Adaptive welding

Abstract

This is a system and method for adaptive welding. In some embodiments, the system includes a welding head, a first welding monitoring camera, and a machine learning system. The machine learning system may be configured to estimate the position of the distal end of the filler wire relative to the groove while the welding head is forming a weld layer in the groove.

Description

Cross - Reference to Related Applications

[0001] This application claims the benefit of, and priority to, U.S. Provisional Application No. 63 / 380,686, entitled "Adaptive Welding," filed on October 24, 2022. The entire content of the application is incorporated herein by reference.

Technical Field

[0002] One or more aspects of the embodiments according to the present disclosure relate to welding, and more specifically, to systems and methods for controlling a welding system.

Background Art

[0003] Welding systems may be used for automatic or semi - automatic welding. In such a welding process, various parameters may be adjusted, for example, the position of a heat source or a filler wire relative to the weld groove, the travel speed, or the supply speed of the filler wire.

[0004] Aspects to which the aspects of the present disclosure relate are related to this general technical environment.

Summary of the Invention

[0005] According to one embodiment of the present disclosure, a welding system is provided that includes a welding head, a first welding monitoring camera, and a machine learning system. The machine learning system is configured to estimate the position of the distal end of a filler wire relative to a weld groove while the welding head is forming a weld layer within the weld groove.

[0006] In some embodiments, the estimation includes estimating the position of a first point of interest based on an image acquired by the first welding monitoring camera, and the first point of interest is the intersection of a first edge of the weld groove and a reference plane.

[0007] In some embodiments, the estimation includes estimating the positions of four points of interest including the first point of interest, and each of the four points of interest is the intersection of an edge of the weld groove and a reference plane.

[0008] In some embodiments, estimating the location of a first point of interest involves estimating the intersection of a line corresponding to a first edge and a line corresponding to a reference plane within the image.

[0009] In some embodiments, the machine learning system is further configured to estimate the depth of the bevel.

[0010] In some embodiments, the machine learning system is further configured to estimate the width of the bottom of the bevel.

[0011] In some embodiments, the estimation further includes estimating the location of a fifth point of interest corresponding to the distal end of the filler wire.

[0012] In some embodiments, the machine learning system is further configured to estimate the error in the position of the distal end of the filler wire relative to the target position of the distal end of the filler wire.

[0013] In some embodiments, the estimation further includes estimating the location of a sixth point of interest corresponding to the tip of the tungsten electrode of the welding head.

[0014] In some embodiments, the machine learning system is further configured to estimate the error in the position of the tip of the tungsten electrode relative to the target position of the tip of the tungsten electrode.

[0015] In some embodiments, the machine learning system includes a first neural network for estimating the position of a tungsten electrode in images acquired by a first welding monitoring camera.

[0016] In some embodiments, the machine learning system further includes a second neural network for estimating the position of the distal end of the filler wire.

[0017] In some embodiments, estimating the position of the distal end of the filler wire involves estimating the position of the distal end of the filler wire based on the estimated position of the tungsten electrode and an image.

[0018] In some embodiments, the machine learning system further includes a third neural network for estimating the position of the upper end of the groove.

[0019] In some embodiments, the machine learning system includes a fourth neural network for estimating the position of the lower end of the groove.

[0020] In some embodiments, estimating the position of the lower end of the groove includes estimating the position of the lower end of the groove based on the estimated position of the upper end of the groove and an image.

[0021] According to one embodiment of the present disclosure, a method is provided which includes calculating the speed at which a substrate moves relative to a welding head based on the feed rate of the filler wire, the dimensions of the filler wire, and the dimensions of the weld layer to be formed.

[0022] In some embodiments, the dimensions of the filler wire are the diameter of the filler wire.

[0023] In some embodiments, the dimensions of the weld layer are the thickness of the weld layer, and the calculation is further based on the width of the weld layer.

[0024] In some embodiments, the calculation is further based on an efficiency coefficient.

