An intelligent welding method for machine tool manufacturing

Abstract

The application discloses an intelligent welding method for machine tool manufacturing, which comprises a multi-sensor fusion module, a welding quality real-time evaluation module, a dynamic path planning module, an adaptive parameter adjustment module and a man-machine interaction interface. The multi-sensor fusion module comprises a visual sensor, a laser ranging sensor and a temperature sensor. The welding quality real-time evaluation module is connected with the dynamic path planning module. A defect detection algorithm based on deep learning is used to unify the data set of the multi-sensor fusion module. The dynamic path planning module combines 3D visual scanning production workpiece three-dimensional point cloud data, generates a welding path through an ant colony optimization algorithm, adjusts the welding state in combination with the data collected by the multi-sensor fusion module, optimizes the training process of the abnormal detection model in combination with an adaptive learning rate and a regularization technology, and adjusts and controls a mechanical welding gun through the man-machine interaction interface.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of welding, in particular to an intelligent welding method for machine tool manufacturing. BACKGROUND

[0002] Industrial robots play an increasingly important role in industrial production and play an increasingly important role. Using robots to weld in production helps to improve the stability of product quality, speed up the response speed of enterprises to orders and the overall production speed, shorten the delivery cycle, and at the same time in countries where the aging of the population is becoming increasingly serious, using robots to assist or replace manual labor helps to reduce production costs and provide more competitive product prices. As a major type of industrial robot, welding robots have been widely used in the field of welding, significantly improving welding quality and efficiency. In order to solve the shortcomings of complex process, poor process linkability, and chaotic welding management in the welding process, the traditional welding management system has emerged. The traditional welding management system realizes the whole process control of the welding process of a single welding machine through software control, collects the welding parameters of each stage of welding, and at the same time integrates each dispersed process section through the welding management system to realize the management of the welding process. However, the traditional welding management system has shortcomings in weld tracking and post-weld quality, and the traditional welding management is not intelligent and automated in the welding process.

[0003] Therefore, it is necessary to provide an intelligent welding method for machine tool manufacturing to solve the above-mentioned technical problems. The intelligent welding management system is developed on the basis of the traditional welding management system, mainly adopts a central server and a field client distribution method, networks the welding machines on the construction site, designs the welding channel, collects and presets the welding parameters, and controls the welding quality in real time. SUMMARY

[0004] This section is intended to summarize some aspects of the embodiments of the present application and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of the specification to avoid obscuring the purpose of this section, abstract and title, and such simplifications or omissions cannot be used to limit the scope of the present application.

[0005] In view of the above-mentioned existing problems, the present application is proposed.

[0006] To solve the above technical problems, the application provides the following technical scheme: an intelligent welding method for machine tool manufacturing, characterized by comprising a multi-sensor fusion module, a welding quality real-time evaluation module, a dynamic path planning module, a self-adaptive parameter adjustment module and a man-machine interaction interface, the multi-sensor fusion module comprising a visual sensor, a laser ranging sensor and a temperature sensor; the welding quality real-time evaluation module is connected with the dynamic path planning module, a defect detection algorithm based on deep learning is used to unify the data set of the multi-sensor fusion module, the dynamic path planning module combines 3D visual scanning production workpiece three-dimensional point cloud data, and an ant colony optimization algorithm is used to make the dynamic path planning module generate a welding path, the self-adaptive parameter adjustment module adjusts the welding state in combination with the data collected by the multi-sensor fusion module, a self-adaptive learning rate and a regularization technique are combined to optimize the training process of an abnormality detection model, and the multi-sensor fusion module, the welding quality real-time evaluation module, the dynamic path planning module, the self-adaptive parameter adjustment module and the man-machine interaction interface are connected, the mechanical arm welding torch is adjusted and controlled through the man-machine interaction interface, and specifically comprises the following steps: S1, fixing a workpiece to be welded on a welding platform, and starting the mechanical arm welding torch through the man-machine interaction interface; S2, starting the multi-sensor fusion module in combination with the dynamic path planning module, setting the walking path of the mechanical arm welding torch through the man-machine interaction interface, and then moving the mechanical arm welding torch at a plurality of pre-welding points on the workpiece to be welded;

[0007] S3, collecting data in the welding process of the mechanical arm welding torch through the multi-sensor fusion module, simultaneously evaluating the quality of the welding points after welding through the welding quality real-time evaluation module, and transmitting the data recorded by the welding quality real-time evaluation module to a background man-machine interaction control unit; S4, programming the data of the dynamic path planning module and the self-adaptive parameter adjustment module in the man-machine interaction control unit through the background man-machine interaction control unit, transmitting the programming program to a mechanical arm welding torch control unit, and then controlling and driving the mechanical arm welding torch through the control unit to perform welding work.

