A composite welding lap weld penetration control method and device

Abstract

The present invention discloses a method and device for controlling the penetration depth of a composite welding lap joint, belongs to the field of welding technology, and solves the problem of online control of the penetration depth of a lap joint in a laser-arc composite welding process. The present invention discloses a method for controlling the penetration depth of a composite welding lap joint, characterized in that it comprises the following steps: S1, adjusting the posture of a laser-arc composite welding gun; S2, starting welding under the conditions that a wire feed speed is V0 and a welding speed is V1; S3, starting welding under the welding parameter conditions of S2 to obtain a data set; S4, building a deep learning model; S5, starting welding under the welding parameter conditions of S2 to achieve online control of the penetration depth of the lap joint structure weld. The present invention introduces a method combining visual sensing with a deep learning model into the process of laser-arc composite welding lap joints, thereby achieving real-time control of the penetration depth of laser-arc composite welding lap joints.

Description

Technical Field

[0001] The present invention relates to the technical field of welding, and in particular to a method and device for controlling the penetration depth of a composite welding lap weld. Background Art

[0002] Due to the synergistic effect of laser and arc, laser-arc hybrid welding fully utilizes the advantages of two heat sources, can significantly improve arc stability and laser energy absorption efficiency, reduce tooling assembly precision requirements, and has broad application prospects in the welding of non-ferrous metals and medium and thick plates.

[0003] Lap welds are a typical weld type in the mechanical manufacturing industry. Characterized by partial overlap of the joined parts at the joint, full penetration of the backing structure is not required. Direct observation of weld penetration in lap joints is difficult, and real-time penetration values ​​are impossible to obtain, making online control of weld penetration challenging. In particular, the weld penetration requirements for lap joints vary continuously as plate thickness changes along the weld path. Therefore, solutions are essential to address the challenges of stable weld penetration control in laser-arc hybrid welding of lap joints.

[0004] At present, there are two main methods to solve the problem of weld penetration control: (1) Under the condition of known weld penetration requirements, the heat source power or welding speed is manually controlled according to the relationship between weld penetration and heat input parameters. However, this method is difficult to apply on a large scale in welding production lines and has low reliability; (2) A regression model is constructed between heat source parameters and weld penetration. According to the different changes in weld depth requirements, the optimal heat source processing parameters are calculated with the help of the regression model. The Chinese invention patent application number 201710798902.8, "Method and system for predicting and controlling penetration of laser welding of thickened plate", proposes to establish a complete quadratic polynomial regression model between laser power and plate thickness. The plate thickness values ​​at different positions are input into the established regression model to calculate the optimal laser power. However, this method takes fewer factors into consideration, and the method of establishing a regression model can only be applied to specific welding conditions. Therefore, it is necessary to seek other methods to achieve precise control of weld penetration. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention provides a hybrid welding lap weld penetration control method, the purpose of which is to solve the problem of online control of lap weld penetration in the laser-arc hybrid welding process.

[0006] The technical solution adopted in the present invention is as follows:

[0007] A method for controlling the penetration depth of a composite welding lap weld comprises the following steps:

[0008] S1. Adjust the posture of the laser-arc hybrid welding gun so that the angle between the arc welding gun and the substrate is θ1, the dry extension of the welding wire is d1, the distance between the end of the welding wire and the focus of the laser beam on the substrate is d2, the angle between the laser beam and the arc welding gun is θ2, the tilt angle of the CCD camera is θ3, and the distance between the CCD camera lens and the molten pool is d3;

[0009] S2. Start welding with the wire feeding speed V0 and welding speed V1. Increase the laser power P gradually from 0 until the lap weld is fully penetrated. Record the maximum laser power P at this time. max , cut the weld along the welding direction, observe the position where the upper lap plate begins to melt through, and record the laser power value P at this position min In [P min , P max ] arbitrarily select the laser power P0 as the basic working point within the interval;

[0010] S3, start welding under the welding parameter conditions of S2, set the laser power P0 from P min Gradually increase to P max , and open the CCD camera to collect the front image of the molten pool in real time. The acquisition frame rate of the CCD camera is f. The weld is cut along the welding direction. The weld penetration value d is measured every 2 mm from the welding starting point. The linear interpolation method is used to calculate the d value at the weld position corresponding to each frame of the molten pool front image. The d value is used to label the molten pool front image to obtain the data set;

[0011] S4. Build a deep learning model, use the front image of the melt pool in the dataset as input, and the d value as output to train the deep learning model. Optimize the model by adjusting the hyperparameters of the deep learning model, and save and call the optimized deep learning model online.

