelligent welding method based on multi-information fusion and adaptive control as described in claim 5, characterized in that: The potential quality deviation, along with the current actual weld tracking deviation and penetration status index, are input into an adaptive fuzzy PID controller to calculate and generate optimized adjustment instructions for welding current, welding speed, and wire feed speed. This includes the following steps: The potential quality deviations, along with the current actual weld tracking deviations and penetration status indicators, are input into the input interface of the adaptive fuzzy PID controller. The precise values ​​of weld tracking deviation, penetration status index, and potential quality deviation are converted into fuzzy linguistic variables through a fuzzy interface. The logical relationship between the typical states of input variables such as weld tracking deviation and penetration status indicators and their corresponding ideal control actions is summarized and categorized into conditional rules in the form of IF...THEN..., forming a fuzzy rule base; Inference is performed on fuzzy linguistic variables based on a fuzzy rule base to generate fuzzy control outputs. The fuzzy control outputs are then converted into precise adjustment values ​​through a defuzzification interface. The precise adjustment values ​​are mapped to the adjustment values ​​of welding current, welding speed, and wire feed speed, respectively. The adaptive fuzzy PID controller outputs optimized adjustment commands for the adjustment values ​​of welding current, welding speed, and wire feed speed.

7. The intelligent welding method based on multi-information fusion and adaptive control as described in claim 6, characterized in that: The welding torch position and welding parameters are adjusted by executing optimization and adjustment commands through the actuator, including the following steps: The actuator receives a set of optimized adjustment instructions from the adaptive fuzzy PID controller. The actuator parses the set of optimized adjustment instructions and extracts the welding current adjustment, welding speed adjustment, and wire feed speed adjustment. The actuator sends the welding current adjustment amount to the welding power source, which adjusts the output current according to the welding current adjustment amount. The actuator sends the welding speed adjustment amount to the robot motion controller, which adjusts the moving speed of the welding torch according to the welding speed adjustment amount. The actuator sends the wire feeding speed adjustment amount to the wire feeder controller, which then adjusts the speed of the wire feeding motor according to the adjustment amount.

8. An intelligent welding system based on multi-information fusion and adaptive control, based on the intelligent welding method based on multi-information fusion and adaptive control as described in any one of claims 1 to 7, characterized in that: This includes a data acquisition module, which acquires images of the molten pool area, welding current and arc voltage signals during the welding process through a vision sensing module and an arc sensing module, and obtains a set of synchronous sensing data. The data processing module preprocesses the molten pool area image in the synchronous sensing data set and extracts the geometric feature parameters of the molten pool. It also analyzes the welding current and arc voltage signals in the synchronous sensing data set to extract arc electrical signal features and obtain a multi-dimensional feature parameter set. The prediction module inputs a multi-dimensional feature parameter set into a pre-generated causal graph model of the welding process for inference calculation, and outputs weld tracking deviation and penetration status indicators. The deviation module, driven by weld tracking deviation and penetration status indicators, simulates the evolution of welding status and predicts potential quality deviations. The adjustment module inputs potential quality deviations, current actual weld tracking deviations, and penetration status indicators into an adaptive fuzzy PID controller to calculate and generate optimized adjustment instructions for welding current, welding speed, and wire feed speed. The adjustment module executes optimization and adjustment commands through the actuator to adjust the welding torch position and welding parameters.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the intelligent welding method based on multi-information fusion and adaptive control as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the intelligent welding method based on multi-information fusion and adaptive control as described in any one of claims 1 to 7.

Images