[0025] These features and advantages of the present disclosure, as well as other features and advantages, will be understood by referring to the specification, claims, and accompanying drawings. [Brief explanation of the drawing]

[0026] [Figure 1A] This is a cross-sectional view of a substrate according to one embodiment of the present disclosure. [Figure 1B] This is a cross-sectional view of a substrate and a weld layer according to one embodiment of the present disclosure. [Figure 1C] Side view of a gas tungsten arc welding system according to an embodiment of the present disclosure. [Figure 1D] Side view of a gas metal arc welding system according to an embodiment of the present disclosure. [Figure 2A] Schematic diagram of an image captured by a welding monitoring camera according to an embodiment of the present disclosure. [Figure 2B] Schematic diagram of an image captured by a welding monitoring camera according to an embodiment of the present disclosure. [Figure 2C] An image captured by a welding monitoring camera according to an embodiment of the present disclosure. [Figure 3A] Block diagram of a neural network according to an embodiment of the present disclosure. [Figure 3B] Block diagram of a neural network according to an embodiment of the present disclosure. [Figure 4] Block diagram of a welding system according to an embodiment of the present disclosure.

Embodiments for Carrying Out the Invention

[0027] The following description, detailed together with the accompanying drawings, is an example of embodiments of a system and method for controlling a welding system provided in accordance with the present disclosure, and is not intended to represent the only form in which the present disclosure may be constructed or utilized. This description explains the features of the present disclosure in relation to the illustrated embodiments. However, it should be noted that the same or equivalent functions and structures can be achieved by other embodiments, and these are also intended to be included within the scope of the present disclosure. Here, as shown elsewhere, the same element numbers are intended to indicate similar elements and features.

[0028] In welding systems, two metal components (e.g., metal plates or pipes) can be joined by filling a groove (which may have a triangular, rectangular, or trapezoidal cross-section) along their contact line with molten metal in one or more welding passes. These metal components (and the unfinished weld) are sometimes collectively referred to as the "base material." The filler metal is supplied by a filler wire, and heat may be supplied by an arc. In gas tungsten arc welding (GTAW) (or tungsten inert gas (TIG) welding), the arc is formed between the tungsten electrode and the molten pool. In contrast, in gas metal arc welding (GMAW) (or metal inert gas (MIG) welding), the arc is formed between the filler wire and the molten pool. In other processes, heat may be supplied by other heat sources, such as lasers.

[0029] For example, referring to Figure 1A, when performing mechanical (i.e., automated) pipe welding, two pipes 100 may be positioned end to end, and their ends may be machined to form a groove 104 along the outer circumference of the pipe ends. As shown in Figure 1B, this groove 104 is formed by partially or completely welding using one or more welding passes, with each pass adding a weld layer 106 within the groove 104. These pipes 100 are sometimes collectively referred to as the base material 102.

[0030] Referring to Figures 1C and 1D, the welding system may include a welding head comprising a heat source and a filler wire supply system. The welding control system may include motors and other actuators to control the relative positions between the elements of the welding head and the base material, as well as the wire supply, as welding progresses. For example, the welding control system may include a movement control system to control the movement speed (e.g., controlling the movement of the welding head by moving the welding head along the length of the groove 104, or by moving the base material). Similarly, the position of the tungsten electrode 105 (e.g., the height and lateral position of the tungsten electrode 105), the distal end position of the filler wire 110 (e.g., the point where the filler wire 110 enters the molten pool), and the filler wire supply speed may also be controlled. The welding control system may also include a welding power source that controls the voltage applied to the arc or the current flowing through the arc.

[0031] Because the groove 104 may not be straight, its width may vary along its length, and it may not perfectly align with the operating axis of the movement control system, simply driving the movement control motor without adjusting the elements of the welding head vertically or horizontally may result in misalignment between the groove 104 and the weld, variations in the thickness of the metal layer welded in each pass, and welding defects as the welding progresses. Therefore, a sensing system can be used to measure various geometric parameters of the process, and these measurements can be used to control the actuators and welding power supply.