[0012] S5. Start welding under the welding parameters of S2, turn on the CCD camera to collect the front image of the molten pool in real time, crop the front image of the molten pool into the region of interest (ROI), and input the ROI image into the deep learning model to obtain the current d p When the laser-arc hybrid welding control device is turned on after welding time t, where t is set to 0.5-1s, the closed-loop controller is based on the predicted value of the penetration depth d p The positive and negative value and size of the deviation value e(t) from the set value ds of the penetration depth are determined. At the same time, feedback control is performed according to the changing trend of the weld penetration depth deviation Δe(t)=e(t)-e(t-1). An adjustment signal is output to the laser to change the laser power P, where e(t-1) is the weld penetration depth deviation at the previous control moment. After welding is completed, the laser-arc composite heat source welding equipment is turned off to realize online control of the weld penetration of the lap structure.

[0013] Preferably, the value range of the angle θ1 in S1 is 60°-80°, the value range of d1 is 10-16 mm, the value range of θ2 is 70°-85°, the value range of θ3 is 45°-70°, the value range of d2 is 2-4 mm, and the value range of d3 is 130-180 mm.

[0014] Preferably, the weld penetration d value corresponding to the weld position in the molten pool front image in S3 is calculated using a linear interpolation method: d=d A +x×(d B –d A ) / 2, where d A and d B is the weld penetration corresponding to two adjacent points P1 and P2, respectively. x is the distance between the current calculation frame and point P1. x is calculated as: x = k × 1 / f × V1, where k is the number of frames from P1 to the current calculation frame. The dataset is randomly divided into training and test sets according to the ratio β:(1-β).

[0015] Furthermore, the value range of the parameter β is 0.7-0.85.

[0016] Preferably, the deep learning model of S4 comprises a convolutional neural network with 1 input layer, n convolutional layers, n pooling layers, m fully connected layers and 1 linear regression layer.

[0017] Preferably, the deep learning model in S4 is tuned by using a root mean square loss function to judge the deviation between the true value and the predicted value, using a gradient momentum descent optimizer to update the model weight coefficient, and using a test set to perform an accuracy test on the trained deep learning model; the adjustable hyperparameters include learning rate, number of hidden layers, activation function, and number of neurons.

[0018] Furthermore, the value range of the parameter f in S4 is 20 Hz-50 Hz.

[0019] Furthermore, the value range of the parameter n in S4 is 3-7, where n is an integer; the value range of the parameter m is 1-4, where m is an integer.

[0020] Preferably, the resolution of the molten pool front image collected by the CCD camera in S5 after ROI cropping is 750 pixels×300 pixels.

[0021] A composite welding lap weld penetration control device comprises: a base plate, an arc power supply, an arc welding gun, a laser, a control terminal, a CCD camera, a control box, and a laser welding torch; wherein the negative pole of the arc power supply is connected to the base plate, and the positive pole is connected to the arc welding gun; the CCD camera is connected to the control terminal via a USB data interface; the CCD camera, the arc welding gun, and the laser welding torch are connected via a fixture; and the laser and the control box are connected to the control terminal via a data acquisition card; the CCD camera collects a front image of the molten pool in real time, and the control terminal inputs the front image of the molten pool into a deep learning model through ROI cropping to predict the weld penetration; the closed-loop control system in the control terminal performs feedback control based on the error between the predicted weld penetration value and the set weld penetration value, sends a signal to change the output laser power of the laser, and controls the weld penetration online.

