Autonomous robotic cutting system

Abstract

Autonomous oxygen gas cutting of a metal substrate or surface along a cutting path employs visual feedback based on camera images from a predetermined, marked cutting path. A method of metal cutting according to a predetermined path includes identifying a path on a substrate for cutting and calculating a set of points based on iterative intervals along the path. The controller arranges the cutting torch based on a tangent to the path at each of the set of points. The controller iteratively advances the welding gun based on consecutive points in the set of points to complete the traversal of the path. The cutting torch control includes moving an oxygen gas cutting jet along a cutting path on the metal surface to achieve efficient and complete cutting. When traversing the cutting path, the controller adjusts the surface hot pool mass by moving the torch tip at an appropriate speed.

Description

Background Art

[0001] Vision-based sensing technology has been widely used in robotic welding and cutting applications due to its non-invasive, high-precision, relatively simple, and low-cost nature. Recent advances in vision-based robotic welding include vision calibration, seam tracking, weld pool monitoring, and associated deformation detection. The "weld pool" refers to the area on the metal surface where heat is concentrated, also known as the melt pool, from which information relevant to the welding task can be extracted. Summary of the Invention

[0002] Autonomous oxyfuel cutting of a metal substrate or surface along a cutting path utilizes visual feedback from a camera-based image of a predetermined, marked cutting path. The method for cutting metal according to the predetermined path includes identifying a path on the substrate for cutting and calculating a set of points based on iteration intervals along the path. A controller positions a cutting torch based on the tangent to the path at each point in the set of points. The controller iteratively advances the torch based on successive points in the set of points to complete traversal of the path. Cutting torch control involves moving an oxyfuel cutting jet or flame along the cutting path on the metal surface to achieve an efficient and complete cut. While traversing the cutting path, the controller adjusts the surface heat pool quality by moving the torch tip at an appropriate speed. Because the desired position of the torch tip (i.e., a point on the cutting path) is known, the controller is constrained to moving the torch tip only along the tangent vector of the cutting path curve. In this manner, the cutting path is presented as a list of poses to be visited. The controller then only needs to determine the rate at which these poses are traversed, expressed as a velocity along the tangent vector.

[0003] This configuration is based in part on the idea that robotic task automation is increasingly being used to free human participants from hazardous tasks. Metal cutting and welding are typical tasks involving extreme heat, radiation, and potentially hazardous fumes that would benefit from such automation. Unfortunately, traditional approaches to automating these tasks suffer from being highly flexible and skill-based, making precision difficult to achieve in programming logic. For example, metal recycling, including scrap cutting, is challenging to automate because the properties of scrap metal fragments vary widely, and it is difficult to predict the cutting behavior that will occur. Automated welding, typically used for new products, relies on precise specifications of the materials being welded. In contrast, disassembly at the other end of the product lifecycle occurs under uncertain materials and conditions. Oxy-fuel (oxygen-driven cutting torches) applied to scrap metal encounter metal of varying thicknesses and shapes. Controlling the cutting speed of the oxy-fuel torch is crucial to ensuring accurate and complete cuts. Therefore, this configuration substantially overcomes the shortcomings of robotic cutting methods by employing camera-based control and presenting an image of the cut heat pool. The system then adjusts the cutting speed based on the convexity and intensity shown in the image, and follows a cutting path based on a predetermined pattern scanned from the cut metal surface.

[0004] More specifically, a method for actuating a robotic cutting torch disclosed herein includes guiding the cutting torch along a predetermined cutting path along a metal surface and receiving an image of a heat pool defined by engagement of the cutting torch with the metal surface. The robotic actuator controls the speed of the cutting torch based on the convexity and intensity of the heat pool calculated from the image. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] The foregoing and other objects, features, and advantages of the present invention will become apparent from the following description of specific embodiments of the present invention, as shown in the accompanying drawings, in which like reference numerals refer to like parts throughout the different views. The drawings are not necessarily drawn to scale, emphasis instead being placed upon illustrating the principles of the invention.

