Method for molten aluminum pouring by using mechanical arm based on multi-sensor fusion

Abstract

The present application relates to the technical field of industrial robots, in particular to a method for aluminum liquid pouring by using a mechanical arm based on multi-sensor fusion, comprising the following steps: acquiring a pouring gate center and a normal line by using a visual sensor, driving the mechanical arm to align and establish an initial aluminum liquid stream, collecting laser point clouds and obtaining a cross-sectional centroid principal axis and an area through principal component analysis, calculating a spatial deviation axis angle and a shrinkage rate to construct a stream state feature set, inputting a fuzzy PID to generate a posture and a valve port adjustment instruction and correcting joint speed and flow to realize stable pouring, in the present application, the pouring gate reference and the aluminum liquid stream shape are synchronously perceived, the cross-sectional centroid principal axis direction and the cross-sectional change are taken as control basis, the mechanical arm posture is directly related to the actual flow state, continuous correction is carried out when deviation or inclination occurs, and flow is adjusted in linkage, so that the forming process is stably controlled, and the influence of posture error and flow fluctuation on forming quality is reduced.

Description

Technical Field

[0001] This invention relates to the field of industrial robot technology, and in particular to a method for casting molten aluminum using a robotic arm based on multi-sensor fusion. Background Technology

[0002] The field of industrial robot technology involves utilizing robotics to perform various tasks in industrial production, including automated processing, handling, assembly, welding, and painting. This field encompasses several core areas, such as robot design and control technology, sensor fusion technology, path planning, artificial intelligence, and machine vision technology. Through precise motion control and automated operation, industrial robots significantly improve production efficiency and product quality, and have wide applications in modern manufacturing. With continuous technological advancements, robots are gradually acquiring more complex task execution capabilities, especially in dynamic and complex environments, where their adaptability and intelligence are constantly improving.

[0003] Traditional methods for aluminum molten metal casting using robotic arms, based on multi-sensor fusion, involve collecting key data from multiple sensors during the casting process, including temperature, flow rate, and location. This data is then processed and analyzed using data fusion technology to precisely control the robotic arm's trajectory and casting parameters. This method employs real-time acquisition and fusion of sensor data, ensuring the aluminum molten metal flows into the casting mold in a predetermined manner through precise control of the robotic arm's movement, thereby improving the stability and accuracy of the casting process. However, traditional aluminum molten metal casting methods rely on a single sensor for control and are prone to deviations and instabilities in dynamic environments.

[0004] Existing aluminum molten casting control methods mainly rely on the integration and judgment of process quantities such as temperature, flow rate, and position. The core of the control is concentrated on the matching of the robotic arm trajectory and fixed parameters. In the actual casting process, it is difficult to directly reflect the real spatial changes of the aluminum molten stream. When the stream is affected by environmental disturbances or flow fluctuations, the control system lacks effective basis for judging the center offset, axial tilt, and cross-sectional changes of the stream. There is a response lag between the robotic arm posture and the aluminum molten flow state. The correlation between valve adjustment and stream state is insufficient, which can easily lead to problems such as mold entry offset, uneven scouring, or unstable flow rate, thereby affecting the continuity of casting and the consistency of casting formation. Summary of the Invention

[0005] To address the technical problems of existing aluminum molten casting control methods that primarily rely on the fusion of process variables such as temperature, flow rate, and position, with the core control focusing on matching the robotic arm trajectory with fixed parameters, it is difficult to directly reflect the actual spatial morphological changes of the aluminum molten flow during the actual casting process. When the flow is affected by environmental disturbances or flow fluctuations, the control system lacks effective criteria for judging the flow center offset, axial tilt, and cross-sectional changes. There is a response lag between the robotic arm posture and the aluminum molten flow state, and the correlation between valve adjustment and flow state is insufficient, which can easily lead to problems such as mold entry offset, uneven scouring, or unstable flow, thereby affecting the continuity of casting and the consistency of casting formation. This invention provides a method for aluminum molten casting using a robotic arm based on multi-sensor fusion.

[0006] To achieve the above objectives, this invention employs a method for aluminum molten casting using a robotic arm based on multi-sensor fusion, comprising the following steps: S1: Use a visual positioning sensor to obtain the center coordinates of the gate and the normal vector of the gate, drive the robotic arm to align with the center coordinates of the gate and make the axis parallel to the normal vector of the gate, and establish the initial aluminum liquid flow stream. S2: Call the initial aluminum liquid stream, use a laser contour sensor to collect the stream contour point cloud data, use principal component analysis to calculate the geometric distribution characteristics of the stream contour point cloud, and generate the centroid coordinates, principal axis vector and contour area of ​​the stream cross section; S3: Calculate the spatial deviation vector between the centroid coordinates of the stream cross section and the center coordinates of the gate, the angle between the axis of the principal axis vector of the stream cross section and the axis of the gate normal vector, calculate the section shrinkage rate based on the profile area of ​​the stream cross section, and construct the stream state feature set; S4: Extract the spatial deviation vector and the angle between the axis and the fuzzy PID control algorithm from the set of flow state features to obtain the end attitude correction matrix, calculate the valve opening adjustment value based on the cross-sectional shrinkage rate, and generate the casting coordinated adjustment command; S5: The end-effector posture correction matrix in the casting coordination adjustment command is used to calculate the joint angular velocity command to correct the posture through the inverse kinematics algorithm of the robotic arm, adjust the aluminum liquid flow rate, and establish a stable casting state for aluminum liquid entering the mold.

