Weld joint quality self-adaptive control system and method based on fuzzy controller

Abstract

The invention belongs to the field of robot welding, relates to robot welding quality control, and discloses a welding seam quality self-adaptive control system and method based on a fuzzy controller. The system comprises a welding power source, a robot controller, an upper computer, a structured light vision device and a welding execution device; the welding execution device comprises a six-axis mechanical arm and a welding gun. The structured light vision device is mounted at the tail end of the welding gun and moves along with the movement of the welding gun; in the welding process, the structured light vision device recognizes a molten pool in the welding process, meanwhile, the welding power source and the robot controller feed back current welding process parameters, and the current welding process parameters are read through the upper computer. On the basis of manually determining the welding path, the technical parameters are adjusted in the welding process.

Description

Technical Field

[0001] The invention belongs to the field of robot welding and relates to robot welding quality control, and in particular to a weld quality adaptive control system and method based on a fuzzy controller. Background Art

[0002] In the process of using robots for automated welding, the weld state changes continuously as the welding progresses due to the influence of processing errors, assembly errors and cyclic heat input during welding. In the initial stage of welding, the welding process parameters set according to the ideal weld state will become unreasonable as the weld state changes during the welding process. Unreasonable welding process parameters are prone to cause welding defects such as bubbles, cracks, slag inclusions, and incomplete penetration. The presence of these defects will reduce the cross-sectional area of ​​the weld joint, reduce the load-bearing capacity and fatigue strength of the weld joint, and thus cause brittle fracture of the weld joint.

[0003] "An Intelligent Welding Quality Monitoring and Adjustment System", publication number: CN202411028614.0, this patent proposes an intelligent welding quality monitoring and adjustment system, which divides the time points based on the estimated welding time of the current welding workpiece, monitors and analyzes the changes in heat input in the time zone before the division in real time, and obtains the heat input change value of the time zone before the current welding workpiece. The heat input change value obtained by analysis provides a quantitative indicator, which can evaluate the heat input amount in the current workpiece welding process, and provides data support for the intelligent adjustment of welding parameters. However, this method relies heavily on mathematical models, and the accuracy of the model has a serious impact on the adjustment accuracy.

[0004] "Control device and method for the crystallization process of plasma arc welding weld based on temperature field", publication number: CN202211210941.9, this patent provides a control device and control method for the crystallization process of plasma arc welding weld based on temperature field, using an infrared monitoring system to monitor the temperature field during plasma arc welding. When the real-time temperature field characteristics or real-time temperature gradient characteristics exceed the corresponding threshold range, the control system adjusts the output parameters of the gas supply system and / or welding system, so that the real-time temperature field characteristics and real-time temperature gradient characteristics of the perforated molten pool are within the corresponding threshold range, thereby ensuring the welding quality. However, the infrared monitoring device used in this patent is expensive, the temperature field image processing algorithm is complex, and the infrared monitoring system is easily blocked by welding equipment such as welding guns, thereby affecting the infrared monitoring system's monitoring accuracy of the molten pool during welding.

[0005] "System and method for monitoring quality of rotary friction welding joints based on multi-information fusion", publication number: CN202411159661.9, this patent provides a system and method for monitoring quality of rotary friction welding joints based on multi-information fusion, which can capture and collect the geometric changes of welds and physical signal changes in real time during welding through the quality monitoring system of rotary friction welding joints, and monitor and evaluate the tensile strength and torsional strength of rotary friction welding joints online through LSTM neural network, thus filling the technical gap of online intelligent monitoring of rotary friction welding quality. However, this method uses a variety of sensor modules, including visual signal acquisition module, physical signal acquisition module, data transmission module, visual signal and physical signal analysis and processing module, quality prediction module, and decision-making module, which makes the equipment complex, costly, and difficult to maintain in the later stage.

[0006] "An Ultrasonic Welding Quality Control System and Method", publication number: CN202410341756.6, this patent discloses an ultrasonic welding quality control system and method, which uses a fuzzy neural network to analyze the welding quality based on the processed signal, predicts the PID parameters based on the analysis results and inputs the PID parameters into the PID controller; the PID controller is used to adjust the welding parameters according to the PID parameters, so that the mechanical actuator performs ultrasonic welding according to the adjusted welding parameters. However, the patent system has a complex structure, which not only includes a fuzzy neural network, but also requires adjustment of the PID parameters, and then adjusts the welding process parameters through the PID controller, and the system response time is slow. Summary of the invention

[0007] In order to solve the above technical problems, the present invention provides a weld quality adaptive control system and method based on a fuzzy controller, which adjusts the process parameters during the welding process on the basis of manually determining the welding path.

[0008] A fuzzy controller-based weld quality adaptive control system, the system comprising: a welding power source, a robot controller, a host computer, a structured light vision device, and a welding execution device;

[0009] The welding execution device includes: a six-axis robot arm and a welding gun;

[0010] The structured light vision device is installed at the end of the welding gun and moves with the movement of the welding gun;

[0011] During the welding process, the structured light vision device identifies the molten pool during the welding process. At the same time, the welding power supply and robot controller provide feedback on the current welding process parameters and read them through the host computer.

[0012] Preferably, the structured light vision device is a laser vision sensor, and the structured light vision device identifies the molten pool during the welding process including:

[0013] During the welding process, the laser vision sensor installed at the end of the welding gun collects the molten pool image, performs image processing on the collected image, and extracts the short axis length and long semi-axis length of the molten pool image.

