CN-121988828-A - Arc welding robot and arc welding method for non-structural environment

CN121988828ACN 121988828 ACN121988828 ACN 121988828ACN-121988828-A

Abstract

The invention discloses an arc welding robot and an arc welding method used in a non-structural environment, and belongs to the technical field of arc welding. The method comprises the steps of synchronously preprocessing multi-source signal information, constructing a multi-module sensing matrix, obtaining a processed multi-module sensing matrix, extracting weld joint characteristics and welding process state characteristics based on the processed multi-module sensing matrix, planning a self-adaptive welding track and initializing a welding process parameter set by taking the weld joint characteristics as tracking targets, creating a self-adaptive swing tracking algorithm based on reinforcement learning, continuously optimizing control parameters, dynamically adjusting the welding process parameter set according to the optimized control parameters, synchronously collecting real-time multi-source feedback signals in the welding process, and constructing a welding process closed-loop control system. The invention thoroughly abandons the traditional teaching-reappearance fixed track control mode, constructs a dynamic regulation and control system adapting to the non-structural environment, solves the problem of poor adaptability of the traditional arc welding robot to dynamic interference, and meets the welding requirement of complex welding seams of the battery box of the new energy automobile.

Inventors

HUANG YILIN
CHENG CHANG
WU ZHITAO
ZHANG CHENGYU
XU JIALIN
CHEN JUN
ZHANG CHAOHAN
CHEN FEIXING
WU FAN
JIANG MENGLONG

Assignees

闽南理工学院

Dates

Publication Date: 20260508
Application Date: 20260311

Claims (10)