[0007] As a further aspect of the present invention, the initial aluminum liquid stream includes the spatial position coordinates of the stream, the consistency of the stream axis direction vector, and the flow continuity characteristics of the stream; the stream cross-sectional characteristic data includes the cross-sectional centroid stability index, the cross-sectional principal axis direction consistency index, and the cross-sectional contour integrity index; the stream state characteristic set includes the spatial position offset, the axis angle deviation, and the cross-sectional area change rate; the casting collaborative adjustment command includes the attitude correction amount, the flow rate adjustment amount, and the collaborative control matching amount; and the stable casting state includes the flow uniformity index, the attitude stability index, and the filling continuity index.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Use a visual positioning sensor to collect image data frames of the gate area, calculate the gradient direction of the pixels at the edge of the image, determine the boundary point set based on the consistency of the gradient direction, perform spatial mapping and converge the center point coordinates, estimate the normal vector based on the surface fitting of the boundary point set, and obtain the center coordinates of the gate and the gate normal vector. S102: Call the center coordinates of the gate and the normal vector of the gate, calculate the direction cosine of the end axis and the normal according to the current pose parameter set of the robot arm, adjust the joint angle according to the angle deviation, perform spatial position vector difference correction, and generate the robot arm alignment pose parameter set; S103: Based on the alignment pose parameter set of the robotic arm, perform consistency judgment on the flow velocity vector parameter of the aluminum liquid conveying channel and the outlet section direction vector, compare the flow direction vector with the direction of the robotic arm axis, and perform stability judgment on the flow state in the continuous time sequence to establish the initial aluminum liquid flow stream.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Call the initial aluminum liquid stream, use a laser contour sensor to collect stream contour point cloud data, perform time alignment based on the point cloud data obtained by continuous scanning, map the three-dimensional coordinates of the points, judge discrete points based on the continuity of spatial position and remove outliers, and generate stream contour point cloud coordinate set. S202: Based on the point cloud coordinate set of the stream contour, construct the point cloud coordinate distribution matrix, perform mean elimination processing on the matrix dimension, calculate the covariance matrix and extract the eigenvalue distribution, compare the size relationship of the eigenvalues, determine the principal component direction according to the direction corresponding to the principal eigenvalue, and generate the principal component direction discrimination vector set. S203: Based on the principal component direction discrimination vector set and the flow stream contour point cloud coordinate set, perform projection operation on the point cloud coordinates along the discrimination direction, calculate the spatial mean of the projected coordinates and aggregate the positions, perform plane segmentation and surface accumulation, and generate the centroid coordinates, principal axis vector and contour area of ​​the flow stream cross section.

[0010] As a further aspect of the present invention, the step of performing time-series alignment based on the point cloud data obtained by continuous scanning and mapping the three-dimensional coordinates of the points refers to matching the timestamps of the point cloud data obtained by two adjacent scans according to the fixed scanning frequency of the laser contour sensor, and completing the coordinate alignment under the condition that the time difference is not greater than the preset sampling period threshold. The step of judging discrete points and removing outliers based on the continuity of spatial location refers to calculating the average Euclidean distance between any point in the coordinate set of the convection stream contour point cloud and a preset number of adjacent points. When the average Euclidean distance exceeds a preset spatial distance threshold, the corresponding point is judged as an outlier and removed.

[0011] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Call the centroid coordinates of the flow section and the center coordinates of the gate, perform a difference operation on the two sets of spatial coordinates in the corresponding coordinate axis directions, perform a direction consistency retrieval on the difference results and construct a vector component combination, perform amplitude merging and direction identification on the vector components, and generate a centroid spatial deviation vector. S302: Based on the centroid space deviation vector, call the principal axis vector of the flow section and the gate normal vector, match the direction components of the two vectors one by one, perform dot product operation on the matched components, and perform ratio conversion in combination with the vector magnitude. Based on the angle definition relationship, perform mapping processing to obtain the characteristic value of the included angle of the axis. S303: Based on the characteristic value of the included angle of the axis and the centroid spatial deviation vector, call the cross-sectional contour area of ​​the flow stream, perform a ratio calculation on the current cross-sectional area and the reference cross-sectional area, and classify and mark the ratio results according to the interval division rules to establish a set of flow stream state features.

[0012] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Extract the spatial deviation vector and the angle between the axis from the set of flow state features, perform fuzzy membership mapping on the spatial deviation vector components and the angle between the axis respectively, substitute the mapping results into the proportional term and differential term rule table to perform rule matching, and generate attitude deviation control quantity; S402: Based on the attitude deviation control quantity, perform deviation accumulation operation between the control quantity and the current state parameter in the attitude dimension to construct the integral parameter sequence, and map the proportional quantity, integral quantity, and differential quantity to the corresponding matrix row and column positions according to the attitude dimension to establish the end attitude correction matrix. S403: Call the end attitude correction matrix to obtain the cross-sectional shrinkage rate parameter, construct it as a change amplitude, perform the association mapping with the attitude correction component in the end attitude correction matrix, and perform interval determination and discrete encoding on the result, calculate the valve opening adjustment value, and generate the casting coordinated adjustment command.

[0013] As a further aspect of the present invention, the spatial deviation vector component and the angle between the axis are respectively subjected to fuzzy membership mapping, which divides the space into at least three continuous membership segments according to a preset angle interval, and the membership degree of the membership segments changes using a monotonic linear mapping method.