[0014] Preferably, during welding, the weld depth and the weld width on the back of the weld joint are set according to the requirements. During the welding process, the molten pool information in the welding process identified by the structured light vision device and the current welding process parameters fed back by the welding power supply and the robot controller are input into the established neural network model, and the neural network model outputs the predicted values ​​of the weld depth and the weld width on the back of the weld joint;

[0015] Among them, the neural network model is to establish a prediction model for the welding joint penetration depth and the welding joint back side penetration width through the BP neural network.

[0016] Preferably, the welding process parameter adjustment variables are output by a multi-input multi-output fuzzy controller, and the welding process parameters are adaptively adjusted according to the information fed back by the structured light vision sensor, the welding power supply and the robot controller, so as to realize the control of the welding quality;

[0017] Among them, in the process of welding joint penetration control, the deviation between the expected welding joint penetration and the online predicted welding joint penetration is used. and the deviation change is the input variable of the fuzzy controller;

[0018] In the process of back weld width control of weld joint, the deviation between the expected back weld width of weld joint and the online predicted back weld width of weld joint is used. and the deviation change is the input variable of the fuzzy controller.

[0019] The present invention also provides a weld quality adaptive control method based on a fuzzy controller, which is implemented by using any one of the weld quality adaptive control systems based on a fuzzy controller, and the method comprises:

[0020] According to the specific scenario, set the required welding joint penetration depth and back weld width value;

[0021] Start the welding robot to weld the welded parts;

[0022] The molten pool information is collected through visual sensors, and the welding process parameter information is collected through arc sensors and robot controllers;

[0023] A prediction model for the weld joint penetration depth and weld joint backside penetration width was established through a neural network.

[0024] The welding process parameter information and molten pool information identified by the arc sensor, robot controller and visual sensor are used as inputs of the prediction model of the weld joint penetration depth and the weld joint back side penetration width;

[0025] The welding joint penetration depth and backside penetration width prediction model are used to predict the welding joint penetration depth and backside penetration width.

[0026] Compare the predicted values ​​output by the prediction model of the weld joint penetration depth and the weld joint backside weld width with the set weld joint penetration depth and backside weld width values ​​that meet the requirements to calculate the deviation and the deviation change, and merge them into the overall error and error change;

[0027] If the overall error between the predicted value and the set value and the error change meet the requirements, the welding process adjustment process is terminated;

[0028] If the overall error and error variation between the predicted value and the set value do not meet the requirements, the overall error and error variation will be input into the designed fuzzy controller, and the welding process parameters will be adjusted by the designed fuzzy controller to achieve the control of the welding joint penetration depth and the back welding width of the welding joint.

[0029] Preferably, establishing a prediction model for the weld joint penetration depth and the weld joint back side penetration width by a neural network includes:

[0030] In the process of using BP neural network to model the back weld width of the weld joint, the middle layer of the BP neural network is activated by S function, and the input and output samples of the BP neural network need to be normalized. The calculation formula is:

[0031] ;

[0032] in, For the The minimum value of the input quantity, For the The maximum value of the input quantity;

[0033] The BP neural network used is a three-layer structure, including an input layer, an intermediate layer, and an output layer; the intermediate layer uses a single intermediate layer neural network;

[0034] For the number of middle-layer neurons, 4-28 middle-layer neurons are usually selected; among them, for the prediction of the weld depth of the weld joint, the number of middle-layer neurons selected is 8; for the weld width on the back of the weld joint, the number of middle-layer neurons selected is 9.

[0035] Preferably, the predicted values ​​output by the prediction model of the weld joint penetration depth and the weld joint backside weld width are compared with the set weld joint penetration depth and backside weld width values ​​that meet the requirements to calculate the deviation and the deviation change, and merge them into the overall error and the error change, including:

[0036] In the process of welding joint penetration control, the deviation and deviation variation between the expected welding joint penetration and the online predicted welding joint penetration are used as system input variables, and the fuzzy controller adopts a multi-input and multi-output structure.

[0037] In the process of controlling the back molten width of the welding joint, the deviation and the deviation variation between the expected back molten width of the welding joint and the online predicted back molten width of the welding joint are used as the input variables of the fuzzy controller, and the fuzzy controller adopts a multi-input and multi-output structure.

[0038] Comprehensive consideration of weld joint penetration deviation and the deviation change And the weld width deviation on the back of the weld joint and the deviation change , setting error And the error change The input variable of the fuzzy controller is the current, welding gun inclination angle and nozzle height are the system outputs. The back molten width of the welding joint is controlled by adjusting the current, welding gun inclination angle and nozzle height so that the back molten width of the welding joint meets the set conditions.

[0039] Preferably, the core fuzzy query table and rule base of the fuzzy controller are set according to the experience gained during the actual welding operation;

[0040] The process of welding process adjustment is based on the values ​​set in the fuzzy query table, and the adjustment standard is to minimize the error between the set value and the predicted value of the welding joint penetration depth and back weld width and the error change by adjusting the actual welding process parameters.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1. The visual sensor is installed at the end of the welding gun, and can identify the welding molten pool during the welding process. However, due to the presence of the welding gun, the front end of the welding gun will block part of the molten pool image. The partial image of the molten pool identified by the visual sensor is used to extract the long semi-axis and short axis lengths of the molten pool image as independent variables, and the back molten width of the weld joint is used as the dependent variable to establish a prediction model for the back molten width of the weld joint. This approach does not require the restoration of the identified molten pool image, but only requires the long semi-axis length and short axis length information, with fast information processing speed and high reliability.