[0008] As a preferred scheme of the intelligent welding method for machine tool manufacturing, the self-adaptive parameter adjustment module comprises closed-loop feedback control of welding current, welding voltage and welding speed.

[0009] As a preferred scheme of the intelligent welding method for machine tool manufacturing, the dynamic path planning module adopts a double-layer optimization mechanism, the first layer generates a rough path based on workpiece geometric characteristics, and the second layer optimizes welding sequence through thermodynamic simulation.

[0010] As a preferred scheme of the intelligent welding method for machine tool manufacturing, the multi-sensor fusion module further comprises a visual sensor, a weld tracking sensor and a temperature sensor.

[0011] As a preferred scheme of the intelligent welding method for machine tool manufacturing, the method further comprises a multi-modal perception unit, the multi-modal perception unit comprises a camera and a laser radar, through fusion of the camera and the laser radar, 0.05mm level weld tracking is realized, and pores and incomplete fusion defects are detected in real time through combination of a thermal imaging and an arc spectrum combined algorithm.

[0012] As a preferred scheme of the intelligent welding method for machine tool manufacturing, weld tracking is combined with weld forming mathematical modeling, and a welding heat transfer process is analyzed, the welding heat transfer process including heat source input height concentration, weld heat source mobility and weld heat transfer complexity.

[0013] As a preferred scheme of the intelligent welding method for machine tool manufacturing, the application adopts mathematical modeling and multivariate regression analysis modeling method, fuses and optimizes the mathematical analysis model based on the multivariate regression model, and compares and analyzes the accuracy of the fused model through actual welding data.

[0014] As a preferred scheme of the intelligent welding method for machine tool manufacturing, the step of establishing the multivariate regression model is:

[0015] (1) Collect data: including dependent variable values corresponding to independent variables at different factor levels;

[0016] (2) Model establishment: input the collected data into professional analysis software to establish a regression model;

[0017] (3) Model verification: key parameters in the multivariate regression model are verified, including significance test of regression coefficients, standard error test and goodness of fit test;

[0018] (4) Through the test result, the fitting effect of the model is evaluated to determine whether it has sufficient confidence.

[0019] As a preferred scheme of the intelligent welding method for machine tool manufacturing, based on the assumption conditions in the welding heat transfer process, if the heat source height concentration is simplified as a concentrated heat source, different analysis models can be established according to the thickness of the plate, including a moving point heat source analysis model and a moving line heat source model.

[0020] As a preferred scheme of the intelligent welding device for machine tool manufacturing, the Gaussian heat source is improved to be used as a heat input model in heat conduction analysis.

[0021] Beneficial effects of the present invention: The present invention adopts multimodal sensing technology and the fusion of RGB-D camera and TOF laser radar to achieve weld tracking accuracy, combines thermal imaging with arc spectrum combination algorithm to detect pores and unfused defects in real time, and constructs a twin model. The fusion optimization model after the multivariate regression analysis model can more accurately predict the arc welding weld formation data, thereby realizing the real-time prediction function of the weld formation of the digital twin system. The temperature distribution function is improved in combination with the Gaussian distribution heat source, which solves the error caused by simplifying the heat source to a concentrated heat source. The welding parameter intelligent optimization digital twin model based on the BP neural network realizes the process parameter optimization function in the welding simulation of the robot workstation digital twin system. Through the analysis of the weld size data, the optimization model can automatically adjust the welding parameters, reduce human intervention, realize the intelligence and automation of the welding process, and thus improve the welding quality and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0023] in:

[0024] Figure 1 A schematic flow chart of an intelligent welding method for machine tool manufacturing according to an embodiment of the present invention;

[0025] Figure 2 A flow chart of an intelligent optimization model of an intelligent welding method for machine tool manufacturing according to an embodiment of the present invention; DETAILED DESCRIPTION

[0026] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0027] Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without making any creative work should fall within the scope of protection of the present invention.