[0022] In summary, the beneficial effects of the present invention are:

[0023] During the laser-arc hybrid welding process, a visual sensing method is used to collect the front image of the molten pool, and the front image of the molten pool is used as the input of the deep learning model to achieve real-time feedback on the penetration depth of the lap weld. Compared with the traditional method, the present invention introduces the method of combining visual sensing with a deep learning model into the laser-arc hybrid welding lap weld process, providing a new idea for solving the problem that the penetration depth of the lap structure weld is difficult to directly observe during the welding process, and realizes real-time control of the penetration depth of the laser-arc hybrid welding lap weld. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The present invention will now be described by way of example with reference to the accompanying drawings, in which:

[0025] Figure 1 This is a schematic diagram of the installation position structure of the CCD camera and heat source of the present invention;

[0026] Figure 2 Schematic diagram of a typical laser-arc hybrid welding overlap structure;

[0027] Figure 3 Schematic diagram of the laser-arc hybrid welding lap weld penetration control device of the present invention;

[0028] Diagram: 1-substrate, 2-arc power supply, 3-arc welding gun, 4-laser, 5-control terminal, 6-CCD camera, 7-control box, 8-laser welding torch. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of this application.

[0030] The following combination Figure 1-Figure 3 The present invention is described in detail.

[0031] Example

[0032] like Figure 1 As shown in the figure, during the laser-arc hybrid welding process of lap welds, a CCD camera is used to collect the front image of the molten pool, and the weld penetration value is used to label the front image of the molten pool to obtain a data set. The constructed deep learning model is trained using the data set. During the welding process, the front image of the molten pool is input into the trained deep learning model to predict the weld penetration value. The control terminal control system adjusts the laser power according to the error between the predicted penetration value and the set penetration value, changes the heat input of the laser-arc composite heat source, and realizes real-time control of the lap weld penetration.

[0033] The lap weld described is a joint where the welded components partially overlap at the joint, with the heat source acting directly on the upper lap plate surface of the lap joint structure. During the laser-arc hybrid welding process, a CCD camera captures a frontal image of the molten pool and labels it with the corresponding weld penetration value d to obtain a dataset. A deep learning model is then built and trained and optimized using the dataset. During the welding process, the frontal image of the molten pool is used as input to the deep learning model to predict the weld penetration value. The error between the predicted penetration value and the set weld penetration value is used as the input signal of a closed-loop controller, which adjusts the laser power to resolve the unstable weld penetration problem of the lap joint structure and achieve online control of the weld penetration of the laser-arc hybrid welding lap joint.

[0034] S1. Adjust the laser-arc hybrid welding gun posture so that the angle between the arc welding gun and the substrate is θ1, the dry extension of the welding wire is d1, the distance between the end of the welding wire and the laser beam focus on the substrate is d2, the angle between the laser beam and the arc welding gun is θ2, the tilt angle of the CCD camera is θ3, the distance between the CCD camera lens and the molten pool is d3, and the overlap plate thickness is H;

[0035] S2. Start welding with the wire feeding speed V0 and welding speed V1. Increase the laser power P gradually from 0 until the lap weld is fully penetrated. Record the maximum laser power P at this time. max , cut the weld along the welding direction, observe the position where the upper lap plate begins to melt through, and record the laser power value P at this position min In [P min , P max ] arbitrarily select the laser power P0 as the basic working point within the interval;

[0036] S3, start welding under the welding parameter conditions of S2, set the laser power P0 from P min Gradually increase to P max , and open the CCD camera to collect the front image of the molten pool in real time. The acquisition frame rate of the CCD camera is f. The weld is cut along the welding direction. The weld penetration value d is measured every 2 mm from the welding starting point. The linear interpolation method is used to calculate the d value at the weld position corresponding to each frame of the molten pool front image. The d value is used to label the molten pool front image to obtain the data set;

[0037] S4. Design a deep learning model, use the front image of the melt pool in the dataset as input and the d value as output to train the deep learning model, optimize the model by adjusting the hyperparameters of the deep learning model, save the optimized deep learning model and call it online;