[0043] 2. In the process of controlling welding quality, the depth of weld joint and the width of weld on the back of weld joint are comprehensively considered. This method can ensure that the weld joint has sufficient depth of weld and penetrates the parent material, while avoiding the defect of weld sinking caused by excessive depth of weld.

[0044] 3. Through the multi-input multi-output fuzzy controller, the welding process parameters current, welding gun inclination and nozzle height are adjusted at the same time. During the adjustment process, the current needs to be adjusted through the welding machine, while the welding gun inclination and nozzle height need to be adjusted through the robot controller to control the robot to change its posture, so as to achieve the adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in 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 creative labor.

[0046] Figure 1 It is a schematic structural diagram of a weld quality adaptive control system based on a fuzzy controller according to an embodiment of the present invention;

[0047] Figure 2 A BP neural network structure diagram of an embodiment of the present invention;

[0048] Figure 3 This is a BP neural network training flow chart of an embodiment of the present invention;

[0049] Figure 4 Schematic diagram of an automatic control system for welding joint quality according to an embodiment of the present invention;

[0050] Figure 5 It is a basic principle block diagram of the fuzzy control system of an embodiment of the present invention;

[0051] Figure 6 4 is a structural diagram of a fuzzy controller according to an embodiment of the present invention.

[0052] Description of the drawings: 100 - welding power source, 200 - robot controller, 300 - host computer, 400 - structured light vision device, 500 - welding execution device. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present disclosure should be understood by people with ordinary skills in the field to which the present disclosure belongs. The "first", "second" and similar words used in the embodiments of the present disclosure do not indicate any order, quantity or importance, but are only used to distinguish different components. "Including" or "comprising" and similar words mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0055] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0056] Embodiment 1

[0057] From the background technology, we know that:

[0058] In the process of using robots for automated welding, the weld state changes continuously as welding progresses due to processing errors, assembly errors, and cyclic heat input during welding. In the initial stage of welding, the welding process parameters set according to the ideal weld state (the ideal state refers to the welding process parameters that meet the regulations and are formulated based on the specific material plate thickness and other known information after the welded parts are fixed) will become unreasonable as the weld state changes during the welding process (due to the heating process and stress during welding, the weld state will change, and the initially formulated welding process parameters will no longer be applicable). Unreasonable welding process parameters are prone to cause welding defects such as bubbles, cracks, slag inclusions, and incomplete penetration. The presence of these defects will reduce the cross-sectional area of ​​the weld joint, reduce the load-bearing capacity and fatigue strength of the weld joint, and thus cause brittle fracture of the weld joint.

[0059] like Figure 1As shown, an embodiment of the present invention provides a weld quality adaptive control system based on a fuzzy controller, which is mainly composed of a welding power supply 100, a robot controller 200, a host computer 300, a structured light vision device 400, and a welding execution device 500, wherein the welding execution device 500 includes a six-axis robot arm and a welding gun. The structured light vision device 400 is installed at the end of the welding gun and moves with the movement of the welding gun. During the welding process, the structured light vision device 400 identifies the molten pool during the welding process, and at the same time, the welding power supply 100 and the robot controller 200 provide feedback on the current welding process parameters.

[0060] In this embodiment, the robot controller 200 is not a fuzzy controller. The robot controller 200 controls the movement of the robot, and each robot has its own control cabinet, which contains controller hardware devices, etc. The fuzzy controller is in the host computer 300, that is, a piece of code, which outputs the corresponding change in the welding process parameters according to the change of the input variable.

[0061] In this embodiment, the structured light vision device 400 is a laser vision sensor, and the structured light vision device 400 identifies the molten pool during the welding process including:

[0062] During the welding process, the laser vision sensor installed at the end of the welding gun collects the molten pool image, performs image processing on the collected image, and extracts the short axis length and long semi-axis length of the molten pool image.

[0063] In this embodiment, during welding, the weld depth and the weld width on the back of the weld are set according to demand. During the welding process, the molten pool information during the welding process identified by the structured light vision device 400 and the current welding process parameters fed back by the welding power supply 100 and the robot controller 200 are input into the established neural network model, and the neural network model outputs the predicted values ​​of the weld depth and the weld width on the back of the weld.

[0064] Specifically, a neural network method was used to establish a high-precision prediction model for the welding joint penetration depth and the welding joint back side penetration width.

[0065] The neural network structure adopts a three-layer structure with multiple inputs and multiple outputs. The first layer is the input layer, which inputs welding parameter information and molten pool information, and the output layer outputs the predicted value of the weld depth of the weld joint and the predicted value of the weld width on the back of the weld joint. The neural network was trained based on the expected experimental data. During the forward propagation process, if the BP neural network fails to obtain the expected result in the output layer, it enters the back propagation stage. At this time, the error signal propagates backward along the network connection path, and the weights of the neurons in each layer are adjusted by applying the gradient descent method, so that the error is gradually reduced until the predetermined conditions are met. The purpose of this process is to approximate the expected output by continuously optimizing the weights. By continuously optimizing the weights of the intermediate layers of the neural network, the neural network error meets the requirements, thereby completing the training of the neural network.