1. A method of arc welding for use in a non-structural environment, comprising the steps of: Collecting multi-source signal information of a welding area in a non-structural environment, carrying out synchronous pretreatment on the multi-source signal information and constructing a multi-module sensing matrix ; For the multi-module sensing matrix The data in the multi-module sensing matrix is subjected to space-time alignment and noise reduction treatment to obtain a multi-module sensing matrix after treatment Based on the processed multi-module sensing matrix Extracting weld joint characteristics and welding process state characteristics; The weld joint characteristics are used as tracking targets, self-adaptive welding tracks are planned, and a welding process parameter set is initialized; and controlling an arc welding executing mechanism to carry out welding operation based on the adjusted control parameters and welding process parameter groups, synchronously collecting real-time multisource feedback signals in the welding process, constructing a closed-loop control system of the welding process, and realizing dynamic regulation and control of arc welding operation in a non-structural environment.
2. The method of arc welding for use in a non-structural environment according to claim 1, wherein the multi-source signal information includes at least visual, arc and voiceprint signals of the welding area; the multi-module sensing matrix The construction flow of (2) is as follows: Arc filtering, image enhancement and downsampling are sequentially carried out on the visual signals, filtering denoising, amplitude calibration and synchronous sampling are sequentially carried out on the arc signals, noise reduction and amplification, frequency domain interception and synchronous sampling are sequentially carried out on the voiceprint signals, and therefore the acquisition time sequence and data dimension of the visual signals, the arc signals and the voiceprint signals are kept uniform; After the synchronization pretreatment, respectively establishing a visual signal two-dimensional matrix Two-dimensional matrix of arc signals Sum voiceprint signal two-dimensional matrix , wherein, For a single frame of pixel dimensions, In order to be in the dimension of the time series, As a dimension of the voiceprint feature, Is the voiceprint feature dimension; the multi-module sensing matrix The expression form of (a) is as follows: 。
3. An arc welding method for use in a non-structural environment according to claim 1, wherein said post-processing multi-module sensing matrix The acquisition process of (a) is as follows: For multi-module sensing matrix Firstly, carrying out noise reduction treatment on the time sequence dimension by adopting Kalman filtering, inhibiting systematic noise caused by welding arc light and mechanical arm vibration, and then denoising and filtering random noise introduced by sensing acquisition and environmental interference through a wavelet threshold value to obtain a multi-module sensing matrix after noise reduction; And (3) taking an acquisition time sequence of the electric arc signal as a time reference axis, calculating time sequence deviation of the visual signal, the voiceprint signal and the electric arc signal through cross-correlation analysis, completing time sequence complement calibration by adopting a cubic spline interpolation method, and simultaneously carrying out space normalization mapping on pixel coordinates of the visual signal, an electric arc sensing acquisition position and a voiceprint sensing acquisition position by taking the center of a welding seam molten pool as a space reference, so as to realize space-time double-dimensional alignment of the multi-source signal and obtain a multi-module sensing matrix after space-time alignment.
4. An arc welding method for use in a non-structural environment according to claim 1, wherein the creation flow of the adaptive wobble tracking algorithm is as follows: Constructing a state space model based on the weld features, wherein the state space model is expressed in the following form: wherein, the method comprises the steps of, Is that A 5-dimensional state vector of the moment in time, Is that A 5-dimensional state transition matrix is provided, Is that The dimensions control the input matrix to be displayed, Is that The 3-dimensional control parameter vector of the moment in time, Is that The 1-dimensional process reduces the noise column vector, To be used for State vector of time of day Constructing a reinforcement learning decision network for input, and outputting a control parameter set consisting of a swing amplitude adjustment amount, a swing frequency adjustment amount and a swing track correction coefficient; The multi-objective rewarding function is utilized, the minimum track deviation, stable swing control and standard reaching of the fusion width and penetration depth are used as optimization targets, and the optimal parameters are selected from the control parameter set to be used as optimization targets Time-of-day control parameter vector 。
5. The arc welding method of claim 1 wherein the welding process parameter set is adjusted as follows: Based on the welding process state characteristics, combining Time-of-day control parameter vector Constructing a linkage adaptation model between control parameters and welding process parameter sets, and realizing dynamic adjustment of the welding process parameter sets by using the linkage adaptation model; the linkage adaptation model is a nonlinear mapping model, and the expression form is as follows: , wherein, The vector is input for the model and, , For the amount of welding current adjustment, For the amount of the welding voltage adjustment, The welding speed adjustment amount is used; For the parameter to be mapped to a weight matrix, The vector is input for the model and, , For the amount of adjustment of the amplitude of the wobble, For the amount of adjustment of the wobble frequency, For the correction factor of the wobble track, In order to correct the vector for the deviation, Is a nonlinear activation function; The parameter maps the weight matrix Deviation correction vector And carrying out real-time iterative updating according to the state characteristics of the welding process and the evaluation results of the multi-objective rewarding function, and realizing the cooperative self-adaptive adjustment of the control parameters and the welding process parameter sets.
6. The arc welding method of claim 4 wherein said optimal parameters are selected as follows: Defining a set of control parameters as Wherein the control parameter Comprises an oscillation amplitude adjustment amount, an oscillation frequency adjustment amount, and an oscillation track correction coefficient, wherein, ; Aggregating the control parameters Inputting the welding simulation environment, and obtaining welding process data corresponding to each parameter With the welding process data Based on the calculation of each set of control parameters using the target reward function Integrated bonus average over all weld cycles : , wherein, Is the duration of the welding cycle for a single set of parameters, Is that Control parameters of time of day A corresponding comprehensive prize value; setting a reward threshold Screening out the products meeting the requirements If the candidate parameter set is empty, then taking the average value of the comprehensive rewards If the candidate parameter set is non-null, further calculating the stability coefficient of the candidate parameter Selecting stability coefficient Maximum parameter as Time-of-day optimal control parameter vector ; Screening out the products meeting the requirements If the candidate parameter set is empty, then taking If the candidate parameter set is not empty, further calculating the stability coefficient of the candidate parameter 。
7. An arc welding method for use in a non-structural environment according to claim 1, wherein the welding process closed loop control system is constructed as follows: Constructing a welding database according to a preset hierarchical data structure, wherein the welding database takes welding material types as primary indexes, and a plurality of sub-databases are formed by dividing; each sub-database takes weld features as a secondary index to construct a weld feature library, and the weld feature library is internally associated with and stores adaptive control parameter vectors and welding process parameter groups which are in one-to-one correspondence with the weld features; during arc welding operation, the current material type and the current welding seam characteristic of a welding area are obtained, and the primary index retrieval is carried out on a welding database based on the current material type: And if the primary index search is unsuccessful, generating an adaptive control parameter vector and a welding process parameter set corresponding to the current material type, and adding a sub-database corresponding to the material type into the welding database according to the preset hierarchical data structure to synchronously construct a welding seam feature database of the sub-database.
8. The arc welding method for use in a non-structural environment according to claim 7, further comprising: if the secondary index search is successful, directly calling an adaptive control parameter vector and a welding process parameter set corresponding to the current weld feature in the weld feature library; if the secondary index search is unsuccessful, an adaptive control parameter vector and a welding process parameter set corresponding to the welding seam feature under the current material type are generated, and the welding seam feature and the adaptive parameter are supplemented into a welding seam feature library corresponding to the sub-database according to the preset hierarchical data structure.
9. An arc welding robot for use in a non-structural environment for implementing the arc welding method according to any one of claims 1 to 8, comprising a lightweight arc welding robot arm body and a system module integrated on the lightweight arc welding robot arm body, the system module comprising: The first module is arranged to collect multi-source signal information of a welding area in a non-structural environment, perform synchronous preprocessing on the multi-source signal information and construct a multi-module sensing matrix ; A second module arranged to sense the matrix for the plurality of modules The data in the multi-module sensing matrix is subjected to space-time alignment and noise reduction treatment to obtain a multi-module sensing matrix after treatment Based on the processed multi-module sensing matrix Extracting weld joint characteristics and welding process state characteristics; The third module is set to take the weld joint characteristics as tracking targets, plan the self-adaptive welding track and initialize the welding process parameter set, create a self-adaptive swing tracking algorithm based on reinforcement learning, continuously optimize the control parameters, dynamically adjust the welding process parameter set according to the optimized control parameters; And the fourth module is arranged for controlling the arc welding execution mechanism to carry out welding operation based on the adjusted control parameters and the welding process parameter groups, synchronously collecting real-time multisource feedback signals in the welding process, constructing a closed-loop control system of the welding process and realizing the dynamic regulation and control of the arc welding operation in a non-structural environment.
10. An arc welding robot for use in a non-structural environment according to claim 9, wherein, The arm body of the lightweight arc welding mechanical arm body adopts a hollow topological optimization structure.