[0028] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0029] Embodiment 1

[0030] With reference to Figure 1 According to the embodiment of the present application, an intelligent welding method for machine tool manufacturing is characterized by comprising a multi-sensor fusion module, a welding quality real-time evaluation module, a dynamic path planning module, an adaptive parameter adjustment module and a man-machine interaction interface. The multi-sensor fusion module comprises a visual sensor, a laser ranging sensor and a temperature sensor. The welding quality real-time evaluation module is connected with the dynamic path planning module. A defect detection algorithm based on deep learning unifies the data set of the multi-sensor fusion module. The dynamic path planning module combines 3D visual scanning production workpiece three-dimensional point cloud data to generate a welding path through an ant colony optimization algorithm. The adaptive parameter adjustment module adjusts the welding state in combination with the data collected by the multi-sensor fusion module. In combination with an adaptive learning rate and a regularization technique, the training process of the abnormal detection model is optimized. The multi-sensor fusion module, the welding quality real-time evaluation module, the dynamic path planning module, the adaptive parameter adjustment module and the man-machine interaction interface are connected to adjust and control the welding gun of the mechanical arm through the man-machine interaction interface. Specifically, the method comprises the following steps: S1, fixing the workpiece to be welded on the welding platform and starting the welding gun of the mechanical arm through the man-machine interaction interface;

[0031] S2, starting the multi-sensor fusion module in combination with the dynamic path planning module, setting the walking path of the welding gun of the mechanical arm through the man-machine interaction interface, and then moving the welding gun of the mechanical arm at multiple pre-welding points on the workpiece to be welded;

[0032] S3, collecting data in the welding process of the mechanical arm welding torch through the multi-sensor fusion module, and evaluating the quality of the welding spot after welding through the welding quality real-time evaluation module, transmitting the data recorded by the welding quality real-time evaluation module to the background man-machine interaction control unit; S4, programming the data of the dynamic path planning module and the self-adaptive parameter adjustment module in the man-machine interaction control unit through the background man-machine interaction control unit, and transmitting the programming program to the mechanical arm welding torch control unit, then controlling the driving unit through the control unit to drive the mechanical arm welding torch to perform welding work. Specifically, the self-adaptive parameter adjustment module includes closed-loop feedback control of welding current, welding voltage and welding speed. The welding parameters (such as current, voltage, wire feeding speed, welding path, etc.) can be dynamically optimized according to the real-time working condition, so as to improve the welding quality and efficiency. The mechanical arm welding torch changes the welding state by itself through the self-adaptive parameter adjustment module during the welding process, records the data of current, voltage and speed, and feeds back the background control unit, then evaluates the quality after welding through the welding quality real-time evaluation module, and adjusts the parameters of the mechanical arm welding torch if it is unqualified. The sensor and data acquisition module includes: current / voltage sensor, temperature sensor including infrared or thermocouple, visual sensor (camera, laser scanner), weld tracking sensor (ultrasonic or laser ranging), gas flow / pressure sensor, which can collect key data in the welding process in real time, such as molten pool shape, arc characteristics, weld position, environmental temperature, etc., and provide input basis for parameter adjustment.

[0033] The data processing and analysis module utilizes signal filtering algorithms (to eliminate noise), feature extraction algorithms (such as molten pool width, weld deviation), and data fusion techniques (integration of multi-sensor information) to preprocess the original data, extract key features (such as molten pool dynamic behavior, weld offset), and provide structured information for subsequent decision-making. The parameter decision module utilizes control algorithms (PID, fuzzy logic, model predictive control), machine learning models (such as neural networks, reinforcement learning), and real-time data and preset targets (such as welding quality, efficiency) to dynamically generate optimal parameter adjustment schemes. For example: current / voltage adjustment: to cope with changes in plate thickness or thermal deformation; welding speed optimization: to adapt to weld track deviation; wire feeding speed matching: to ensure the stability of deposited metal. The actuator module converts the output instructions of the decision module into physical actions through servo motors (adjusting the position of the welding torch), inverter power supplies (adjusting current / voltage), wire feeders (controlling wire feeding speed), and mechanical arm motion controllers (adjusting welding path), to realize real-time adjustment of parameters. The feedback control loop includes closed-loop control logic (real-time comparison of target and actual values), dynamic compensation mechanisms (such as heat input compensation), which continuously monitor the difference between output results and expected targets, correct parameter adjustment strategies, and ensure system stability. For example: when the molten pool temperature is too high, automatically reduce the current or increase the welding speed; when the weld deviates, adjust the robot path tracking accuracy. The user interaction and monitoring interface is an HMI (human-machine interface) with data visualization tools (such as dynamic display of the molten pool) and parameter manual override functions, allowing operators to monitor system status, intervene in parameter adjustment strategies, and set process constraints (such as maximum current limit). In addition, there is a safety protection module that can detect abnormalities and trigger emergency stops, such as arc breaking, gas leakage, etc. The safety protection module triggers protective measures when it detects abnormal conditions (such as parameter over-limit, equipment failure), to avoid equipment damage or welding defects. At the same time, the control unit has a system calibration and self-learning module, which includes offline calibration tools, historical data storage and analysis, and online model update functions; through long-term data accumulation, the decision model is optimized to adapt to new materials, new processes or environmental changes, and to improve the generalization ability of the system.