[0066] In this embodiment, a multi-input and multi-output fuzzy controller is designed. According to the welding process parameter information identified by the arc sensor and the molten pool information identified by the visual sensor, a neural network prediction model is used to make real-time predictions on the welding joint penetration depth and the back molten width. The effects of the welding joint penetration depth and the back molten width on the welding quality are comprehensively considered, and the errors and error changes of the welding joint penetration depth and the back molten width are integrated and used as the input of the fuzzy controller. The fuzzy controller is used to adjust the current, welding gun inclination angle and nozzle height, thereby achieving control of the welding joint quality.

[0067] Specifically, in the process of welding joint penetration control, the deviation between the expected welding joint penetration and the online predicted welding joint penetration is used. and the deviation change The controller uses a multi-input and multi-output structure as the input variable of the system. In the process of controlling the back molten width of the weld joint, the deviation between the expected back molten width of the weld joint and the online predicted back molten width of the weld joint is used. and the deviation change is the system input variable, and the fuzzy controller adopts a multi-input and multi-output structure. The fuzzy controller structure diagram is shown in Figure 6 As shown in the figure, the fuzzy controller uses the deviation between the set value and the predicted value of the weld formation parameter and the deviation change As input variables, The difference between the current moment and the previous moment, plus the deviation change The purpose of using the fuzzy controller as an input variable is to consider the variation trend of the deviation, so as to more accurately control the welding process. In the entire system, the fuzzy controller outputs the control quantity as the welding process parameter adjustment value to adjust the welding process parameters. During the robot welding process, the welding process is monitored by sensors to obtain the welding process current, welding gun inclination, and nozzle height parameters. The established neural network prediction model is used to predict the weld joint penetration in real time and compare it with the preset value to calculate the weld joint penetration deviation. and the deviation change The molten pool information identified by the visual sensor is used to predict the back molten width of the weld joint in real time, and compared with the set value of the back molten width of the weld joint to calculate the back molten width deviation of the weld joint. and the deviation change , comprehensive calculation of weld formation quality deviation and the deviation change , and use it as the control input of the fuzzy controller to form a closed-loop control, realize the adaptive adjustment of welding process parameters, and thus control the quality of weld formation.

[0068] When using Figure 1 When the welding equipment shown is performing welding operation, when the arc is started and welding begins, the structured light vision sensor 400 installed at the end of the welding gun can identify the molten pool image during the welding process, and process the identified molten pool image to obtain the short axis and long semi-axis data of the molten pool, while the welding power supply 100 and the robot controller 200 provide feedback on the current welding process parameters (including: current, welding speed, nozzle height and welding gun inclination). The obtained molten pool information and welding process parameters are input into the trained neural network, and the neural network outputs the predicted value of the weld depth and the weld width on the back of the weld joint. The predicted value output by the neural network is compared with the predetermined value set according to the requirements, and the error and the error change are input into the fuzzy controller. According to the set fuzzy query table, the welding process parameter adjustment amount that needs to be adjusted is output to achieve adaptive adjustment of the welding process parameters.

[0069] Technical solution of the present invention:

[0070] The present invention discloses a weld quality intelligent adaptive control method based on a fuzzy controller, and designs a multi-input and multi-output fuzzy controller, which can output welding process parameter adjustment variables according to the error between the predicted value and the set value of the weld joint penetration depth and the weld width on the back of the weld joint, as well as the error change.

[0071] The present invention can effectively predict the molten pool information recognized by the structured light vision sensor and the welding process parameter information fed back by the welding power supply 100 and the robot controller 200 through a trained neural network, and the molten depth of the welding joint and the molten width on the back of the welding joint can be predicted.

[0072] The present invention can output welding process parameter adjustment variables through a multi-input and multi-output fuzzy controller, and adaptively adjust the welding process parameters according to information fed back by the structured light vision sensor, the welding power supply 100 and the robot controller 200, thereby achieving control of the welding quality.

[0073] Embodiment 2

[0074] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments, an embodiment of the present invention further provides a weld quality adaptive control method based on a fuzzy controller, which is implemented by applying any of the above-mentioned weld quality adaptive control systems based on a fuzzy controller, and the method includes:

[0075] According to the specific scenario, set the required welding joint penetration depth and back weld width value;

[0076] Start the welding robot to weld the welded parts;

[0077] The molten pool information is collected through the visual sensor, and the welding process parameter information is collected through the arc sensor and the robot controller 200;

[0078] A prediction model for the weld joint penetration depth and weld joint backside penetration width was established through a neural network.

[0079] The welding process parameter information and the molten pool information identified by the arc sensor, the robot controller 200 and the visual sensor are used as inputs of the prediction model of the weld joint penetration depth and the weld joint back side width;

[0080] The welding joint penetration depth and backside penetration width prediction model are used to predict the welding joint penetration depth and backside penetration width.

[0081] Compare the predicted values ​​output by the prediction model of the weld joint penetration depth and the weld joint backside weld width with the set weld joint penetration depth and backside weld width values ​​that meet the requirements to calculate the deviation and the deviation change, and merge them into the overall error and error change;

[0082] If the overall error between the predicted value and the set value and the error change meet the requirements, the welding process adjustment process is terminated;

[0083] If the overall error and error variation between the predicted value and the set value do not meet the requirements, the overall error and error variation will be input into the designed fuzzy controller, and the welding process parameters will be adjusted by the designed fuzzy controller to achieve the control of the welding joint penetration depth and the back welding width of the welding joint.