[0014] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Extract the end attitude correction matrix from the casting coordination adjustment command, calculate the end attitude angular velocity sequence based on the time derivative of adjacent attitude matrices, use the inverse kinematics algorithm of the robotic arm to convert the end angular velocity into joint angular velocity command, perform sign consistency check and interval determination, and generate joint angular velocity correction pose sequence. S502: Based on the joint angular velocity correction pose sequence, extract the valve opening adjustment value flow velocity vector parameter, convert the joint angular velocity into the end linear velocity through the kinematic forward solution, establish the spatial correspondence between the flow velocity vector and the end motion velocity, perform amplitude normalization mapping, and obtain the aluminum liquid flow coordinated adjustment configuration. S503: Based on the aluminum liquid flow rate coordinated adjustment configuration, call the attitude change component in the joint angular velocity correction pose sequence, perform state combination calculation of flow velocity vector parameter and attitude change component, perform interval judgment and state identifier mapping according to the state transition rules of the mold entry process, and establish a stable pouring state of aluminum liquid entering the mold.

[0015] As a further aspect of the present invention, during the process of executing symbol consistency verification and interval determination, a predetermined threshold range is set for the joint angular velocity command. When the amplitude of the joint angular velocity changes within the threshold range, it is determined to be a valid adjustment range. When it exceeds the threshold range, the execution of the joint angular velocity correction command is paused and an exception handling mechanism is triggered.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by synchronously sensing the spatial reference of the gating gate and the geometric shape of the molten aluminum stream, the centroid of the stream cross-section, the direction of the main axis, and the changes in the cross-section are introduced into the casting control criteria. This directly links the robot arm's posture adjustment with the actual flow state of the molten aluminum, enabling continuous correction when the stream experiences spatial offset or directional change. Simultaneously, the flow rate of the molten aluminum is adjusted in conjunction with the changes in the stream cross-section, keeping the mold entry process stable and controlled. This reduces the impact of posture errors and flow fluctuations on the forming quality during casting, and improves the stability and consistency of molten aluminum casting under complex working conditions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the accompanying drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0019] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0020] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0021] Please see Figure 1 This invention provides a method for casting molten aluminum using a robotic arm based on multi-sensor fusion, comprising the following steps: S1: Use a visual positioning sensor to obtain the center coordinates of the gate and the normal vector of the gate, drive the robotic arm to align with the center coordinates of the gate and make the axis parallel to the normal vector of the gate, and establish the initial aluminum liquid flow stream. S2: Call the initial aluminum liquid stream, use a laser profile sensor to collect stream profile point cloud data, use principal component analysis to calculate the geometric distribution characteristics of the stream profile point cloud, and generate the centroid coordinates, principal axis vector and profile area of ​​the stream cross section. S3: Calculate the spatial deviation vector between the centroid coordinates of the stream section and the center coordinates of the gate, the angle between the axis of the principal axis vector of the stream section and the axis of the gate normal vector, calculate the section shrinkage rate based on the profile area of ​​the stream section, and construct the stream state feature set; S4: Extract the spatial deviation vector and the angle between the axis and the fuzzy PID control algorithm from the set of flow state features to obtain the end attitude correction matrix, calculate the valve opening adjustment value based on the cross-sectional shrinkage rate, and generate the casting coordinated adjustment command; S5: The end-effector posture correction matrix in the casting coordination adjustment command is used to calculate the joint angular velocity command to correct the posture, adjust the aluminum liquid flow rate, and establish a stable casting state for aluminum liquid entering the mold.

[0022] The initial aluminum liquid stream includes the spatial coordinates of the stream, the consistency of the stream axis direction vector, and the flow continuity characteristics of the stream. The stream cross-sectional characteristic data includes the cross-sectional centroid stability index, the cross-sectional principal axis direction consistency index, and the cross-sectional profile integrity index. The stream state characteristic set includes the spatial position offset, the axis angle deviation, and the cross-sectional area change rate. The casting coordination adjustment commands include the attitude correction amount, the flow rate adjustment amount, and the coordination control matching amount. The stable casting state includes the flow uniformity index, the attitude stability index, and the filling continuity index.

[0023] Please see Figure 2 The specific steps of S1 are as follows: S101: Use a visual positioning sensor to collect image data frames of the gate area, calculate the gradient direction of the pixels at the edge of the image, determine the boundary point set based on the consistency of the gradient direction, perform spatial mapping and converge the center point coordinates, estimate the normal vector based on the surface fitting of the boundary point set, and obtain the center coordinates of the gate and the gate normal vector. Image data frames of the gate area are acquired using a visual positioning sensor. An industrial-grade CMOS visual positioning sensor continuously captures images of the gate area at a frequency of 60 frames per second, obtaining raw grayscale image data frames with a resolution of 2048×2048 pixels. The process first performs Gaussian filtering to denoise the acquired raw images, then uses a 5×5 convolution kernel to smooth the image pixels. Subsequently, the Sobel operator is used to calculate the pixel gradient values ​​along the horizontal and vertical directions. The arithmetic square root of the sum of the squared horizontal and vertical gradients is taken as the gradient magnitude of that pixel. Simultaneously, the arctangent function is used to calculate the gradient direction angle. The process of determining the boundary point set based on gradient direction consistency involves setting a gradient magnitude threshold of 150. Pixels with a gradient magnitude greater than this threshold are marked as candidate edge points. The candidate points are then traversed, and the gradient magnitude of the current point is compared with the gradient magnitude of its two adjacent pixels along the gradient direction. If the current point has the largest gradient magnitude, it is retained as an edge. If a point is selected, it is suppressed, thus refining the gate edge contour with a single pixel width. Then, spatial mapping is performed and the center point coordinates are converged. This process calls the pre-calibrated camera intrinsic matrix and hand-eye extrinsic matrix to convert the edge point coordinates in the image pixel coordinate system into three-dimensional spatial coordinates in the robot arm base coordinate system. The least squares method is used to perform circle fitting on the converted three-dimensional edge point set. By minimizing the sum of squared distances from the edge points to the center of the fitted circle, the center coordinates of the fitted circle are calculated as the gate center coordinates. The gate normal vector is estimated based on the surface fitting method of the boundary point set. In the set of normal vectors of the fitted plane, the dot product of the normal vector and the gravity direction vector is calculated, and the vector with a positive dot product result is selected as the vertically upward gate normal vector. For example, in an actual acquisition process, the calculated gate center coordinates are 500.5 mm on the X-axis, 200.2 mm on the Y-axis, and 150.0 mm on the Z-axis, and the gate normal vectors are 0.0, 0.0, and 1.0.