[0034] In order to improve the rationality of the dynamic path planning module, the dynamic path planning module adopts a two-layer optimization mechanism. The first layer generates a rough path based on the geometric features of the workpiece, and the second layer optimizes the welding sequence through thermodynamic simulation. The first step is to extract the 3DCAD model or scanned point cloud data of the workpiece with geometric features. It can identify the type of weld (butt joint, fillet joint, overlap joint, etc.); extract geometric features such as weld position, length, groove shape; and mark complex areas (such as corners, surfaces, and intersections of multiple welds). The second step is path topology planning to generate a continuous trajectory covering all welds. Through graph theory optimization: decompose the weld into nodes and connect them through the shortest path algorithm (such as Dijkstra); use the block strategy to plan the blocks of complex geometries (such as symmetrical structure segmented welding); the robot arm kinematic constraints are combined with the welding gun posture (TCP posture) and the range of joint motion to generate a collision-free path. The third step, the coarse path output can generate a geometrically feasible welding path sequence (such as the coordinate points of the welding gun movement, the welding direction, and the segmentation mark); the second layer is based on the optimization of the welding sequence of thermodynamic simulation, which predicts the heat accumulation and deformation of different welding sequences through simulation, and selects the optimal sequence to minimize residual stress and deformation. The first step, thermal-mechanical coupling modeling, the model input, the welding segment division of the coarse path, the material properties (thermal conductivity, thermal expansion coefficient), and the process parameters (current, voltage, speed). Finite element analysis establishes a three-dimensional transient heat conduction model to simulate the evolution of the temperature field. The second step, welding sequence optimization, the optimization variables adjust the execution order of each welding segment in the coarse path, the optimization algorithm heuristic search: the genetic algorithm (GA) generates the candidate sequence, the simulation results are evaluated and iteratively optimized; the symmetry principle gives priority to the alternating welding sequence for symmetrical structures to offset thermal stress. Dynamic parameter adjustment is performed through the human-computer interaction control unit,

[0035] Collaborative Optimization: While adjusting the sequence, welding parameters may be fine-tuned (e.g., reducing the current in segment B by 10% to balance heat input). The final welding sequence and supporting parameters (e.g., segment order, speed, and energy) are adjusted based on the output. The dynamic path planning module uses a two-layer collaborative mechanism for data transmission. The first layer outputs a rough path (geometric segmentation), which the second layer uses as input for simulation optimization. If the second layer finds that a certain path segment will inevitably lead to severe deformation (e.g., welding a long straight weld in one direction), it may trigger the first layer to re-segment (e.g., split into multiple reverse welding segments). Iterative optimization is very suitable for different materials. The iterative optimization process is geometric planning → thermodynamic optimization → output of the final path. A feedback mechanism is used for supervision. If the simulation shows excessive deformation, the second layer can request the first layer to adjust the path topology (e.g., increase the heat dissipation gap). Thermodynamic optimization is completed in advance for fixed workpieces through offline precalculation. Online adjustments are made based on the welding conditions at the welding site. Lightweight models are used to quickly respond to dynamic working conditions (e.g., temporary weld changes). The dynamic path planning module decouples geometry and thermodynamics and can reduce computational complexity through hierarchical processing. It utilizes multi-objective balance to simultaneously meet path feasibility and welding quality. In addition, it has strong adaptability and is compatible with complex workpieces and multiple processes.

[0036] In addition, the multi-sensor fusion module also includes a visual sensor, a weld tracking sensor and a temperature sensor. It also includes a multimodal sensing unit, which includes a camera and a laser radar. Through the fusion of the camera and the laser radar, 0.05mm-level weld tracking can be achieved. At the same time, the thermal imaging and arc spectrum combined algorithm can be combined to detect pores and unfused defects in real time. The weld tracking is combined with the mathematical modeling of the weld formation to analyze the welding heat transfer process. The welding heat transfer process includes highly concentrated heat source input, mobility of the welding heat source, and complexity of the welding heat transfer. For the heat conduction problem in general welding, when the thermal conductivity coefficient, density ρ, and specific heat capacity c of the known object are known, the welding heat transfer process is analyzed. p In the case of , the differential equation is shown in formula (1.1):