[0038] S5. Start welding under the welding parameters of S2, turn on the CCD camera to collect the front image of the molten pool in real time, crop the front image of the molten pool into the region of interest (ROI), and input the ROI image into the deep learning model to obtain the current d p When the laser-arc hybrid welding control device is turned on after welding time t, where t is set to 0.5-1s, the closed-loop controller is based on the predicted value of the penetration depth d p The positive and negative value and magnitude of the deviation value e(t) from the set value ds of the penetration depth are controlled. At the same time, feedback control is performed based on the variation trend of the weld penetration depth deviation Δe(t)=e(t)-e(t-1). An adjustment signal is output to the laser to change the laser power P, where e(t-1) is the weld penetration depth deviation at the previous control moment. After welding is completed, the laser-arc hybrid heat source welding equipment is turned off to achieve online control of the weld penetration of the overlap structure.

[0039] As a preferred embodiment, the angle θ1 described in S1 has a value range of 60°-80°, the value range of d1 is 10-16 mm, the value range of θ2 is 70°-85°, the value range of θ3 is 45°-70°, the value range of d2 is 2-4 mm, and the value range of d3 is 130-180 mm. The value range of the parameter β described in S3 is 0.7-0.85. The value range of the parameter f described in S4 is 20 Hz-50 Hz. The resolution of the CCD camera described in S5 after collecting the molten pool front image ROI cropped is 750 pixels × 300 pixels.

[0040] As a preferred embodiment, the laser-arc hybrid welding equipment includes: a substrate 1, an arc power supply 2, an arc welding gun 3, a laser (4), a control terminal 5, a CCD camera 6, a control box 7, and a laser welding torch 8; wherein the negative pole of the arc power supply 2 is connected to the substrate 1, and the positive pole is connected to the arc welding gun (3); the CCD camera 6 is connected to the control terminal 5 via a USB data interface; the CCD camera 6, the arc welding gun 3 and the laser welding torch 8 are connected via a clamp; the laser 4 and the control box 7 are connected to the control terminal 5 via a data acquisition card; the CCD camera 6 collects the front image of the molten pool in real time, and the control terminal 5 inputs the front image of the molten pool into the deep learning model through ROI clipping to predict the weld penetration depth; the closed-loop control system in the control terminal 5 performs feedback control according to the error between the weld penetration prediction value and the weld penetration set value, and sends a signal to change the output laser power of the laser 4, thereby realizing online control of the weld penetration depth.

[0041] The above-described embodiments merely represent specific implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of protection of the present application. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the technical concept of the present application, and all such variations and improvements fall within the scope of protection of the present application.

Claims

1. A method for controlling the penetration depth of a composite lap weld, characterized in that: The steps include: S1. Adjust the laser-arc hybrid welding gun posture so that the angle between the arc welding gun and the substrate is θ 1. Wire extension length d 1. The distance between the end of the welding wire and the focus of the laser beam on the substrate is d 2. The angle between the laser beam and the arc welding gun is θ 2. The tilt angle of the CCD camera is θ 3. The distance between the CCD camera lens and the molten pool is d 3; S2, when the wire feeding speed is V 0 , welding speed is V 1 Start welding under the condition of P Start from 0 and gradually increase until the lap weld is fully penetrated, and record the maximum laser power at this time P max , cut the weld along the welding direction, observe the position where the upper lap plate begins to melt through, and record the laser power value here P min， exist[ P min , P max ] Select the laser power arbitrarily within the range P 0 As the basic working point; S3. Start welding under the welding parameters of S2 and set the laser power P 0 Depend on P min Gradually increase to P max , and open the CCD camera to collect the front image of the molten pool in real time. The frame rate of the CCD camera is f , cut the weld along the welding direction, and measure the weld penetration value at intervals from the welding starting point d , and use linear interpolation to calculate the weld position corresponding to each frame of the weld pool front image d Value, use d The values ​​are used to label the front images of the melt pool to obtain the dataset; S4. Build a deep learning model, using the front image of the melt pool in the dataset as input. d The deep learning model is trained using the value as the output, and the model is optimized by adjusting the hyperparameters of the deep learning model. The optimized deep learning model is saved and called online. S5. Start welding under the welding parameters of S2, turn on the CCD camera to collect the front image of the molten pool in real time, crop the front image of the molten pool into the region of interest (ROI), and input the ROI image into the deep learning model to obtain the current d p Value, when welding t After the moment, the laser-arc hybrid welding control equipment is turned on, where t Set to 0.5-1s, the closed-loop controller predicts the depth of penetration. d p and penetration setting value ds Deviation value e(t) The positive and negative and size of the weld penetration deviation, and the trend of the change of the weld penetration deviation, Δ e ( t )= e ( t )- e ( t -1) Perform feedback control and output adjustment signals to the laser to change the laser power P ,in e ( t -1) is the weld penetration deviation at the previous control moment. After welding is completed, the laser-arc composite heat source welding equipment is turned off to achieve online control of the weld penetration of the overlapped structure.