[0024] S102: Call the gate center coordinates and gate normal vector, calculate the direction cosine of the end axis and normal according to the current pose parameter set of the robot arm, adjust the joint angle according to the angle deviation, perform spatial position vector difference correction, and generate the robot arm alignment pose parameter set; The process involves calling the gate center coordinates and gate normal vector, initiating the forward kinematics algorithm for the robotic arm, and reading the real-time angle values ​​from the robotic arm joint encoders. Combined with the robotic arm link length parameters (e.g., upper arm length 800 mm, lower arm length 600 mm), the current spatial coordinates of the tool center point (TCP) and the end effector axis attitude vector are derived using the homogeneous transformation matrix multiplication rule. The process of calculating the direction cosine of the end effector axis and normal vector based on the current pose parameter set involves performing a dot product operation between the end effector axis vector and the gate normal vector, dividing the dot product by the product of the magnitudes of the two vectors to obtain the cosine of the angle between them. The process of adjusting the joint angle based on the angle deviation involves setting the alignment accuracy threshold to 0.5 degrees. When the calculated included angle is greater than the threshold, a differential motion equation is constructed using the inverse of the Jacobian matrix. The position deviation vector and the attitude deviation vector are mapped to the angular velocity command of the joint. The process of correcting the spatial position vector difference refers to calculating the algebraic difference between the coordinates of the gate center and the current coordinates of the end of the robot arm in the X, Y, and Z directions, and generating a position compensation vector. For example, if the current end coordinates are 510.0 mm on the X axis, 205.0 mm on the Y axis, and 300.0 mm on the Z axis, then the position difference vector is -9.5 mm, -4.8 mm, and -150.0 mm. This process plans a motion trajectory that smoothly transitions from the current pose to the target alignment pose through an iterative interpolation algorithm, and generates a set of robot arm alignment pose parameters.

[0025] S103: Based on the alignment pose parameter set of the robotic arm, perform consistency judgment on the velocity vector parameter of the aluminum liquid conveying channel and the direction vector of the outlet section, compare the flow direction vector with the direction of the robotic arm axis, and perform stability judgment on the flow state in a continuous time sequence to establish the initial aluminum liquid flow stream; Based on the alignment pose parameter set of the robotic arm, the aluminum liquid delivery pump is activated, and the real-time reading of the electromagnetic flowmeter is used as the flow rate parameter of the aluminum liquid delivery channel. Simultaneously, the outlet cross-sectional direction vector is obtained through a miniature inertial measurement unit (IMU) installed at the outlet. The flow velocity vector parameter is calculated by combining this with the outlet cross-sectional area. The process of verifying the consistency between the flow velocity vector parameter of the aluminum liquid delivery channel and the outlet cross-sectional direction vector involves setting a flow threshold of 1.5 kg / s. When the flow velocity vector parameter reaches this threshold and the angle between the outlet cross-sectional direction vector and the gravity vector is less than 2 degrees, the outflow condition is deemed met. The flow direction vector is then compared with the direction of the robotic arm axis. The process of calculating the parallelism between the outflow velocity vector of the calculated fluid and the axis vector of the robotic arm's end effector requires that the angle deviation between the two be kept within 1.0 degree. The process of performing stability judgment on the flow state within a continuous time sequence refers to establishing a sliding time window of length 50, calculating the variance of the flow data within the window, and determining that the flow is stable if the variance is less than 0.05. The process of establishing the initial aluminum liquid flow stream refers to maintaining the current pumping power and robotic arm posture after confirming that the above conditions are met, forming a continuous and stable columnar aluminum liquid flow. For example, under the condition of a flow rate of 2.0 kg per second and a flow velocity direction deviation of 0.5 degrees, the initial flow stream is confirmed to be established.