[0006] Figure 1 This is the background image of the robot cutting environment configured in this article;

[0007] Figure 2 Shown by Figure 1 The process is undertaken by robots;

[0008] Figure 3 Shows the use of Figure 1 The device traverses a cutting path along the metal surface to cut;

[0009] Figures 4A-4E Shows the use of Figure 1-Figure 3 The intensity group calculated by the method of ; and

[0010] Figures 5A-5C Shown Figure 1 - Image of the burning area of ​​the welding torch of Figure 4, Figure 1 - Figure 4 shows convexity and strength. DETAILED DESCRIPTION

[0011] The use of robots in industrial applications has significantly contributed to productivity growth across the economy. The widespread adoption of industrial robots has been shown to improve worker safety and reduce the global ecological footprint. With demand for primary metals expected to increase, the industrial operations associated with them are becoming increasingly important. In particular, the automation of metal cutting can greatly benefit cutting-intensive industries such as manufacturing and scrap metal recycling, which includes shipbreaking (breaking large metal structures into smaller pieces) and lead to more efficient steel production compared to iron ore processing. Metal cutting technologies and their associated robots are diverse, with distinct advantages and limitations. The configurations discussed below utilize combustion-based thermal cutting media for automated oxy-fuel cutting operations.

[0012] Figure 1 This is the background image of the robot cutting environment used in this article. Figure 1 , a mobile actuator 120 includes a drive 122 (such as a set of rails) and a robotic arm 124 including one or more robotic members 126-1...126-2 (collectively referred to as 126), which is used to approach an object 102 that defines a recyclable item and a proposed cutting path 110. An end effector 130 is connected to the end of the robotic arm 124 and includes an optical sensor or camera 132 and a cutting tool or torch 134. The camera 132 receives an image 133 of a heat pool 150 defined by the engagement of a flame 135 or combustion source of a cutting torch 134 with a metal surface 101. The torch 134 is adapted to follow a predetermined cutting path 110 along the surface 101 of the scrap object 102. The flame 135 is adapted to sever the material constituting the recyclable item by reaching a combustion temperature sufficient to melt and cut the metal surface 101.

[0013] The robotic controller 140 in the motion actuator 120 includes an image processor 144 for receiving the image 133 of the heat pool 150. The cutting logic 142 includes instructions for analyzing the image and directing the end effector 130 or the robotic gripper holding the welding gun 134 using a processor 146 including a power source and memory.

[0014] In operation, a motion actuator 120 traverses a predetermined cutting path 110 along a metal surface 101 for guiding a cutting torch 134. The cutting path 110 has been pre-scanned based on a colored or visual line on the metal surface 101, as described in co-pending U.S. patent application Ser. No. 18 / 119,547, filed on March 9, 2023, entitled “FEATUREDRIVEN NEXT VIEW PLANNING OF 3-DIMENSIONAL SURFACES,” which is incorporated herein by reference in its entirety. The cutting tool 134 is then moved along a curve defined by a set of points defining the cutting path 110, with the cutting tool 134 responding to the actuator 130. The actuator 130 is driven by cutting logic 142 for controlling the speed of the cutting torch 134 based on the convexity and intensity of the heat pool calculated from the image.

[0015] Figure 2 Shown by Figure 1 The process is undertaken by robots. Figure 2 The vision-based control method shown performs autonomous combustion cutting using an oxy-fuel torch 134. This method is inspired by the techniques of skilled cutting workers, particularly in shipbreaking yards. Conventional workers track the formation and evolution of the heat pool 150 at the junction 50 of the flame 135 with the metal surface 101 to adjust their torch movement to achieve a successful cut. The disclosed method implements a vision system that encodes the visual characteristics (shape, size, brightness and color) of the heat pool 150 by calculating two features: the convexity and intensity of the heat pool. These two features are combined to describe the combustion state of the heat pool, which is used to control the movement of the torch 134. An example configuration implements a method for Figure 1 Vision-based control method for a 1-DOF cutting robot is shown. Any suitable robot may be used to follow the predetermined cutting path 110 while grasping the welding gun 132 .

[0016] refer to Figure 1 and Figure 2The cutting operation 200 begins with calibration at step 202 , which detects the torch flame 135 and records its center of mass by averaging the coordinates of the center of mass of the maximum heat region of the stable torch flame over a set number of frames. The intensity is recorded to serve as a baseline for subsequent measurements. As shown in step 204 , the metal surface 101 is initially heated by maintaining the torch flame 135 at an initial position along the cutting path 110 , thereby establishing visual feedback. As the heat pool 150 forms and evolves, the pool is monitored until it is sufficiently burned. A filtered image 210 depicts the intensity based on the center of mass of the flame 135 . A color map 212 shows colored or shaded regions based on the temperature range surrounding the center of mass. Regions 151 - 1 through 151 - 3 (collectively, 151 ) depict the temperature range of the heat pool 150 , with the highest temperature region 151 - 1 represented in the blue channel (also including the center of mass), surrounded by the lower temperature green region 151 - 2 and the outermost red region 151 - 3. The welding gun control based on the cutting logic 142 includes advancing the welding gun 134 through the cutting path 110 by adjusting the speed to maintain combustion of the desired heat pool 150 indicated by the image 133, as shown in step 206. As shown in step 208, the welding gun 134 continues to traverse the surface 101 at a forward speed based on the cutting logic 142 operating in an iterative feedback loop.