Description

Arc welding robot and arc welding method for non-structural environment Technical Field The invention belongs to the technical field of arc welding, and particularly relates to an arc welding robot and an arc welding method used in a non-structural environment, which are particularly suitable for high-precision welding scenes of complex components such as a battery box of a new energy automobile. Background Along with the high-speed development of the new energy automobile industry, the battery box is used as a core bearing component of the whole automobile power system, and the welding quality of the battery box directly determines the structural safety, the sealing reliability and the service life of the whole automobile. The 6061-T6 aluminum alloy becomes a main stream material of the battery box due to the characteristics of light weight and high strength, but the material has high heat conductivity and low melting point, and is easy to generate defects such as air holes, cracks, undercut and the like in the welding process, and strict requirements are put on the accuracy and stability of the welding process. In actual production, the battery box is welded in a non-structural environment, and complex working conditions such as thermal deformation of a workpiece, assembly errors (+ -1.5 mm), dynamic change of curvature of a curved weld joint and the like exist, so that the technical limitation of the traditional arc welding robot is increasingly remarkable. The existing arc welding robots mostly adopt a fixed track control mode of teaching-reproduction, have extremely poor adaptability to dynamic interference of non-structural environments, and are difficult to realize accurate tracking of curved-surface welding seams (with curvature radius not less than 5 mm) and T-shaped fillet welding seams (with gap not more than 2 mm), so that the welding defect rate is high. In the aspect of sensing technology, domestic equipment mostly depends on a single visual sensing mode, the cooperative sensing capability of electric arc and voiceprint signals is lacked, the process state and environment information in the welding process cannot be comprehensively captured, closed-loop control of welding quality is difficult to form, few imported equipment (such as German KUKA ArcTech series) for multi-mode sensing is introduced, although vision and electric arc signal fusion is realized, the problems of closed core algorithm, high equipment cost, insufficient process suitability for a new energy battery box and the like exist, and the voiceprint signals are not incorporated into a sensing system, so that microscopic process characteristics such as penetration fluctuation and the like cannot be analyzed. Meanwhile, the existing welding control algorithm is mainly controlled by traditional PID, has weak generalization capability on different welding seam forms and welding materials, has high parameter sensitivity, ensures that penetration fluctuation can reach +/-1.2 mm, and needs a large amount of manual debugging, so that the production efficiency is low and the labor cost is high. Disclosure of Invention The invention provides an arc welding robot and an arc welding method used in a non-structural environment for solving the technical problems in the background technology. The invention adopts the following technical scheme that the arc welding method used in the non-structural environment comprises the following steps: Collecting multi-source signal information of a welding area in a non-structural environment, carrying out synchronous pretreatment on the multi-source signal information and constructing a multi-module sensing matrix ; For the multi-module sensing matrixThe data in the multi-module sensing matrix is subjected to space-time alignment and noise reduction treatment to obtain a multi-module sensing matrix after treatmentBased on the processed multi-module sensing matrixExtracting weld joint characteristics and welding process state characteristics; The weld joint characteristics are used as tracking targets, self-adaptive welding tracks are planned, and a welding process parameter set is initialized; and controlling an arc welding executing mechanism to carry out welding operation based on the adjusted control parameters and welding process parameter groups, synchronously collecting real-time multisource feedback signals in the welding process, constructing a closed-loop control system of the welding process, and realizing dynamic regulation and control of arc welding operation in a non-structural environment. In a further embodiment, the multi-source signal information includes at least a visual signal, an arc signal, and a voiceprint signal of the welding area; the multi-module sensing matrix The construction flow of (2) is as follows: Arc filtering, image enhancement and downsampling are sequentially carried out on the visual signals, filtering denoising, amplitude calibration and synchronous sampling are se