[0026] Please see Figure 3 The specific steps of S2 are as follows: S201: Call the initial aluminum liquid flow stream, use the laser contour sensor to collect the flow stream contour point cloud data, perform time alignment based on the point cloud data obtained by continuous scanning, map the three-dimensional coordinates of the points, judge discrete points based on the continuity of spatial position and remove outliers, and generate the flow stream contour point cloud coordinate set. The initial aluminum molten stream is invoked, activating the line laser contour sensor installed at the end of the robotic arm. The scanning frequency is set to 200 Hz. The laser contour sensor collects point cloud data of the stream contour. This process involves projecting a laser beam onto the surface of the aluminum molten stream to form a light stripe. The sensor's photosensitive chip records the deformation information of the light stripe and converts it into depth data. Based on the point cloud data obtained from continuous scanning, a time-series alignment process is performed. The nanosecond-level hardware timestamp attached to each frame of point cloud data is read, and the time axis of the laser sensor is synchronized with the time axis of the robotic arm controller. The timestamps of point cloud data obtained from two adjacent scans are matched. Only when the absolute value of the time difference between two timestamps is less than a preset sampling period threshold of 5 milliseconds are these two frames of data considered valid data from the same moment for coordinate alignment. The three-dimensional coordinates of the points are mapped according to the calibrated hand-eye relationship matrix. Point cloud data in the sensor coordinate system is uniformly converted to the robotic arm base coordinate system. The process of judging discrete points and removing outliers based on the continuity of spatial position refers to traversing each data point in the flow contour point cloud coordinate set, taking the point as the center and a radius of 5 mm as the search range, and searching for the number of neighboring points within this range. If the number of neighboring points is less than 3, or if the average Euclidean distance between the point and a preset number of adjacent points (e.g., 10) exceeds a preset spatial distance threshold of 2.0 mm, the point is marked as a splash noise point or invalid reflection point and removed from the dataset, generating a clean flow contour point cloud coordinate set. For example, for a test point with coordinates of X-axis 500.0, Y-axis 200.0, and Z-axis 250.0, the average distance between it and 10 neighboring points is calculated to be 2.5 mm, which is greater than the threshold of 2.0 mm. Therefore, the point is judged as an outlier and removed.

[0027] S202: Based on the point cloud coordinate set of the stream contour, construct the point cloud coordinate distribution matrix, perform mean elimination processing on the matrix dimension, calculate the covariance matrix and extract the eigenvalue distribution, compare the relationship between the eigenvalues, determine the principal component direction according to the direction corresponding to the principal eigenvalue, and generate the principal component direction discrimination vector set. Based on the point cloud coordinate set of the flow contour, a point cloud coordinate distribution matrix is ​​constructed. This matrix is ​​an N x 3 matrix, where N is the total number of points in the point cloud, and the three columns correspond to the X, Y, and Z axis coordinate values, respectively. The process of performing mean elimination on the matrix involves calculating the arithmetic mean of each column and subtracting the mean of its column from each element to obtain a centered coordinate matrix. The process of calculating the covariance matrix and extracting the eigenvalue distribution involves multiplying the transpose of the centered matrix with the centered matrix itself and dividing the product by N minus 1 to obtain a 3 x 3 covariance matrix. Subsequently, the Jacobi iteration method or the QR algorithm is used to refine this covariance matrix. The process involves performing eigenvalue decomposition to extract three non-negative eigenvalues ​​and their corresponding eigenvectors. Comparing the magnitudes of these eigenvalues ​​involves arranging them in descending order of value. Determining the principal component direction based on the direction corresponding to the principal eigenvalue involves selecting the eigenvector corresponding to the largest eigenvalue as the principal axis direction of the flow (i.e., the longitudinal direction of the flow), and selecting the eigenvector corresponding to the second largest eigenvalue as the major axis direction of the flow cross-section. This generates a set of principal component direction discrimination vectors. For example, if the calculated three eigenvalues ​​are 150.0, 20.0, and 5.0, then the vector corresponding to the eigenvalue 150.0 is determined as the principal flow direction vector.

[0028] S203: Based on the principal component direction discrimination vector set and the flow stream contour point cloud coordinate set, perform projection operation on the point cloud coordinates along the discrimination direction, calculate the spatial mean of the projected coordinates and aggregate the positions, perform plane segmentation and surface accumulation, and generate the centroid coordinates, principal axis vector and contour area of ​​the flow stream section; Based on the principal component direction discrimination vector set and the flow stream contour point cloud coordinate set, a virtual cross-sectional plane perpendicular to the main flow direction is constructed. The process of projecting the point cloud coordinates along the discrimination direction refers to projecting each point in the point cloud onto the virtual plane along the main flow direction vector, transforming the three-dimensional coordinates into two-dimensional plane coordinates. The spatial mean of the projected coordinates is calculated and the position is aggregated, which means calculating the geometric center of the projected points on the two-dimensional plane. The process of performing plane segmentation and surface accumulation refers to using the AlphaShape algorithm or convex hull algorithm to extract the contour boundary of the projected point set, dividing the region within the contour into several non-overlapping micro-triangles, calculating the area of ​​each micro-triangle using Heron's formula and adding them together to obtain the total area of ​​the flow stream cross-section. The centroid coordinates, principal axis vectors, and contour area of ​​the flow stream cross-section are generated. For example, after projection and surface accumulation calculation, the contour area of ​​the flow stream cross-section is 314.1 square millimeters, and the centroid coordinates in the base coordinate system are 501.2 mm on the X-axis, 200.8 mm on the Y-axis, and 250.0 mm on the Z-axis.