[0037]

[0038] Where λ, ρ and c p are functions that vary with temperature. If we assume they are constants, equation (1.1) can be simplified to:

[0039]

[0040] To solve equation (1.2), constraints must be added, such as the boundary conditions and initial conditions of the material. The welding heat transfer process is a complex three-dimensional heat conduction process. It is mainly manifested in:

[0041] (1) Highly concentrated heat source input

[0042] The heat input in welding comes from the welding torch tip, which makes only local areas of the workpiece be heated by the heat source during the heating process, and the heating and cooling of the whole workpiece is extremely uneven. Generally, the temperature changes greatly in the area around the welding torch, and the temperature changes become smaller and smaller as the distance from the welding torch increases. Moreover, the heating and temperature change curve gradient of the workpiece during the heating process is large.

[0043] (2) Movement of welding heat source

[0044] Since the welding torch always moves during the welding process, the welding heat source also remains in a moving state. This makes the heat-affected zone change constantly, and the points in the heat-affected zone are heated and cooled rapidly when they leave the heat-affected zone, so that the process can only reach a "quasi-stable state" at the end.

[0045] (3) Complexity of welding heat transfer

[0046] Due to the movement of the welding torch, the molten pool also remains in a moving state during welding. In the molten pool, the heat transfer is mainly convective heat transfer due to the phase change of the workpiece; in the non-melted area, it is mainly solid heat conduction. On the boundary element of the workpiece, there are also forms of heat conduction such as convective heat dissipation and radiative heat dissipation. At the same time, the change of the stress field of the workpiece also makes the temperature calculation more complex. Therefore, in order to realize real-time prediction of the weld formation, assumptions are generally made to simplify the calculation during the welding heat conduction calculation:

[0047] (1) In the process of robot arc welding, the heat input of the workpiece is only affected by the heat source parameters;

[0048] (2) The thermal physical properties of the material do not change with temperature, and there is no phase change after reaching the melting point; (3) The thermal conductivity of the material in different directions is the same, i.e., isotropic.

[0049] The present application improves the Gaussian heat source, and combines the improved analytical formula of the Gaussian heat source. During the process of heating and welding the plate by the welding torch, the heat flow is mainly concentrated in a small area, and the radius of this area is the effective radius r h of the heating spot. Since the heat flow on the heating spot can be approximately represented by a Gaussian function, the Gaussian heat source is used as the heat input model in heat conduction analysis. The mathematical expression of the Gaussian heat source is:

[0050]

[0051] In the formula, q(r) is the heat flux density at a distance Q from the center; Q is the effective power of the welding arc; K is the heat energy concentration coefficient. There is also the following expression for the Gaussian heat source:

[0052]

[0053] σ q is the parameter of the Gaussian heat source distribution, which is related to the heat concentration coefficient K and the effective radius r h of the Gaussian heat source as follows:

[0054]

[0055] On an infinite plate, assume that there is a Gaussian distribution heat source with an effective power of Q on the surface of the welding part. A coordinate system is established with the center of the Gaussian distribution heat source as the origin. At any point (x', y') in the heat source action area, the heat input dQ = q(r') dx' dy' dt, which causes a temperature change of

[0056]

[0057] wherein

[0058]

[0059] Substituting equation (1.4) into equation (1.6) gives:

[0060]

[0061] The Gaussian heat source can be regarded as the sum of the action of infinite point heat sources. The analytical expression of the temperature change caused by the Gaussian distribution heat source acting on the thick welding part at point (x, y, z) at time t is obtained as follows:

[0062]

[0063] When the Gaussian distribution heat source moves along the surface of the welding part at a speed v0, the mathematical expression of the temperature change caused by the Gaussian distribution heat source acting on the thick welding part at point (x, y, z) at time t can be obtained by dividing the time period into infinite micro-intervals, integrating them, and converting them to the moving coordinate system as follows:

[0064]

[0065] When σ q = 0, equation (1.9) becomes the mathematical expression of the heating process of the welding material under the action of the moving point heat source as follows:

[0066] T| t＝0 = δ(x, y, z)

[0067] By combining the Gaussian heat source with the analytical solution model, an improved analytical solution model is obtained, which can be used as the basis for designing an analytical algorithm to calculate the weld size during the welding process. The real-time prediction model of weld formation has great significance in the welding robot workstation. By predicting the size and shape of the weld in real time, the process parameters can be adjusted in a timely manner during the welding process to ensure the quality and consistency of the weld. This predictive ability can identify potential welding defects in advance during complex welding tasks, thereby reducing waste rates and improving production efficiency and product quality. Digital twin technology provides strong support for this, enabling virtual simulation to be closely integrated with actual operations to achieve efficient and precise welding process control.