[0032] The sensing system may use a welding monitoring camera 115. This camera may also be a video camera configured to acquire images of the molten pool, tungsten electrode 105, filler wire 110, and a portion of the substrate as the welding progresses. In some embodiments, the system includes multiple welding monitoring cameras 115 (e.g., two welding monitoring cameras as shown in Figures 1C and 1D). The images may be analyzed by a machine learning system (e.g., a system including one or more neural networks, as will be described in more detail below). Figures 2A and 2B show examples of such images (for GTAW and GMAW, respectively), with points of interest in the images that can be identified by the machine learning system labeled. Figure 2C is a frame from a video taken by a welding monitoring camera 115 in a GTAW system, with points of interest 205, 210, and 215 superimposed on the image.

[0033] The sensing system (for example, using a machine learning system) can estimate (I) the position of the tip on the tungsten electrode 105 (e.g., height and lateral position) (indicated by reference numeral 205 in Figure 2A) (in the GTAW system), (II) the position of the distal end of the filler wire 110 (e.g., height and lateral position) (indicated by point 210 in Figure 2A and point 260 in Figure 2B), and (III) the lateral position of each of the four intersections between the four edges of the groove 104 and the reference plane (indicated by reference numeral 215 in Figure 2A and reference numeral 265 in Figure 2B). The four edges of the groove 104 are two upper edges where the wall of the groove 104 intersects the upper surface of the base material, and two lower edges where the wall of the groove 104 intersects the lower surface of the groove 104 (the lower surface formed when the groove 104 is machined, or the upper surface of the pass before welding, e.g., a previously welded layer 106). The reference plane is defined by the horizontal line in the image, so that all four intersection points lie on the horizontal line in the image.

[0034] From the estimated parameters, the following can be calculated: (i) the position error of the tungsten electrode 105 relative to the target position (in a GTAW system), (ii) the position error of the distal end of the filler wire 110 relative to the target position, (iii) the depth of the groove (104), and (iv) the width of the bottom of the groove (104). The target position of the tip of the tungsten electrode 105 in a GTAW and the target position of the distal end of the filler wire 110 in a GMAW may be at a constant height above the center of the groove 104, or they may move from side to side as welding progresses, for example, when using a weaving bead. In some embodiments, in a GMAW system, the height of the distal end of the filler wire 110 may be estimated based on the arc voltage instead of, or in addition to, the image from the welding monitoring camera 115. In GTAW applications, the target position of the distal end of the filler wire 110 may be offset by a predetermined amount vertically from the tip of the tungsten electrode 105 and aligned horizontally with the tip of the tungsten electrode 105. The depth and bottom width of the groove 104 can be calculated from: (i) the distance between the intersections corresponding to the two upper edges of the groove 104, (ii) the distance between the intersections corresponding to the two lower edges of the groove 104, and (iii) known information regarding the inclination angle of the side walls of the groove 104.

[0035] Commands can be sent to the welding control system to minimize positional errors of the tungsten electrode 105 and the distal end of the filler wire 110, or to keep these errors small enough to be acceptable for the weld quality. The estimated depth of the groove 104 can be used to select the thickness of the metal layer to be welded during the current pass, so that after the current pass is completed, the remaining depth of the groove 104 is equal to (or approximately equal to) the target depth. For example, in the case of flash welding (e.g., welding with a bead without a crown), the target thickness of the metal layer to be welded during the current pass can be calculated by dividing the estimated depth of the groove 104 by the number of remaining passes, including the current pass.

[0036] The estimated width of the groove 104 can be used to calculate the metal deposition rate per unit length of weld, corresponding to the target thickness of the metal layer to be welded in the current pass. This deposition rate can then be achieved by adjusting the wire feed rate and travel rate (details below), and the heating power can be adjusted according to the wire feed rate and travel rate by adjusting the voltage or current (or both) of the welding power supply. When using a weaving motion pattern, the width can also be used to calculate the vibration stroke, speed, and residence time. When using a split bead weld configuration (multiple beads per layer), the width can also be used to calculate the position of each weld bead relative to other weld beads or the side of the weld groove 104.