[0084] In this embodiment, the arc sensor is installed on the welding machine, and the welding machine can provide feedback on the current welding process parameters (including current). That is, the current welding process parameters can be directly extracted and used. The remaining welding process parameters (including welding speed, welding gun inclination angle and nozzle height) are fed back through the robot controller 200. The welding speed is the movement speed of the robot end, and the welding gun inclination angle and nozzle height are obtained through the robot end posture.

[0085] In this embodiment, the identified molten pool information includes the minor axis length and the major semi-axis length of the molten pool.

[0086] In this embodiment, before welding begins, the welding process parameters are set according to the specific conditions of welding. The main purpose of setting the welding process parameters is to ensure sufficient penetration and weld width on the back of the weld joint. The size of these two values ​​directly affects the quality of the weld joint. Insufficient penetration indicates that the parent material has not been fully melted, and insufficient weld width on the back of the weld joint indicates that there is too little molten metal during the welding process and a strong connection strength cannot be guaranteed. Once the welding material, plate thickness, weld gap and other conditions are determined, the required weld penetration and weld width on the back of the weld joint are set according to the required welding quality. This value is the value to be achieved. The predicted value is the current weld penetration and weld width on the back of the weld joint predicted according to the actual welding state during the welding process.

[0087] Specifically, the forming parameter prediction model based on BP neural network:

[0088] BP Neural Network (Backpropagation Neural Network, BPNN) is an error back propagation neural network proposed by Rumelhart in 1986. Its structure is as follows Figure 2 As shown, is the input layer, For the middle layer, The input information is input by the input layer neurons, processed by the intermediate layer neurons, and transmitted to the output layer. BP neural network is one of the most classic and widely used models in artificial neural networks. A large number of studies have confirmed that it not only has powerful pattern recognition capabilities, but also can excellently complete the mapping task of multidimensional functions, thereby effectively solving complex problems. This network has shown excellent performance in a wide range of application scenarios.

[0089] In the BP neural network learning process, the input layer has neurons, and the middle layer has neurons, and the output layer has neurons, then the input of the middle layer neurons is:

[0090]

[0091] in, is the middle layer weight, Input to the input layer.

[0092] Output of neurons in the middle layer for:

[0093]

[0094] but

[0095]

[0096] The output of the neuron in the output layer of the BP neural network is:

[0097]

[0098] During the forward propagation process, if the BP neural network fails to obtain the expected result in the output layer, it enters the back propagation stage. At this time, the error signal propagates backward along the network connection path, and the weights of neurons in each layer are adjusted by applying the gradient descent method to gradually reduce the error until the predetermined conditions are met. The purpose of this process is to approach the expected output by continuously optimizing the weights.

[0099] The difference between the BP network outputs is:

[0100]

[0101] in, is the expected value of the output layer.

[0102] The error criterion function is:

[0103]

[0104] Weight The update method is:

[0105]

[0106] In the formula, is the learning rate of the network. Usually, .

[0107] The time weight update method is:

[0108]

[0109] Connection weight The update method is:

[0110]

[0111] In the formula, .

[0112] The network weight at this moment is:

[0113]

[0114] In order to avoid oscillation or slow convergence during weight learning, it is necessary to consider the impact of the previous weight change on the current weight adjustment, so the momentum factor needs to be introduced. This strategy can speed up the convergence process and improve the stability of training. The weights at this time are:

[0115]

[0116]

[0117] In the formula, is the momentum factor, .

[0118] Take the first input of the BP neural network as ,Right now ,but:

[0119]

[0120] Establishment of BP neural network model for welding forming parameters: In the process of using BP neural network to model the backside weld width of welding joint, since the middle layer of BP neural network uses S function for activation, it is necessary to normalize the input and output samples of BP neural network. Normalization uses the following formula:

[0121]

[0122] in, For the The minimum value of the input quantity, For the The maximum value of the input quantity.

[0123] After normalization, the input and output samples of the BP neural network are shown in Table 3.

[0124] The BP neural network used is a three-layer structure, including input layer, middle layer and output layer. The BP neural network training process is as follows Figure 3 As shown. Since weld formation prediction is a relatively simple regression problem, a single hidden layer can already achieve continuous mapping from a finite space to another finite space, which is sufficient to meet the requirements of this task. Therefore, a single intermediate layer neural network model is adopted. For the number of intermediate layer neurons, 4-28 intermediate layer neurons are usually selected. In order to determine the number of intermediate layer neurons, different numbers of intermediate layer neurons are set, and the root mean square error between the predicted value and the true value of the forming parameter is calculated. The root mean square error table is shown in Table 1 and Table 2:

[0125] Table 1 Root mean square error between predicted and true values ​​of weld joint penetration

[0126]

[0127] Table 2 Root mean square error between the predicted value and the true value of the weld width on the back of the weld joint

[0128]

[0129] Table 3. Normalized neural network input and output samples

[0130]

[0131] According to Table 1, for the prediction of the penetration depth of the welding joint, as the number of neurons in the middle layer increases, the root mean square error gradually decreases. The minimum root mean square error occurs when the number of neurons in the middle layer is 8, and the minimum root mean square error is 0.63511. However, when the number of neurons in the middle layer reaches 9, the root mean square error increases. Although the root mean square error decreases when the number of neurons in the middle layer reaches 11, considering the calculation speed and the root mean square error, the number of neurons in the middle layer is selected as 8.