[0017] A particular advantage of the disclosed method is its adaptability to different thicknesses of the metal surface 101. The actuator 130 performs a cut by advancing the welding gun 134 on the metal surface 101 of a first thickness, and then advances the welding gun 134 on the metal surface of a second thickness different from the first thickness by adjusting only the cutting speed to maintain the strength between an upper strength limit and a lower strength limit and to maintain the crown above a crown minimum value.

[0018] Figure 3 Shows the use of Figure 1 A device that traverses a cutting path along a metal surface. Figure 1-3 , the configuration herein demonstrates the separation of the welding gun control problem from the motion planning and control tasks of the actuator 134. The robot 120 achieves this by formulating and solving the welding gun control problem in a specific task space. In practice, the cutting logic 142 expects the cutting path 110 to be predetermined by an appropriate cutting path generation method, which is discussed in the co-pending patent application referenced above. The forward motion maintains the welding gun 134 perpendicular to the metal surface 101 while traversing the cutting path 110, as shown by the normal vector 150'. Assuming that the robot's 3D cutting path is given, the purpose of the vision-based cutting logic is to determine the cutting speed, that is, the tangential speed of the welding gun flame as the robot traverses the cutting path 110.

[0019] It is worth noting that the cutting speed determines the success of the cutting operation. Setting a fixed cutting speed does not guarantee a successful cut because the speed must be adjusted according to the burning conditions of the surface.

[0020] Therefore, the visual feedback provided to the controller is a heat pool burn state descriptor calculated from the camera's RGB feed. Specifically, the camera image is processed (simplified and filtered), and two heat pool features (convexity and intensity) are then used to calculate the forward velocity. One might intuitively think that a slow cut would completely penetrate each plate along the cut path. In reality, moving the torch too slowly can lead to excessive heat accumulation in the area surrounding the cut path, causing the cut portion to re-seal itself with creep deposits from the heated steel.

[0021] Conventional approaches do not employ a vision-based controller that regulates the firing state by updating the torch speed while maintaining a consistent height and tangential attitude above the cutting surface. Our configuration demonstrates that this approach can successfully cut steel plates of varying thicknesses relying solely on visual feedback, without requiring prior knowledge of the thickness.

[0022] The control problem involves moving the torch's flame jet along a reference spatial path across the metal surface so that the cut is both successful and efficient. The controller must regulate the burning state of the surface heat pool while traversing the cutting path by moving the torch tip at the appropriate speed.

[0023] This particular configuration assumes that the cutting path is predetermined by a suitable cutting path generation method and that the welding gun tip remains perpendicular to the metal surface while moving along the path. Gun control focuses on determining the cutting speed, i.e., the speed at which the welding gun moves along the path 110. Moving the welding gun at a sufficient cutting speed is crucial for a successful cut. It should be emphasized that the vision-based cutting logic 142 does not use any a priori information about plate thickness or temperature, but relies entirely on visual feedback.

[0024] Since the desired gun tip position is known from a point on the reference cutting path 110, the controller 140 is constrained to move the gun tip only along the tangent vector of the curve. Essentially, the reference path is a series of gun tip positions to be visited, where the controller determines the rate at which these positions are traversed, expressed as Figure 3 The velocity of the tangential vector 150'.

[0025] The cutting logic 142 uses visual feedback from the RGB camera to set the acceleration (update velocity) of the tangent vector 150' along the path. Expressing control actions in terms of local tangent vectors separates the combustion control problem from the robot motion planning problem. In a general setting, the tangential motion commands can be decomposed into joint space using the inverse kinematics of the system.

[0026] Since the goal of the cutting logic 142 is to maintain sufficient combustion for cutting, the visual feedback from the camera 132 should be designed to describe the combustion state of the heat pool. It should be noted that excessive combustion indicates that the welding gun speed is too slow, while insufficient combustion indicates that the speed is too fast, and sufficient speed results in the desired combustion state. The extracted visual features must then reflect this negative correlation between cutting speed and combustion state.