[0068] The present application adopts mathematical modeling and multiple regression analysis modeling methods, and fuses and optimizes the mathematical analytical model based on the multiple regression model. The accuracy of the fused model is compared and analyzed with actual welding data.

[0069] Multiple regression analysis model

[0070] As a regression analysis model, it differs from general regression analysis in the number of independent variables. Generally speaking, the dependent variable in a two or more regression model is called a multiple regression model. The mathematical expression for analyzing the correlation between the effective radius and the independent variables using a multiple linear regression model is:

[0071] y = β0+ β1x1+ β2x2+ … + β k x k + ε

[0072] In the formula, y is the effective radius r h of the heat source; x i (i = 1, 2, …, k) is the welding process parameter;

[0073] β i (i = 1, 2, …, k) is the regression coefficient; ε is the random error variable, which represents the difference between the actual observed value and the predicted value of the regression model. According to the least squares method, the estimate of the regression coefficient is β = (X'X) -1 X' y

[0074] The steps for establishing a multiple regression model are generally as follows:

[0075] (1) Collect data: including the dependent variable values corresponding to the independent variables at different factor levels;

[0076] (2) Establish a model: input the collected data into professional analysis software such as SPSS, Stata, EXCEL, and establish a regression model;

[0077] (3) Model verification: Key parameters in the multiple regression model are verified, including significance test of regression coefficient, standard error test, goodness-of-fit test, etc.

[0078] (4) Based on the test results, the fitting effect of the model is evaluated to determine whether it has sufficient confidence. If the effect is poor, the problem needs to be found out and the model needs to be re-established. The digital twin model based on the multiple regression analysis model of the weld forming real-time prediction is used to solve the real-time generation and size prediction function of the weld in the digital twin system.

[0079] Digital twin has high requirements for real-time performance in order to realize real-time mapping of physical entities, online monitoring of physical data, etc. For example: the sampling frequency of robot running data is 20Hz, and the basic time delay between the digital twin and the physical entity is 50ms. This time delay is controllable and can be adjusted according to actual needs. If the real-time requirement is high, 10kHz or higher sampling frequency data can be used to drive the digital twin within the allowable range of Redis read-write speed, so that the delay is reduced to 0.1ms or less.

[0080] The collection of mechanical arm twin evolution data has multiple evaluation methods for welding quality. The commonly used methods include qualitative evaluation of macroscopic morphology, such as whether the weld surface is smooth and whether there are cracks and pores; quantitative evaluation of macroscopic morphology, such as the width, height and penetration of the weld; mechanical property evaluation, such as the tensile strength and hardness of the weld; qualitative evaluation of microscopic morphology, such as the morphology of grains in the weld; and quantitative evaluation of microstructure, such as the composition ratio of elements in the weld. In this paper, the commonly used evaluation indexes of weld quality, such as weld width, weld height and weld penetration, are selected to evaluate the weld quality obtained by welding. Welding quality is affected by many factors. In theory, any factor involved in the welding process will affect the welding quality. This application also relates to the influence of controllable welding parameters on welding quality, including six parameters such as welding current and voltage, welding gun moving speed and angle, welding wire length, and protective gas flow. At the same time, the influence of three groups of interaction factors, such as welding current and welding speed, welding current and wire length, and welding current and welding angle, on welding quality is considered. In the design of welding test, the combination of welding parameters is determined to make the determined parameter combination representative in the combination of all factor levels, so that the data samples obtained by the test can reflect the overall relationship between the parameters and the indexes, and the overall description of the relationship between the parameters and the indexes can be realized with fewer test times, shorter test time and lower test cost. The commonly used methods include the orthogonal design method in the field of engineering test and the Pairwise algorithm in the field of software test. The orthogonal design method is a classical test design method, and the combination of test parameters designed by the method has the characteristics of "uniform dispersion and neat comparison", that is, the number of different levels of each parameter is equal, and the level combination of any two parameters is comprehensive and balanced. The Pairwise algorithm is a modification of the orthogonal test method based on statistics. It is believed that the research targets such as quality and failure usually involve the combination of two factors. Compared with the orthogonal test, the Pairwise algorithm can further reduce the number of tests, and the generated parameter combination still has a certain representativeness, and is especially suitable for mixed level test design. In order to determine the influence of the three groups of interaction factors, such as welding current and welding speed, welding wire length and welding angle, on welding quality, the corresponding interaction table needs to be determined at the same time as the orthogonal table, so as to determine the position of the interaction column in the orthogonal table. The allowed values of the welding test parameters are shown in Table 1:

[0081] Table 1

[0082] Item Welding current Welding voltage Welding gun speed Welding gun angle Welding wire length Gas flow Level 1 120 14 9 60 10 14 Level 2 170 20 11 90 15 20 Level 3 250 24 13 120 20 23

[0083] From the above table, it can be seen that the six factors in the experiment are all three levels, and the degrees of freedom of the six factors are all two. The degrees of freedom of the three pairs of interaction factors are four, and the mathematical expression is:

[0084] f A×B =f A·f B =2×2=4

[0085] Among them, f A 、f B are the degrees of freedom of two factors A and B, respectively, A×B The degrees of freedom of its interaction factors. Therefore, there are two options for selecting an orthogonal array when designing an orthogonal experiment: (1) Select a single-level orthogonal array, where the levels of each column in the table are 3 and the degrees of freedom are 2. Since the degrees of freedom of each interaction factor are 4, they need to occupy 2 columns in the orthogonal array. The mathematical expression for the number of columns in the orthogonal array is:

[0086] k min =6+2×3=12

[0087] The minimum specification of the orthogonal array is L N1 (3 12 );

[0088] (2) Select an orthogonal table with mixed levels, where the level of the columns containing the six single factors is 3 and the level of the columns containing the three pairs of interaction factors is 5. The mathematical expression for the number of columns in the orthogonal table is:

[0089] k min =6+3=9

[0090] The minimum specification of the orthogonal array is L N2 (3 6 ,5 3 ).

[0091] In the above two schemes, k min This is because the interaction columns cannot be arranged arbitrarily in the orthogonal table to prevent confounding. Therefore, the position of the interaction column needs to be determined according to the positions of its two single factors. When it conflicts with the positions of other factors, an orthogonal table with more columns is required.

[0092] Among them, the acquisition of BP neural network learning samples mainly includes data acquisition and cleaning, data transformation, data normalization and other operations. Data cleaning refers to finding some identifiable errors in the data and removing them, thereby providing more accurate and reliable data for subsequent data analysis, mainly including invalid values, duplicate values, etc. After removing the repeated and abnormal data through data cleaning, the remaining data is saved as a neural network training and verification dataset. Data transformation mainly includes transformation of non-numeric data, and here the One-hot encoding technology is mainly used. This encoding technology is mainly suitable for non-numeric data between categories without size relationship, such as plate thickness, joint form, welding pose, etc. The basic principle is to represent each category as a binary vector, where only one element is 1 and the other element is 0. Combined with the actual situation of the BP neural network parameter optimization model, when using the One-hot encoding technology, it is adjusted to 1 and -1 respectively. This transformation can improve the feature discrimination of the data, because the difference between -1 and 1 is greater than the difference between 0 and 1, which improves the noise immunity of the model.

[0093] Data normalization is a common data preprocessing technique that scales the original data obtained to a small specific interval according to a certain mathematical transformation method, thereby adjusting the data to a unified scale, making the features in the data comparable, so as to better analyze and process the data. The commonly used normalization interval is [0, 1] or [-1, 1]. Since in data transformation, non-numeric data is converted to -1 and 1 using One-hot encoding, in order to reduce the amount of normalization operation, the normalization interval is set to [-1, 1].

[0094] The formula used for data normalization is:

[0095]

[0096] Normalization processing can effectively eliminate the influence of different magnitudes and dimensions between data, and at the same time, the solving speed of gradient descent is also improved, so that the convergence speed of the model is correspondingly improved.

[0097] In summary, the weld forming real-time prediction model is based on the Gaussian heat source model and multiple regression analysis, which can predict the size and shape of the weld in real time during the welding process. The welding process parameter optimization model uses BP neural network to analyze the input weld size data such as fusion width and stack height, and then designs and optimizes more reasonable welding current, welding voltage and welding speed, thereby ensuring welding quality and efficiency.

[0098] It is to be understood that the development of the particular implementations described herein was motivated by the desire to solve real-world problems, and as such the claimed implementations can be susceptible to further implementation while still being generically consistent with the descriptions provided herein. Specifically, although many of the examples provided herein describe one or more implementations with any particular feature, an individual feature can be replaced by alternative features within the scope of the application. Thus, features discussed in one example can be interchanged with features in another example. Any implementation of more than one feature disclosed herein is specifically referenced within the scope of the application.