[0037] Referring to Figure 3A, multiple interconnected neural networks may be used to make the following estimations: the position of the tungsten electrode 105, the position of the distal end of the filler wire 110, the position of the intersection corresponding to the upper edge of the groove 104, and the position of the intersection corresponding to the lower edge of the groove 104. The first neural network 305 receives images from the welding monitoring camera and estimates the coordinates of a point that determines the tip position of the tungsten electrode 105. The estimated tip position of the tungsten electrode 105 is used by the welding control system as described above and is also input to the second neural network 310. This second neural network 310 may use both the estimated tip position of the tungsten electrode 105 and the image to estimate the distal end position of the filler wire 110. The third neural network 315 receives the image and estimates the position of the upper edge of the groove 104 (for example, it may estimate the horizontal coordinates of the intersection corresponding to the upper edge of the groove 104). The fourth neural network 320 may receive (i) estimated horizontal coordinates of the intersection corresponding to the upper edge of the groove 104 and (ii) an image, and estimate the position of the lower edge of the groove 104 (for example, it may estimate the horizontal coordinates of the intersection corresponding to the lower edge of the groove 104). Figure 3B shows the internal structure of the first neural network 305 and the second neural network 310 in one embodiment. These neural networks may be implemented in one of the following ways: processing circuits (described later), analog circuits, or analog-digital hybrid circuits.

[0038] Each neural network can be trained using supervised learning. This training generates a set of labeled images by labeling the coordinates of points of interest in images acquired from a welding monitoring camera during welding. Labeling can be done manually by an experienced operator. The size of the training dataset can be increased by adding modified labeled images. For example, additional labeled images can be created by moving a previously labeled image horizontally by a certain amount and adjusting the horizontal coordinates of the labels accordingly. The result of the training, when loaded into the neural network, is a set of filters, weights, and biases (or neural network model) that the neural network uses to estimate the coordinates of points of interest in images during welding. The set of weights, filters, and biases is sometimes called the "model".

[0039] A similar approach can be used to train a neural network to recognize (and determine the coordinates of) points of interest in images of the welding setup before welding begins, taken from a welding monitoring camera. Such images may differ from images of welding in progress in that the lighting is different (provided by a suitable light source rather than primarily by the arc) and there is no molten pool (and the point where the distal end of the filler wire 110 enters the molten pool). In some embodiments, the neural networks in Figures 3A and 3B are trained twice, once with images of welding in progress and once with images of the welding setup before welding begins, generating a first model and a second model, respectively. The second model is then used (i.e., loaded into the neural network) to align the tungsten electrode 105 and filler wire 110 to the groove 104 before welding begins. Once this adjustment is complete, the first model is loaded into the neural network, the arc is started, and welding proceeds using the first set of weights.

[0040] In a neural network-controlled welding system, or in a welding system without such control, the travel speed can be calculated using an equation derived as follows, based on the diameter of the filler wire, the cross-sectional area of ​​the weld layer 106, and the filler wire supply rate:

[0041] As long as the volume of filler metal is preserved, the rate at which the volume of filler metal is added to the molten pool via the filler wire 110 is equal to the rate at which the filler metal is removed from the molten pool in the form of a completed weld layer 120. The rate at which the volume of filler metal is added can be calculated as the product of the cross-sectional area Af of the metal in the filler wire 110 and the feed rate vf of the filler wire using the following formula.

[0042]

number

[0043] In the case of a round wire (i.e., the wire being fed out has a circular cross-section), the cross-sectional area Af is expressed by the following formula.

[0044]

number

[0045] Here, φ is the diameter of the filler wire 110. The rate at which the metal is removed from the weld is expressed by the following formula.

[0046]

number

[0047] Here, AL is the step area of ​​the weld layer, and vt is the travel speed. For a rectangular weld groove, the step area is expressed by the following formula.

[0048]

number

[0049] Here, W is the width of the groove (and the layer), and T is the thickness or "height" of the layer.

[0050] If we assume that the rate at which metal is added is equal to the rate at which metal is removed, then we obtain the following equation.

[0051]

number

[0052] Therefore, given the weld width, layer height, cross-sectional area of ​​the filler rod, and filler wire supply rate, the travel speed that yields the desired layer height can be determined from equation (1) (the slower the travel speed, the higher the layer height; the faster the travel speed, the lower the layer height).