[0027] Therefore, the visual feedback provided to the controller is a thermal pool combustion state descriptor s calculated from the camera's RGB feed. Specifically, the camera image 133 is processed (simplified and filtered) and then two thermal pool features (convexity and intensity) are calculated and combined to describe the pool's combustion state s.

[0028] The image is simplified to highlight the features of the heat pool. The features of interest are visual surrogates for shape and temperature. In a heat pool, the color distribution of light emission can be used to estimate the relative temperature distribution. This is because higher frequency visible light is emitted in a higher energy state. The cutting logic 142 restricts the RGB color space to four discrete colors arranged in order of decreasing temperature: blue, green, red, and black, however any suitable number of layers or temperature thresholds can be used. In this discrete color space, blue light emission is associated with the highest temperature (higher electromagnetic frequency), green light is associated with medium temperature (medium frequency), and red light is associated with the lowest temperature (lower frequency). Black represents low light emission, so the temperature is negligible for metal cutting.

[0029] This phenomenon is Figures 4A-4E , depicting the original camera image and its color channels. Figure 1-Figure 3 and Figures 4A-4E , the highest heat intensity is closest to the flame center ( Figure 4A ) and decays with increasing distance. In fact, the blue channel ( Figure 4D ) has the highest brightness closest to the center of the pool, while the green channel ( Figure 4C ) and the red channel ( Figure 4B ) includes the moderate and lower temperature parts of the pool.

[0030] Thus, the original image 133 is processed as follows:

[0031] Binary thresholding was performed on each channel using a high cutoff value;

[0032] Color the non-zero pixels of each channel (red, green, blue);

[0033] Eliminate channel overlap through XOR operation;

[0034] Merge channels to generate segmented heat pools, e.g. Figure 4EAs shown, different shading represents the various temperature levels exhibited by the thermal pool 150.

[0035] This simple and efficient per-channel thresholding procedure highlights the pool's geometric and thermal information. Because the image contains only four colors, the pool's shape and size are preserved, and its color information is quantized. This enables efficient representation and tractable reasoning about the thermal pool 150, its convexity characteristics, and its combustion.

[0036] Each image of the thermal pool 150 thus represents a plurality of temperature regions 151, such that each temperature region represents a temperature threshold. Within the respective shaded area, each temperature region defines a group of pixels in the image representing a temperature above the threshold, such that all pixels within that temperature region represent a temperature above the threshold. Based on the threshold, the temperature region 151 represents a group of pixels within a specified temperature range between a minimum threshold value and a maximum threshold value, with the centermost pixel (including the centroid) having the highest temperature.

[0037] Convexity and intensity are calculated based on the image of the region. Using the processed image, the cutting logic 142 uses its convexity to define a metric for the shape of the pool. During the cutting process, when the shape of the pool is relatively more convex, the surface conditions are more favorable for combustion. Conversely, when the shape of the pool is more concave or exhibits significant convexity defects, the surface conditions are less favorable for combustion. This is because a higher relative convexity indicates that the heat is more concentrated in the hot pool. While there are sophisticated methods to quantify the convexity of a closed contour, the cutting logic 142 calculates the area ratio between the pool contour and its convex hull. Formally, let K be the contour of the hot pool in the image; this is a set of pixel coordinates of a simple closed curve. We define the pool convexity as c = |K| / |convK|, where |K| represents the area of ​​the contour and |convK| represents the area of ​​its convex hull. This yields the following desirable properties: (1) c∈(0, 1], where c=1 is a completely convex shape. Higher values ​​of c indicate higher convexity and therefore correlate with higher flare. The ratio c is invariant to the size or color of the pool; it simply quantifies the shape of the pool. More precisely, it is size-invariant, pose-invariant, and color-invariant.

[0038] In practice, the profile K is restricted to the hottest layer in the pool, i.e., the blue or innermost channel of the pool, to produce a more robust convexity value. Furthermore, K is calculated as the largest profile closest to the centroid of the welding gun flame (previously obtained from the calibration step).