[0029] Please see Figure 4 The specific steps of S3 are as follows: S301: Call the centroid coordinates of the flow section and the center coordinates of the gate, perform a difference operation on the two sets of spatial coordinates in the corresponding coordinate axis directions, perform a direction consistency search on the difference results and construct a vector component combination, perform amplitude merging and direction identification on the vector components, and generate a centroid spatial deviation vector. The process involves calculating the differences between the X-coordinate of the flow centroid and the X-coordinate of the gate center, the Y-coordinate of the flow centroid and the Y-coordinate of the gate center, and the Z-coordinate of the flow centroid and the Z-coordinate of the gate center, respectively. The process then performs direction consistency retrieval on the difference results and constructs a vector component combination. The process confirms the offset direction represented by the difference sign, combines the three difference components into a three-dimensional deviation vector, performs amplitude merging and direction identification on the vector component, calculates the magnitude of the deviation vector, and records its projection azimuth angle in the horizontal plane, generating a centroid spatial deviation vector. For example, if the gate center is (500, 200, 150) and the flow centroid is (502, 202, 150), the vector component obtained from the difference operation is (2, 2, 0), the magnitude of this vector is 2.828 mm, and the direction identification is 45 degrees northeast.

[0030] S302: Based on the centroid space deviation vector, call the principal axis vector of the flow section and the gate normal vector, match the direction components of the two vectors one by one, perform dot product operation on the matched components, and perform ratio conversion in combination with the vector magnitude. Based on the angle definition relationship, perform mapping processing to obtain the characteristic value of the included angle of the axis. The process involves extracting the X, Y, and Z components of two vectors, performing a dot product operation on the matched components, multiplying the corresponding X, Y, and Z components, summing the results to obtain the dot product value, and then performing a ratio conversion based on the vector magnitudes. This involves calculating the product of the two vector magnitudes, dividing the dot product by this product to obtain the cosine value, and then performing a mapping process based on the defined angle relationship to obtain the characteristic value of the axis angle. This process uses the inverse cosine function to convert the cosine value into an angle value, with a range of 0 to 180 degrees. For example, if the main axis vector of the flow stream is (0.1, 0, 0.995) and the gate normal vector is (0, 0, 1), the dot product is 0.995, the magnitude product is 1, then the cosine value is 0.995, and the corresponding characteristic value of the axis angle is 5.73 degrees.

[0031] S303: Based on the characteristic value of the included angle of the axis and the centroid spatial deviation vector, call the cross-sectional area of ​​the flow stream, perform a ratio calculation on the current cross-sectional area and the reference cross-sectional area, and classify and mark the ratio results according to the interval division rules to establish a set of flow stream state features; The reference cross-sectional area is a theoretical value (e.g., 310 square millimeters) calculated in advance based on ideal flow rate and velocity. The calculation formula is the current area divided by the reference area to obtain the area ratio. The process of classifying and labeling the ratio result according to the interval division rules refers to the following: if the area ratio is between 0.95 and 1.05, it is labeled as "stable"; if it is less than 0.95, it is labeled as "contraction"; if it is greater than 1.05, it is labeled as "diffusion". A set of flow state characteristics is established. For example, if the current area is 314.1 square millimeters and the reference area is 310.0 square millimeters, the calculated ratio is 1.013, which falls within the interval of 0.95 to 1.05. Therefore, this characteristic is labeled as "stable". The final generated set of flow state characteristics includes: centroid deviation of 2.828 mm, axis angle of 5.73 degrees, and flow state of "stable".

[0032] Please see Figure 5 The specific steps of S4 are as follows: S401: Extract the spatial deviation vector and the angle between the axis from the flow state feature set, perform fuzzy membership mapping on the spatial deviation vector components and the angle between the axis respectively, substitute the mapping results into the proportional and differential term rule tables to perform rule matching, and generate attitude deviation control quantity; First, a fuzzy controller is established, dividing the magnitude of the spatial deviation vector component into three fuzzy subsets: "small," "medium," and "large." The included angle of the axis is also divided into three fuzzy subsets: "small," "medium," and "large." Fuzzy membership mapping is then applied to the spatial deviation vector component and the included angle of the axis, dividing them into at least three consecutive membership segments according to a preset angle range. For example, for the included angle of the axis, 0 to 2 degrees belongs to "small," 2 to 5 degrees belongs to "medium," and above 5 degrees belongs to "large." Furthermore, the change in membership degree of the membership segments adopts a monotonically linear mapping method, i.e., using... Using triangular or trapezoidal membership functions, specific numerical values ​​are converted into membership values ​​between 0 and 1. The mapping results are then substituted into the proportional and differential term rule tables to perform rule matching. The rule tables pre-set logic such as "if the deviation is large and the included angle is large, then the output control quantity is a strong correction" to generate attitude deviation control quantities. For example, when the included angle of the axis is 5.73 degrees, its membership degree on the "medium" fuzzy set is 0.2, and its membership degree on the "large" fuzzy set is 0.8. According to the rule weighted calculation, a normalized attitude deviation control quantity value of 0.75 is output.

[0033] S402: Based on the attitude deviation control quantity, perform deviation accumulation calculation between the control quantity and the current state parameters in the attitude dimension to construct the integral parameter sequence, and map the proportional quantity, integral quantity, and differential quantity to the corresponding matrix row and column positions according to the attitude dimension to establish the end attitude correction matrix. Based on the attitude deviation control quantity, the control quantity and the current state parameters are subjected to deviation accumulation calculation in the attitude dimension. This process adopts the discrete-time integration method, which weights and sums the attitude deviation control quantity at the current moment with the cumulative error at the previous moment to construct an integral parameter sequence to eliminate steady-state error. The proportional quantity (directly affected by the current deviation), the integral quantity (accumulated historical deviation), and the differential component (deviation change rate) are mapped to the corresponding matrix row and column positions according to the attitude dimension (rotation around the X, Y, and Z axes) to establish the end attitude correction matrix. This matrix is ​​a 4x4 homogeneous transformation matrix, where the rotation submatrix is ​​generated by Euler angle transformation and the translation vector is temporarily set to zero. For example, if the calculated correction angle around the X-axis is 1.5 degrees and the correction angle around the Y-axis is -0.5 degrees, then the corresponding rotation correction matrix is ​​constructed.