[0099] It should be noted that the above examples are merely used to illustrate the technical solutions of the present application, rather than limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and all should be covered in the scope of the claims of the present application.

Claims

1. An intelligent welding method for machine tool manufacturing, characterized in that: It includes a multi-sensor fusion module, a real-time welding quality assessment module, a dynamic path planning module, an adaptive parameter adjustment module and a human-computer interaction interface. The multi-sensor fusion module includes a visual sensor, a laser ranging sensor and a temperature sensor; the real-time welding quality assessment module is connected to the dynamic path planning module, and the multi-sensor fusion module is unified into a data set based on a deep learning defect detection algorithm. The dynamic path planning module combines 3D visual scanning to produce three-dimensional point cloud data of the workpiece, and generates a welding path through the ant colony optimization algorithm. The adaptive parameter adjustment module adjusts the welding state in combination with the data collected by the multi-sensor fusion module, and optimizes the training process of the anomaly detection model by combining adaptive learning rate and regularization technology. The multi-sensor fusion module, the real-time welding quality assessment module, the dynamic path planning module, the adaptive parameter adjustment module and the human-computer interaction interface are connected, and the robot arm welding gun is adjusted and controlled through the human-computer interaction interface. Specifically, it includes the following steps: S1. Fix the workpiece to be welded on the welding platform and start the robotic arm welding gun through the human-machine interface; S2. Start the multi-sensor fusion module in combination with the dynamic path planning module, set the walking path of the robotic arm welding gun through the human-computer interaction interface, and then move the robotic arm welding gun to multiple pre-welding points on the workpiece to be welded; S3, collecting data from the robotic arm welding gun during the welding process through the multi-sensor fusion module, and simultaneously evaluating the quality of the welds after welding through the welding quality real-time evaluation module, and transmitting the data recorded by the welding quality real-time evaluation module to the background human-computer interaction control unit; S4. Programming the data of the dynamic path planning module and the adaptive parameter adjustment module in the human-machine interaction control unit through the human-machine interaction control unit, and transmitting the programming program to the robotic arm welding gun control unit, and then controlling the driving unit through the control unit to drive the robotic arm welding gun to perform welding operations; The dynamic path planning module uses a two-layer optimization mechanism. The first layer generates a rough path based on the workpiece geometry, and the second layer optimizes the welding sequence through thermodynamic simulation. In addition, mathematical modeling and multivariate regression analysis modeling methods are used to fuse and optimize the mathematical analytical model based on the multivariate regression model, and the accuracy of the fused model is analyzed by comparing it with actual welding data. It also includes improvements to the Gaussian heat source as a heat input model in heat conduction analysis; By combining the Gaussian heat source with the analytical solution model, an improved analytical solution model is obtained. Based on the Gaussian heat source model and multiple regression analysis, the size and shape of the weld can be predicted in real time during the welding process. The adaptive parameter adjustment module includes closed-loop feedback control of welding current, welding voltage and welding speed; The method also includes a multimodal sensing unit, which includes a camera and a lidar. Through the cooperation of the camera and the lidar, 0.05mm level weld tracking is achieved. At the same time, thermal imaging and arc spectrum combined algorithms are combined to detect pores and unfused defects in real time.

2. The intelligent welding method for machine tool manufacturing according to claim 1, characterized in that: Weld seam tracking is combined with mathematical modeling of weld formation to analyze the welding heat transfer process, which includes the highly concentrated heat source input, the mobility of the welding heat source, and the complexity of the welding heat transfer.

3. The intelligent welding method for machine tool manufacturing according to claim 1, characterized in that: The steps to build a multiple regression model are: (1) Collect data: including the values ​​of the dependent variable corresponding to the independent variable at different factor levels; (2) Model building: Input the collected data into professional analysis software to build a regression model; (3) Model testing: testing the key parameters in the multiple regression model, including significance test of regression coefficients, standard error test, and goodness of fit test; (4) Evaluate the model’s fitting effect through the test results to determine whether it has sufficient confidence.

4. The intelligent welding method for machine tool manufacturing according to claim 1, characterized in that: Based on the assumptions in the welding heat transfer process, if the heat source is simplified to a concentrated heat source according to its highly concentrated characteristics, different analytical models can be established according to the thickness of the plate. Including moving point heat source analytical model and moving line heat source model.

Images