[0039] Figures 5A-5C Shown Figure 1-4E Image of the burning area of ​​the welding torch, Figure 1-4E Convexity and intensity are shown. Typically, the heat pool image represents a temperature region 151, such that convexity is based on the percentage of the bounding convex hull surrounding the region that the temperature region occupies. Figure 5AA heat pool 150 is shown that is approximately elliptical. Using the innermost region, corresponding to the blue channel, a line is drawn around the outer edge, much like an elastic band being applied around the outer edge. Concave regions 154-1 and 154-2 (collectively 154) reduce the convexity so that the convexity is a percentage of the band-shaped area occupied by the blue channel region. That is, the area of ​​concave region 154 represents the percentage area reduction from 100% convexity, resulting in a value of approximately 78%. Figure 5B A greater convexity is defined, and a more rounded, less elliptical shape is consistent with a single small concave region 154 having a convexity of approximately 90%. Figure 5C It appears to be essentially circular, with a correspondingly high convexity of approximately 98%. In practice, limiting the contour K to the hottest layer in the pool (i.e., the blue channel of the pool) produces more robust convexity values. Furthermore, K is calculated by retrieving the largest contour closest to the centroid of the torch flame.

[0040] Once the convexity is calculated, the intensity of the thermal pool follows. The intensity is calculated by grouping the pixels in each region and weighting each group of pixels. The weighted pixels are then aggregated according to the radial falloff to calculate the intensity. The pool intensity is designed to convey the color and size of the pool from the processed image. The pool intensity calculation is summarized below and detailed below:

[0041] Pixels are weighted based on color (black, red, green, blue).

[0042] The pixels are weighted by a radial falloff function centered around the centroid of the welding torch flame (obtained by calibration).

[0043] All weighted pixels are summed to produce the unscaled intensity.

[0044] The sum was scaled by the calibration baseline, a saturation cutoff was applied, and normalized to produce pooled intensities.

[0045] The first step of the intensity calculation approximates the relative temperature differences in a quantized color space. This captures both magnitude (the total number of non-zero pixels) and color (via weights). The second step approximates the nonlinear radial decay effect of heat transfer, i.e., the heat at a pixel decays rapidly with distance from the centroid of the welding torch flame. Furthermore, this radial decay provides robustness to noise and undesirable effects such as: (1) light contamination, (2) sudden sparks, slag, or streaks, and (3) residual heated areas far from the heat pool. In this paper, a bivariate Gaussian function is suitable because it uses simple parameters to model the radial exponential decay.

[0046] Formally, let M be the processed image in array form, which contains pixel p, where pixel p is considered as a tuple (x p ,y p, color(p)). The 2D Gaussian function g(p) assigns an attenuation factor to each pixel based on the distance of each pixel from the centroid (x c , y c ) of the torch flame, which is determined during calibration. The color weighting function w(p) assigns a constant weight based on the color of the pixel. The functions g(p) and w(p) are:

[0047]

[0048] In this paper, σX and σY determine the radial attenuation rate of each axis. The weights wR < wG < wB are constants. Using these definitions, the absolute sum is defined as:

[0049] I = ∑p∈M g(p)w(p).

[0050] In the disclosed method, σX = σY = 30 px and (wR, wG, wB) = (0.01, 0.04, 0.16) to approximate the non-linear temperature difference between colors.

[0051] During calibration, we perform these intensity calculations and obtain a baseline intensity value I cal , and all other intensities are scaled according to this value. Let I ∽ = I / Ical be the intensity expressed relative to the baseline. This expresses the intensity in physical terms, for example, I ∽ = 1 is the intensity equal to that of the torch flame, and I ∽ = 5 is five times the intensity of the torch flame. This simplifies the interpretation of intensity values and the adjustment of desired intensity values.

[0052] To design the pool combustion state descriptor, the pool intensity must be normalized to be combined with the pool convexity. For this purpose, the saturation intensity I ∽ sat is set to be greater than the maximum value of I ∽ observed during the test cutting runs. In the disclosed method, I ∽ sat = 10, that is, the intensity saturates at 10 times the baseline. Therefore, we normalize the pool intensity to i = min(I ∽ sat, I ∽ ) / I ∽ sat, where i ∈ (0, 1].

[0053] Finally, the pool burning state descriptor is defined as s = λc + (1-λ)i, where the relative emphasis on pool convexity or pool intensity is adjusted by λ∈[0, 1]. This produces the desired normalized state descriptor s∈(0, 1]. In our experiments, we set λ = 1 / 2. By design, this descriptor captures the positive correlation between convexity and burning state (higher convexity indicates more concentrated burning), as well as the positive correlation between intensity and burning state (higher intensity indicates more intense burning).