[0034] S403: Call the end attitude correction matrix, obtain the cross-sectional shrinkage rate parameter, construct it as the change amplitude, perform the association mapping with the attitude correction component in the end attitude correction matrix, and perform interval judgment and discrete encoding on the result, calculate the valve opening adjustment value, and generate the casting coordinated adjustment command. The end attitude correction matrix is ​​called to obtain the cross-sectional shrinkage rate parameter, which comes from the aforementioned area ratio data. This parameter is constructed as a change amplitude and is associated with the attitude correction component in the end attitude correction matrix. That is, the coupling relationship between cross-sectional area change and attitude adjustment is analyzed. If the attitude adjustment amplitude is too large, it will cause severe deformation of the flow cross-section. Therefore, the attitude correction amount needs to be limited, and the result is interval-determined and discretely encoded. The correction amplitude is divided into 5 levels. The process of calculating the valve opening adjustment value refers to calculating the opening increment of the flow valve through the PID algorithm based on the difference between the area ratio and its target value of 1.0, and generating the casting coordinated adjustment command. This command contains attitude correction matrix data and valve opening adjustment percentage data, as shown in Table 1 for example of coordinated adjustment parameters.

[0035] Table 1: Casting Coordination Adjustment Parameter Table

[0036] As shown in Table 1, the data in the table shows the specific execution amount of the coordinated regulation. The positive correction angle around the X-axis and the negative correction angle around the Y-axis work together to calibrate the flow attitude. At the same time, the flow rate is compensated by a 2.3% valve opening increment, so that the predicted cross-sectional ratio returns to the ideal stable range of 1.005.

[0037] Please see Figure 6 The specific steps of S5 are as follows: S501: Extract the end-effector attitude correction matrix from the casting coordination adjustment command, calculate the end-effector attitude angular velocity sequence based on the time derivative of adjacent attitude matrices, use the inverse kinematics algorithm of the robotic arm to convert the end-effector angular velocity into joint angular velocity commands, perform sign consistency check and interval determination, and generate joint angular velocity correction pose sequence. The Cartesian space end-effector attitude correction is converted into joint angular velocity commands through the inverse operation of the Jacobian matrix. Specifically, the rotation matrix is ​​extracted from the 4×4 homogeneous transformation matrix and converted into Euler angle representation (Roll, Pitch, Yaw). Based on the difference between the current end-effector attitude and the target attitude, and combined with the preset adjustment time, the angular velocity vector ω=[ω x ω y ω z Then, through the inverse kinematics Jacobian matrix J... -1 The angular velocity vector in Cartesian space is mapped to a six-dimensional joint angular velocity vector θ = [θ1, θ2, θ3, θ4, θ5, θ6,]ᵀ. When performing sign consistency checks and interval determination, a zero threshold is used as the criterion for positive and negative sign determination to check whether the calculated joint angular velocity direction conflicts with the physical motion constraint direction of the joint. The interval determination uses symmetrical closed intervals for validity screening. For example, the upper limit of the joint angular velocity is set to 0.524 radians / second (30 degrees / second), and the lower limit is -0.524 radians / second. If the calculated value exceeds this range, it is truncated to the boundary value, generating a joint angular velocity correction pose sequence. For example, the generated sequence is (0.1, 0.2, -0.1, 0.5, 0.0, -0.3) radians per second.

[0038] S502: Based on the joint angular velocity correction pose sequence, extract the flow velocity vector parameters of the valve opening adjustment value, convert the joint angular velocity into the end linear velocity through the kinematic forward solution, establish the spatial correspondence between the flow velocity vector and the end motion velocity, perform amplitude normalization mapping, and obtain the aluminum liquid flow coordinated adjustment configuration. The three-dimensional vector components of the target flow velocity change were extracted from the casting coordination control command. x v y v z The joint angular velocity is converted into the terminal linear velocity using forward kinematics, and the terminal Cartesian space linear velocity vector V = [V] is obtained by forward operation of the Jacobian matrix J·θ. x V y V zTo establish the spatial correspondence between the flow velocity vector and the end effector velocity, the flow velocity vector is transformed from the fluid coordinate system to the end effector coordinate system of the robotic arm through coordinate transformation. The spatial angle and amplitude ratio of the two velocity vectors are calculated. When the angle is less than a set threshold (e.g., 15 degrees) and the amplitude ratio is within a reasonable range (e.g., 0.5-2.0), the spatial matching degree is considered to be up to standard. Amplitude normalization mapping is then performed, mapping the change in flow velocity vector parameters (e.g., an increase of 0.1 m / s per second) and the end effector linear velocity to dimensionless control signals between 0 and 1. The component with the largest absolute value is used as the normalization benchmark to obtain the aluminum liquid flow coordinated adjustment configuration, ensuring that the flow velocity adjustment and attitude adjustment match in response characteristics and avoiding flow instability caused by speed incoordination.