2. The method for controlling the penetration depth of a composite lap weld according to claim 1, wherein: The angle in S1 θ The value range of 1 is 60°-80°, d The value range of 1 is 10-16mm, θ The value range of 2 is 70°-85°, θ The value range of 3 is 45°-70°, d The value range of 2 is 2-4mm, d The value range of 3 is 130-180mm.

3. The method for controlling the penetration depth of a composite lap weld according to claim 1, wherein: The weld penetration depth corresponding to the weld position of the molten pool front image in S3 d The values ​​are calculated using linear interpolation: d = d A + x × ( d B – d A ) / 2, where d A and d B For two adjacent points P 1 and P 2 The weld penetration corresponding to each position, x The current calculated frame position and P 1 The distance between the point locations, where x The calculation method is: x = k × 1 / f × V 1 ,in k For P 1 The number of frames from the position to the current calculation frame position, the data set is randomly divided into a training set and a test set according to the ratio of β: (1-β).

4. A hybrid welding lap weld penetration control method according to claim 3, characterized in that: The value range of β is 0.7-0.

85.

5. The method for controlling the penetration depth of a composite lap weld according to claim 1, wherein: The S4 deep learning model contains 1 input layer, n convolutional layers, n Pooling layers, m A convolutional neural network with 1 fully connected layer and 1 linear regression layer.

6. A method for controlling the penetration depth of a composite lap weld according to claim 1 or 4, characterized in that: The deep learning model in S4 is tuned by using the root mean square loss function to judge the deviation between the true value and the predicted value, using the gradient momentum descent optimizer to update the model weight coefficient, and using the test set to test the accuracy of the trained deep learning model; the adjustable hyperparameters include learning rate, number of hidden layers, activation function, and number of neurons.

7. A method for controlling the penetration depth of a composite lap weld according to any one of claims 1 to 5, characterized in that: The parameters in S3 f The value range is 20Hz-50Hz.

8. The method for controlling the penetration depth of a composite lap weld according to claim 5, wherein: The parameters in S4 n The value range is 3-7. n is an integer, parameter m The value range is 1-4. m is an integer.

9. The method for controlling the penetration depth of a composite lap weld according to claim 1, wherein: The CCD camera in the S5 collects the front image of the molten pool, and the resolution of the ROI after cropping is 750 pixels × 300 pixels.

10. A composite welding lap weld penetration control device, characterized in that: A method for controlling the weld penetration depth of a composite welding overlap joint according to any one of claims 1 to 9, comprising: a substrate (1), an arc power supply (2), an arc welding gun (3), a laser (4), a control terminal (5), a CCD camera (6), a control box (7), and a laser welding torch (8); wherein the negative electrode of the arc power supply (2) is connected to the substrate (1), the positive electrode is connected to the arc welding gun (3), the CCD camera (6) is connected to the control terminal (5) via a USB data interface, the CCD camera (6), the arc welding gun (3) and the laser welding torch (8) are connected via a fixture, and the laser (4) and the control box (7) are connected to the control terminal (5) via a data acquisition card; the CCD camera (6) collects a front image of the molten pool in real time, the control terminal (5) inputs the front image of the molten pool into a deep learning model through ROI clipping to predict the weld penetration depth, and the closed-loop control system in the control terminal (5) performs feedback control according to the error between the weld penetration prediction value and the weld penetration set value, sends a signal to change the output laser power of the laser (4), and controls the weld penetration depth online.

Images