[0053] If the filler wire 110 is hollow rather than solid (for example, a hollow flux-cored wire), the metal cross-sectional area of ​​the filler wire 110 is, for example, 25% of the total cross-sectional area of ​​the filler wire 110 (the remaining 75% of the cross-sectional area is flux), that is, the cross-sectional area Af is expressed by the following formula.

[0054]

number

[0055]

number

[0056] (Here, for example, in the case of filler wire 110 where 75% by volume is flux and 25% is metal, m = 0.25).

[0057] Equation (1) may not hold perfectly if the volume of the filler metal is not retained (for example, if some of the filler metal is lost due to sputtering or oxidation, or as a result of other mechanisms, such as changes in crystal structure that affect density). Equation (1) may also not hold perfectly if the weld layer is so high that some of the filler metal overflows from the groove 104. However, if the equation holds approximately, satisfactory results can be obtained using it, or corrections can be made to the equation if the extent to which the process deviates from the ideal is known. For example, equation (2) can be used, along with a correction factor m (or alternatively), to compensate for various mechanisms that prevent equation (1) from holding perfectly. For example, if 5% of the metal in the filler wire 110 is lost due to sputtering, a value of m = 0.95 can be used (or, if m is not 1 for other reasons, the value of m can be reduced by 5% to account for sputtering losses).

[0058] If the end of the pipe is shaped such that the groove is V-shaped, the weaving may be increased in each pass around the pipe so that the width of the weld is substantially equal to the width of the partially filled V-shaped bottom. In this case, equation (1) can be used, where W is the width of the partially filled V-shaped bottom (or, more precisely, W is the width of V at a height of half the thickness of the layer from the partially filled V-shaped bottom). In some embodiments, the substrate is any substrate (e.g., a plate to be joined, or a substrate on which a metal overlay layer is formed), and the travel speed can be calculated using equation (1) or equation (2) based on the diameter of the filler wire and the feed rate of the filler wire, as well as the height and width of the weld layer 120 being formed (or, more generally, use the following equation if the filler wire 110 is neither circular nor solid, or if the cross-section of the weld layer 120 is not rectangular).

[0059]

number

[0060] During operation, the welding system can calculate the target travel speed to be used and automatically set the travel actuators (rotating the pipe or moving the welding head outside the pipe at the joint) to move at the target travel speed. The system can obtain the input variables of equation (1) from various sources, such as operator input (supplied to the welding system via an input device such as a keyboard) and sensors. For example, the operator can adjust the wire feed rate by instructing the welding system to increase or decrease the feed rate of the filler wire until the molten pool has the desired characteristics. Thus, the welding system can know the feed rate of the filler wire from the operator's latest instruction. The width of the weld can be measured by the welding system using machine learning-based image analysis disclosed herein or by other methods.

[0061] Figure 4 shows a welding system that can implement movement speed control based on the above equation. In some embodiments, such a welding system includes a welding head 410 which includes a filler wire supply unit 412 that supplies filler wire 110 to a molten pool 205. A system for heating the molten pool is not explicitly shown. It may be, for example, an arc or a laser (by current flowing through the filler wire 110 or through a separate (e.g., tungsten) electrode). The system may further include a moving actuator 415 that controls the relative motion between the welding head 410 and the substrate 110. The welding system may also include a control circuit 425 which includes (i) a moving actuator drive circuit 430 that interfaces with the moving actuator 415, (ii) a wire feed drive circuit 435 that controls the feed rate of the filler wire 110, and (iii) a processing circuit 445 (described in more detail below) that performs high-level control functions such as commands to the moving actuator 415. The system may also include a welding power supply and controller, and other elements (not shown) such as one or more user interface devices such as a display, keyboard, or mouse.

[0062] In this specification, “part” of something means “at least a part” of that thing, and therefore may mean less than or the whole of that thing. Thus, “part” of something includes, in special cases, the whole thing; that is, the whole thing is an example of a part of it. In this specification, the word “or” is inclusive, for example, “A or B” means (i) A, (ii) B, and (iii) A and B.