[0054] In operation, the combustion control tasks performed by the cutting logic 142 assume that calibration and adjustment have been completed, meaning that the centroid and baseline intensity of the welding gun flame have been determined (so that the combustion state s can be calculated) and the metal surface has been sufficiently preheated for combustion to occur.

[0055] The cutting logic can be performed under the following assumptions:

[0056] The pose transformation between the welding torch flame and the camera is fixed: the camera is firmly attached to the welding torch, observing its tip from a fixed viewpoint throughout the cutting process, and keeping the welding torch flame stationary in the image frame. After calibration, the center of mass of the welding torch flame in the image frame does not change.

[0057] Pool burn state and torch speed have an inverse relationship: pool burn state is highly correlated with pool temperature. Faster torch speeds reduce the amount of heat transferred to the local metal surface, resulting in lower pool temperatures and less pool burn state. This is also confirmed by experience with cutting torches and sheet metal. Pool burn state decreases at higher torch speeds and increases at lower speeds. Note that burn state is the most direct, instantaneous measurement of the pool and is independent of residual heated areas elsewhere on the surface.

[0058] There is a desired speed for maintaining the desired combustion state: this is due to the inverse relationship between combustion state and torch speed. Moving the torch too quickly results in incomplete pool combustion, resulting in poor or unreliable cuts. Conversely, moving too slowly results in excessive combustion, producing an inefficient and rough cut. There is a range of intermediate speed values ​​that produce adequate combustion, within which a pair of desired speed and combustion states exists.

[0059] Typically, controlling the cutting speed involves reducing the cutting speed as the crown decreases and increasing the cutting speed as the crown increases. When the thickness of the cutting surface remains substantially constant, the cutting logic 142 can maintain the cutting speed at a rate that achieves an intensity between an upper and lower intensity limit, performing a complete cut at a crown above the minimum crown value. This occurs while maintaining the welding gun's launch angle perpendicular to the metal surface and at a substantially constant height above the metal surface.

[0060] Looking further into actual deployment, the robotic scrap metal cutting apparatus employs a robotic actuator 130 and a cutting torch 134, which is connected to and responsive to the actuator for movement along a predetermined cutting path 110 of a metal surface 101. A camera 132 is positioned to receive an image 133 of a heat pool 150, such that the heat pool is defined by the engagement of the cutting torch with the metal surface. Cutting logic 142 controls the speed of the cutting torch based on the convexity and intensity of the heat pool calculated from the image 133, and iterates in a control loop to continuously update the forward cutting speed, which typically varies based on the thickness of the cutting surface 101.

[0061] We formally define the variables of the tangential velocity of the heat source and the combustion state of the heat pool as follows:

[0062] v(t)∈R >0 It is the tangential velocity of the heat source (the centroid of the welding torch flame) on the reference cutting path.

[0063] s(t)=φ(v(t))R>0 is the pool burning state in the image frame, where φ models the mapping from v to s.

[0064] s * is the desired state, and v * is the desired speed.

[0065] The function φ:v1→s maps from the welding gun speed to the combustion state, thus allowing the following conditions:

[0066] φ is strictly positive: φ(v)>0, This is because, as defined above, c,i>0 leads to s>0.

[0067] φ is monotonically decreasing with respect to v:

[0068] d / dvφ(v)<0,, Let limφ(v)=0 as v→∞.

[0069] φ allows the desired degree constraint φ(v * )=s * .

[0070] The control input is the tangential acceleration v˙(t) that updates the velocity v, and the state s is a function of v. Therefore, we control the combustion state through acceleration. φ(v * )=s * The meaning is that by tracking s * , the welding gun will move at the desired speed v * Move. Set e s (t) = s * -s(v(t)) is the state relative to the desired state s* The combustion state error. Its dynamic equation is: Apply control input v˙(t)=-ke s (t), where k > 0 is a strictly positive gain. This results in the desired behavior of accelerating when s is too large and decelerating when s is insufficient.

[0071] Substituting t into the kinetic equation yields:

[0072]

[0073] Stability can be shown using the Lyapunov candidate function:

[0074]

[0075] Compute its derivative:

[0076]

[0077] The above result holds for all non-zero error states, since

[0078] k>0, and and

[0079] Therefore, the controlled system is asymptotically stable.