[0132] According to Table 2, for the weld width on the back side of the weld joint, as the number of neurons in the middle layer increases, the root mean square error also gradually decreases. However, when the number of neurons in the middle layer reaches 10, the root mean square error increases. Although the root mean square error decreases when the number of neurons in the middle layer reaches 12, considering the calculation speed and the root mean square error, the number of neurons in the middle layer is selected as 9.

[0133] In this embodiment, the predicted values ​​output by the prediction model of the weld joint penetration depth and the weld joint backside weld width are compared with the set weld joint penetration depth and backside weld width values ​​that meet the requirements to calculate the deviation and the deviation change, and merge them into the overall error and the error change, including:

[0134] In the process of welding joint penetration control, the deviation and deviation variation between the expected welding joint penetration and the online predicted welding joint penetration are used as system input variables, and the fuzzy controller adopts a multi-input and multi-output structure.

[0135] In the process of controlling the back molten width of the welding joint, the deviation and the deviation variation between the expected back molten width of the welding joint and the online predicted back molten width of the welding joint are used as the system input variables, and the fuzzy controller adopts a multi-input and multi-output structure.

[0136] Comprehensive consideration of weld joint penetration deviation and the deviation change And the weld width deviation on the back of the weld joint and the deviation change , setting error And the error change The system input variables are the current, welding gun inclination angle and nozzle height, and the back molten width of the welding joint is controlled by adjusting the current, welding gun inclination angle and nozzle height so that the back molten width of the welding joint meets the set conditions.

[0137] In this embodiment, the core fuzzy query table and rule base of the fuzzy controller are set according to the experience gained in the actual welding operation process. For example, if the welding joint has a large penetration depth, the welding current needs to be reduced, which is an empirical rule.

[0138] The process of welding process adjustment is based on the values ​​set in the fuzzy query table. The standard of adjustment is to minimize the error between the set value and the predicted value of the welding joint penetration depth and back weld width when adjusting the actual welding process parameters, as well as the error change.

[0139] In the control of the welding process by the fuzzy controller, the deviation and the deviation change are input into the fuzzy controller, and the fuzzy controller outputs the corresponding adjustment amount through the fuzzy rule base and the fuzzy query table. The current welding process parameters are adjusted by the adjustment amount, thereby adjusting the welding parameters to control the welding quality.

[0140] Specifically, in the intelligent welding process, the most important link is to correct the welding process parameters in real time through the controller to control the quality of the welded joint. Based on the welded joint forming parameter prediction model, an automatic control system for welded joint quality is established. The control system structure is as follows: Figure 4As shown. For butt welds, it is ultimately necessary to ensure that the weld joint has a certain depth of penetration and back molten width to meet the welding quality grade requirements. Therefore, the depth of penetration of the weld joint and the back molten width of the weld joint belong to the controlled quantities, while the welding process parameters and the molten pool information belong to the controlled quantities. During the welding process, in order to ensure the stability of the welding process, the welding speed is usually constant. The adjustment of the welding heat input by the welding speed can be achieved by changing the current. And the molten pool geometry is closely related to the welding process parameters. Therefore, in the actual welding process, only the current, welding gun inclination and nozzle height need to be adjusted to achieve the control of the quality of the welded joint during the welding process. Among them, the information processing module mainly processes the molten pool image of the welding process identified by the visual sensor, extracts the long semi-axis and short axis length information of the molten pool, and extracts the welding process parameters identified by the arc sensor, and uniformly inputs all the identified parameters into the neural network prediction model.

[0141] Fuzzy Control is an intelligent control method based on fuzzy set theory, fuzzy language variables and fuzzy logic reasoning, which simulates the process of human fuzzy reasoning and decision-making. This method effectively imitates human judgment and decision-making behavior under uncertain and fuzzy conditions. The birth of fuzzy set theory solves the mapping problem between clear numerical values ​​and fuzzy concepts. Fuzzy mathematics based on this establishes a connection between the fuzzy things in reality and classical mathematics, so that fuzzy problems can be expressed in a quantitative way, thereby using mathematical methods to reveal their essence and laws. The reasoning results are then applied to the actuator to achieve system control and operation. Figure 5 The basic principle block diagram of the fuzzy control system is shown. The core component is the fuzzy controller. The part in the dotted box in the block diagram represents the controller, and its control rules are implemented through computer programs. This controller is the key module for the system to implement fuzzy reasoning.

[0142] As can be seen from the figure, the dotted box is the structure of the controller, which includes a calculation control variable module, a fuzzy quantization processing module, a fuzzy rule module, a fuzzy decision module, and a non-fuzzy processing module. In the process of welding joint penetration control, the deviation between the expected welding joint penetration and the online predicted welding joint penetration is used. and the deviation change The controller uses a multi-input and multi-output structure as the input variable of the system. In the process of controlling the back molten width of the weld joint, the deviation between the expected back molten width of the weld joint and the online predicted back molten width of the weld joint is used. and the deviation change The input variables of the system are used, and the controller adopts a multi-input multi-output structure.

[0143] As shown in Figure 5, the core of the fuzzy system lies in the design and construction of the fuzzy controller. The design process is as follows:

[0144] 1. Determine the structure of the controller and select appropriate input and output variables.