[0063] In this specification, the term “processing circuit” is used to mean any combination of hardware, firmware, and software used to process data or digital signals. Processing circuit hardware includes, for example, application-specific integrated circuits (ASICs), general-purpose or dedicated central processing units (CPUs), digital signal processors (DSPs), graphics processing units (GPUs), and programmable logic devices such as field-programmable gate arrays (FPGAs). In the processing circuits used herein, each function is performed by hardware configured to perform that function, i.e., hardwired hardware, or by more general-purpose hardware such as a CPU configured to execute instructions stored in a non-temporary storage medium. A processing circuit may be manufactured on a single printed circuit board (PCB) or distributed across multiple interconnected PCBs. A processing circuit may include other processing circuits, for example, two processing circuits interconnected on a PCB, i.e., an FPGA and a CPU.

[0064] While exemplary embodiments of systems and methods for controlling welding systems have been specifically described and illustrated herein, many modifications and variations will be apparent to those skilled in the art. Therefore, it should be understood that systems and methods for controlling welding systems configured according to the principles of this disclosure can be implemented in forms other than those specifically described herein. The present invention is also defined by the following claims and equivalents.

Claims

1. A system for welding, Welding head and The first welding monitoring camera, Equipped with a machine learning system, The machine learning system is configured to estimate the position of the distal end of the filler wire relative to the groove while the welding head is forming a weld layer in the groove.

2. The system according to claim 1, wherein the estimation includes estimating the position of a first point of interest based on an image acquired by the first welding monitoring camera, the first point of interest being the intersection of a first edge of the groove and a reference plane.

3. The system according to claim 2, wherein the estimation comprises estimating the locations of four points of interest, including the first point of interest, each of the four points of interest being an intersection of the edge of the groove and the reference plane.

4. The system according to claim 3, wherein the estimation of the position of the first point of interest includes estimating the intersection of a line corresponding to the first edge and a line corresponding to the reference plane in the image.

5. The system according to claim 4, wherein the machine learning system is further configured to estimate the depth of the groove.

6. A system according to claim 4 or 5, wherein the machine learning system is further configured to estimate the width of the bottom of the groove.

7. A system according to any one of claims 4 to 6, wherein the estimation further comprises estimating the location of a fifth point of interest, the fifth point of interest corresponding to the distal end of a filler wire.

8. A system according to any one of claims 4 to 7, wherein the machine learning system is further configured to estimate the error in the position of the distal end of the filler wire with respect to a target position of the distal end of the filler wire.

9. A system according to any one of claims 4 to 8, wherein the estimation further comprises estimating the location of a sixth point of interest, the sixth point of interest corresponding to the tip of the tungsten electrode of the welding head.

10. The system according to claim 9, wherein the machine learning system is further configured to estimate the error in the position of the tip of the tungsten electrode relative to a target position of the tip of the tungsten electrode.

11. A system according to any one of claims 1 to 10, wherein the machine learning system comprises a first neural network that estimates the position of a tungsten electrode in an image acquired by the first welding monitoring camera.

12. A system according to any one of claims 1 to 11, wherein the machine learning system further comprises a second neural network for estimating the position of the distal end of the filler wire.

13. A system according to any one of claims 1 to 12, wherein estimating the position of the distal end of the filler wire includes estimating the position of the distal end of the filler wire based on the estimated position of the tungsten electrode and the image.

14. A system according to any one of claims 1 to 13, wherein the machine learning system further comprises a third neural network for estimating the position of the upper edge of the groove.

15. A system according to any one of claims 1 to 14, wherein the machine learning system comprises a fourth neural network for estimating the position of the lower edge of the groove.

16. A system according to claim 15, wherein estimating the position of the lower edge of the groove includes estimating the position of the lower edge of the groove based on the estimated position of the upper edge of the groove and the image.

17. The movement speed of the substrate relative to the welding head, The feed rate of the filler wire and The dimensions of the aforementioned filler wire and The dimensions of the welded layer that is formed and A method that includes calculation based on [a specific factor / method].

18. A method according to claim 17, wherein the dimensions of the filler wire are the diameter of the filler wire.

19. The method according to claim 17 or 18, The dimensions of the welded layer are the thickness of the welded layer. The calculation is further based on the width of the weld layer.

20. A method according to any one of claims 17 to 19, wherein the calculation is further based on an efficiency coefficient.

Images