[0080] It should be noted that the desired combustion state s * Determined by experiment. Specifically, for a specific setting, a suitable range of combustion conditions will be found, and this suitable range of combustion conditions will not change during the cutting process. After a certain number of trials, the desired state s is determined. * The exact value of .

[0081] It should be readily understood by those skilled in the art that the procedures and methods defined herein can be delivered to a user processing and reproduction device in a variety of forms, including but not limited to: a) information permanently stored on a non-writable storage medium (such as a ROM device), b) information replaceably stored on a writable non-transitory storage medium (such as a solid-state drive (SSD) and media, flash drives, floppy disks, tapes, CDs, RAM devices, and other magnetic and optical media), or c) information transmitted to a computer via a communication medium in an electronic network (such as the Internet or a telephone modem line). These operations and methods can be implemented in a software executable object, or as a set of coded instructions executed by a processor in response to instructions, including a virtual machine and an execution environment controlled by a hypervisor. Alternatively, the operations and methods disclosed herein can be implemented in whole or in part using hardware components, such as an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a state machine, a controller, or other hardware component or device, or a combination of hardware, software, and firmware components.

[0082] While the systems and methods defined herein have been particularly shown and described with reference to embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the scope of the invention as encompassed by the appended claims.

Claims

1. A method for actuating a robotic cutting welding gun, comprising: guiding the cutting torch along the metal surface on a predetermined cutting path; receiving an image of a heat pool defined by engagement of the cutting torch with the metal surface; as well as The speed of the cutting torch is controlled based on the convexity and intensity of the heat pool calculated from the image.

2. The method according to claim 1, wherein The image of the heat pool represents a plurality of temperature zones, each temperature zone representing a temperature threshold.

3. The method according to claim 2, further comprising calculating the intensity by: Group the pixels in each region; weighting each group of pixels; and The intensity is calculated by clustering pixels weighted according to radial falloff.

4. The method according to claim 1, wherein The heat pool image represents a temperature region, and the convexity is based on the percentage of the bounding convex hull surrounding the region that the temperature region occupies.

5. The method according to claim 1, wherein The temperature region defines a group of pixels in the image that represent temperatures above a threshold value, such that all pixels in that temperature region represent temperatures above the threshold value.

6. The method according to claim 1, wherein The temperature region represents a group of pixels within a range of temperature values ​​between a specified threshold minimum value and a threshold maximum value.

7. The method of claim 4, further comprising reducing the cutting speed as the crown decreases.

8. The method of claim 4, further comprising increasing the cutting speed as the crown increases.

9. The method of claim 1 further comprising maintaining the cutting speed at a rate that achieves an intensity between an upper intensity limit and a lower intensity limit so that a complete cut is performed with a crown above a crown minimum value.

10. The method of claim 1 further comprising maintaining a firing angle of the welding gun perpendicular to the metal surface and at a substantially constant height above the metal surface.

11. The method according to claim 9, further comprising: performing the cutting by advancing a welding torch across a metal surface of a first thickness; advancing the welding gun against a metal surface of a second thickness different from the first thickness; as well as Adjust the cutting speed for: maintaining the intensity between an upper intensity limit and a lower intensity limit, and The convexity is maintained above a convexity minimum value.

12. The method of claim 1, further comprising filtering the image based on distance from a centroid of highest temperature pixels.

13. A robotic scrap metal cutting device comprising: Robotic actuators; a cutting torch connected to the actuator and moving along a predetermined cutting path of the metal surface in response to the actuator; a camera for receiving an image of a heat pool defined by engagement of the cutting torch with the metal surface; as well as Cutting logic for controlling the speed of the cutting torch based on the convexity and intensity of the heat pool calculated from the image.

14. The apparatus according to claim 13, wherein The robotic actuator is adapted to position the cutting torch perpendicular to a cutting surface and at a predetermined height above the cutting surface to advance the torch in a direction parallel to a tangent to the cutting surface.

15. The apparatus according to claim 12, wherein The cutting surface is metal and the cutting torch is an oxy-fuel torch.

16. A computer program embodying program code on a non-transitory computer readable medium, which, when executed by a processor, performs steps for implementing a method of actuating a robotic cutting welding gun, the method comprising: guiding the cutting torch along the metal surface on a predetermined cutting path; receiving an image of a heat pool defined by engagement of the cutting torch with the metal surface; as well as The speed of the cutting torch is controlled based on the convexity and intensity of the heat pool calculated from the image.

Images