[0145] According to the variables to be controlled in this paper, a multi-input multi-output fuzzy controller is used to comprehensively consider the welding joint penetration deviation. and the deviation change And the weld width deviation on the back of the weld joint and the deviation change , setting error And the error change The system input variables are the current, welding gun inclination angle and nozzle height, and the current, welding gun inclination angle and nozzle height are the system outputs. The back molten width of the welding joint is controlled by adjusting the current, welding gun inclination angle and nozzle height to meet the set conditions.

[0146] 2. Define input and output fuzzy sets.

[0147] In the design and construction of fuzzy control systems, precise values ​​are converted into fuzzy quantities through fuzzification. In a multi-input multi-output fuzzy controller, the choice of deviation and the deviation change As input variables, and The range of change is defined as the basic domain of fuzzy language, and the domain of its fuzzy set is expressed as:

[0148]

[0149] in and For discretization in and The number of files formed after the continuity error is combined into the domain and Elements, generally speaking, and The values ​​of are consistent. Fuzzy subsets are usually divided in the following three ways:

[0150]

[0151] 3. Define input and output membership functions

[0152] After determining the fuzzy sets and domains of deviation, deviation change, and control quantity, it is necessary to select appropriate membership functions for the fuzzy variables. This step involves assigning specific values ​​to the fuzzy variables and setting the membership of the elements in the domain to ensure that the system can correctly process the input data. In order to meet the actual work needs and take into account the simplicity of calculation, the objective data is usually approximated as a common analytical function form, which is convenient for selection according to needs. Commonly used membership functions include triangular function, bell function, Gaussian function, trapezoidal function and Sigmoid function. Considering the convenience and speed of calculation, the triangular function is selected as the membership function. The basic form of the triangular function is as follows:

[0153]

[0154] in, .

[0155] 4. Establish fuzzy rules

[0156] Based on human intuitive reasoning, fuzzy rules for eliminating system errors are designed through system output errors and their variations. These rules constitute fuzzy models that describe various controlled processes. During the welding process, operators can adjust welding parameters in real time by observing the shape of the molten pool, thereby achieving control of welding quality. In quality control, welders observe the size of the molten pool and related parameters intuitively. If the molten pool is too large, the welding heat input is reduced, and if the molten pool is too small, the heat input is increased. Fuzzy rules are established by imitating this thinking and reasoning process.

[0157] 5. Establish fuzzy rule table

[0158] Based on the mentioned, fuzzy rules can be represented by a fuzzy rule table. Table 4 shows the rule table, which contains 49 fuzzy rules.

[0159] Table 4 Fuzzy rules table

[0160]

[0161] 6. Fuzzy Reasoning

[0162] Fuzzy reasoning is the core of fuzzy control system, which uses fuzzy reasoning algorithm combined with fuzzy rules to calculate and finally get the result of control quantity. This process is the key step to realize fuzzy control.

[0163] 7. Defuzzification

[0164] The output of fuzzy reasoning is a fuzzy set, but in actual control, it needs to be converted into a specific numerical value in order to control or drive the welding robot system to adjust the process parameters in real time. The process of converting the results of fuzzy reasoning into precise values ​​is called defuzzification. There are three common defuzzification methods, namely the maximum membership method, the center of gravity method, and the weighted average method. These methods are suitable for different application scenarios and should be selected according to needs. The weighted average method is used to transform the results of fuzzy reasoning. The weighted average formula is as follows:

[0165]

[0166] Among them, the coefficient The choice depends on the actual situation.

[0167] Fuzzy query table establishment: In order to ensure that the welding depth and the back molten width of the welding joint reach a certain value, thereby ensuring the welding quality, the welding parameters need to be adjusted during the welding process. In the actual welding process, the welding depth and the back molten width of the welding joint are related to the welding process parameters (current, welding gun inclination, nozzle height). Therefore, a multi-input multi-output fuzzy controller is used to control the welding process, and the deviation is selected. and the deviation change As input variables, current A, welding gun inclination C, and nozzle height D are selected as output variables.

[0168] when deviation is negative (NB), the deviation change is negative (NB), the output variable For Zhengda (PB), For Zhengda (PB), is negative (NB); when the deviation is negative (PB), the deviation change is negative (PB), the output variable For Zhengda (NB), For Zhengda (NB), The complete fuzzy control query table is shown in Table 5, Table 6, and Table 7:

[0169] Table 5 Fuzzy rule table

[0170]

[0171] Table 6 Fuzzy rule table

[0172]

[0173] Table 7 Fuzzy rule table

[0174]

[0175] Based on the previous experiments, the deviation The range of variation is: , the deviation change The range of variation is: , the range of change of current A is: , the range of the change of the welding gun inclination angle C is: , the range of the nozzle height D is: The fuzzy control query tables generated by the fuzzy controller toolbox in MATLAB are shown in Table 8, Table 9, and Table 10:

[0176] Table 8 Fuzzy rule query table

[0177]

[0178] Table 9 Fuzzy rule query table

[0179]

[0180] Table 10 Fuzzy rule query table

[0181]

[0182] The embodiments of the present invention are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present disclosure should be included in the scope of protection of the present disclosure.

Claims

1. A weld quality adaptive control system based on fuzzy controller, characterized in that: The system includes: a welding power source, a robot controller, a host computer, a structured light vision device, and a welding execution device; The welding execution device includes: a six-axis robot arm and a welding gun; The structured light vision device is installed at the end of the welding gun and moves with the movement of the welding gun; During the welding process, the structured light vision device identifies the molten pool during the welding process. At the same time, the welding power supply and robot controller provide feedback on the current welding process parameters and read them through the host computer.

2. The system according to claim 1, characterized in that The structured light vision device is a laser vision sensor. The structured light vision device identifies the molten pool during the welding process, including: During the welding process, the laser vision sensor installed at the end of the welding gun collects the molten pool image, performs image processing on the collected image, and extracts the short axis length and long semi-axis length of the molten pool image.

3. The system according to claim 2, characterized in that During welding, the penetration depth of the welding joint and the molten width on the back of the welding joint are set according to the requirements. During the welding process, the molten pool information in the welding process identified by the structured light vision device and the current welding process parameters fed back by the welding power supply and the robot controller are input into the established neural network model, and the neural network model outputs the predicted values ​​of the penetration depth of the welding joint and the molten width on the back of the welding joint; Among them, the neural network model is to establish a prediction model for the welding joint penetration depth and the welding joint back side penetration width through the BP neural network.

4. The system according to claim 3, characterized in that The welding process parameters are adjusted by outputting the welding process parameters adjustment variables through the multi-input multi-output fuzzy controller, and the welding process parameters are adaptively adjusted according to the information fed back by the structured light vision sensor, welding power supply and robot controller to achieve the control of welding quality. Among them, in the process of welding joint penetration control, the deviation between the expected welding joint penetration and the online predicted welding joint penetration is used. and the deviation change is the input variable of the fuzzy controller; In the process of back weld width control of weld joint, the deviation between the expected back weld width of weld joint and the online predicted back weld width of weld joint is used. and the deviation change is the input variable of the fuzzy controller.

5. A method for adaptive control of weld quality based on a fuzzy controller, which is implemented by using the adaptive control system for weld quality based on a fuzzy controller according to any one of claims 1 to 4, characterized in that: The method comprises: According to the specific scenario, set the required welding joint penetration depth and back weld width value; Start the welding robot to weld the welded parts; The molten pool information is collected through visual sensors, and the welding process parameter information is collected through arc sensors and robot controllers; A prediction model for the weld joint penetration depth and weld joint backside penetration width was established through a neural network. The welding process parameter information and molten pool information identified by the arc sensor, robot controller and visual sensor are used as inputs of the prediction model of the weld joint penetration depth and the weld joint back side penetration width; The welding joint penetration depth and backside penetration width prediction model are used to predict the welding joint penetration depth and backside penetration width. Compare the predicted values ​​output by the prediction model of the weld joint penetration depth and the weld joint backside weld width with the set weld joint penetration depth and backside weld width values ​​that meet the requirements to calculate the deviation and the deviation change, and merge them into the overall error and error change; If the overall error between the predicted value and the set value and the error change meet the requirements, the welding process adjustment process is terminated; If the overall error and error variation between the predicted value and the set value do not meet the requirements, the overall error and error variation will be input into the designed fuzzy controller, and the welding process parameters will be adjusted by the designed fuzzy controller to achieve the control of the welding joint penetration depth and the back welding width of the welding joint.

6. The method according to claim 5, characterized in that The prediction model of weld joint penetration depth and weld joint backside penetration width established by neural network includes: In the process of using BP neural network to model the back weld width of the weld joint, the middle layer of the BP neural network is activated by S function, and the input and output samples of the BP neural network need to be normalized. The calculation formula is: ； in, For the The minimum value of the input quantity, For the The maximum value of the input quantity; The BP neural network used is a three-layer structure, including an input layer, an intermediate layer, and an output layer; the intermediate layer uses a single intermediate layer neural network; For the number of middle-layer neurons, 4-28 middle-layer neurons are usually selected; among them, for the prediction of the weld depth of the weld joint, the number of middle-layer neurons selected is 8; for the weld width on the back of the weld joint, the number of middle-layer neurons selected is 9.

7. The method according to claim 5, characterized in that The predicted values ​​output by the prediction model of weld joint penetration and weld joint backside weld width are compared with the set weld joint penetration and backside weld width values ​​that meet the requirements to calculate the deviation and deviation change, and merge them into the overall error and error change, including: In the process of welding joint penetration control, the deviation and deviation variation between the expected welding joint penetration and the online predicted welding joint penetration are used as system input variables, and the fuzzy controller adopts a multi-input and multi-output structure. In the process of controlling the back molten width of the welding joint, the deviation and the deviation variation between the expected back molten width of the welding joint and the online predicted back molten width of the welding joint are used as the input variables of the fuzzy controller, and the fuzzy controller adopts a multi-input and multi-output structure. Comprehensive consideration of weld joint penetration deviation and the deviation change And the weld width deviation on the back of the weld joint and the deviation change , setting error And the error change The input variable of the fuzzy controller is the current, welding gun inclination angle and nozzle height are the system outputs. The back molten width of the welding joint is controlled by adjusting the current, welding gun inclination angle and nozzle height so that the back molten width of the welding joint meets the set conditions.

8. The method according to claim 5, characterized in that The core fuzzy query table and rule base of the fuzzy controller are set according to the experience gained during the actual welding operation; The process of welding process adjustment is based on the values ​​set in the fuzzy query table, and the adjustment standard is to minimize the error between the set value and the predicted value of the welding joint penetration depth and back weld width and the error change by adjusting the actual welding process parameters.

Images