[0039] S503: Based on the coordinated adjustment configuration of aluminum liquid flow, call the joint angular velocity to correct the posture change component in the pose sequence, perform state combination calculation of flow velocity vector parameter and posture change component, perform interval judgment and state identifier mapping according to the state transition rule of the mold entry process, and establish a stable pouring state of aluminum liquid entering the mold. This process constructs a two-dimensional state vector from the normalized flow signal and attitude signal. It performs interval judgment and state identifier mapping according to the state transition rules of the mold entry process. The state transition rules define the transition conditions of stages such as "injection", "filling" and "holding pressure". For example, when the cumulative injection volume reaches 80% of the mold volume and the variance of the attitude change component is less than 0.01, the state switches from "injection" to "filling" to establish a stable pouring state of aluminum liquid. This process is continuously monitored and dynamically adjusted until the pouring is completed. For example, in the current case, the two-dimensional state vector (0.8, 0.15) determines that the current state is "efficient and stable injection" and maintains this control output.

[0040] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the described technical solutions.

Claims

1. A method for casting molten aluminum using a robotic arm based on multi-sensor fusion, characterized in that, Includes the following steps: S1: Use a visual positioning sensor to obtain the center coordinates of the gate and the normal vector of the gate, drive the robotic arm to align with the center coordinates of the gate and make the axis parallel to the normal vector of the gate, and establish the initial aluminum liquid flow stream. S2: Call the initial aluminum liquid stream, use a laser contour sensor to collect the stream contour point cloud data, use principal component analysis to calculate the geometric distribution characteristics of the stream contour point cloud, and generate the centroid coordinates, principal axis vector and contour area of ​​the stream cross section; S3: Calculate the spatial deviation vector between the centroid coordinates of the stream cross section and the center coordinates of the gate, the angle between the axis of the principal axis vector of the stream cross section and the axis of the gate normal vector, calculate the section shrinkage rate based on the profile area of ​​the stream cross section, and construct the stream state feature set; S4: Extract the spatial deviation vector and the angle between the axis and the fuzzy PID control algorithm from the set of flow state features to obtain the end attitude correction matrix, calculate the valve opening adjustment value based on the cross-sectional shrinkage rate, and generate the casting coordinated adjustment command; S5: The end-effector posture correction matrix in the casting coordination adjustment command is used to calculate the joint angular velocity command to correct the posture through the inverse kinematics algorithm of the robotic arm, adjust the aluminum liquid flow rate, and establish a stable casting state for aluminum liquid entering the mold.

2. The method for aluminum molten casting using a robotic arm based on multi-sensor fusion according to claim 1, characterized in that, The initial aluminum liquid stream includes the stream's spatial position coordinates, the consistency of the stream's axial direction vector, and the stream's flow continuity characteristics. The stream's cross-sectional characteristic data includes the cross-sectional centroid stability index, the cross-sectional principal axis direction consistency index, and the cross-sectional contour integrity index. The stream's state characteristic set includes spatial position offset, axial angle deviation, and cross-sectional area change rate. The casting coordination adjustment command includes attitude correction, flow rate adjustment, and coordination control matching. The stable casting state includes flow uniformity index, attitude stability index, and filling continuity index.

3. The method for aluminum molten casting using a robotic arm based on multi-sensor fusion according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Use a visual positioning sensor to collect image data frames of the gate area, calculate the gradient direction of the pixels at the edge of the image, determine the boundary point set based on the consistency of the gradient direction, perform spatial mapping and converge the center point coordinates, estimate the normal vector based on the surface fitting of the boundary point set, and obtain the center coordinates of the gate and the gate normal vector. S102: Call the center coordinates of the gate and the normal vector of the gate, calculate the direction cosine of the end axis and the normal according to the current pose parameter set of the robot arm, adjust the joint angle according to the angle deviation, perform spatial position vector difference correction, and generate the robot arm alignment pose parameter set; S103: Based on the alignment pose parameter set of the robotic arm, perform consistency judgment on the flow velocity vector parameter of the aluminum liquid conveying channel and the outlet section direction vector, compare the flow direction vector with the direction of the robotic arm axis, and perform stability judgment on the flow state in the continuous time sequence to establish the initial aluminum liquid flow stream.

4. The method for aluminum molten casting using a robotic arm based on multi-sensor fusion according to claim 3, characterized in that, The specific steps of S2 are as follows: S201: Call the initial aluminum liquid stream, use a laser contour sensor to collect stream contour point cloud data, perform time alignment based on the point cloud data obtained by continuous scanning, map the three-dimensional coordinates of the points, judge discrete points based on the continuity of spatial position and remove outliers, and generate stream contour point cloud coordinate set. S202: Based on the point cloud coordinate set of the stream contour, construct the point cloud coordinate distribution matrix, perform mean elimination processing on the matrix dimension, calculate the covariance matrix and extract the eigenvalue distribution, compare the size relationship of the eigenvalues, determine the principal component direction according to the direction corresponding to the principal eigenvalue, and generate the principal component direction discrimination vector set. S203: Based on the principal component direction discrimination vector set and the flow stream contour point cloud coordinate set, perform projection operation on the point cloud coordinates along the discrimination direction, calculate the spatial mean of the projected coordinates and aggregate the positions, perform plane segmentation and surface accumulation, and generate the centroid coordinates, principal axis vector and contour area of ​​the flow stream cross section.

5. The method for aluminum molten casting using a robotic arm based on multi-sensor fusion according to claim 4, characterized in that, The step of performing time-series alignment based on point cloud data obtained by continuous scanning and mapping the three-dimensional coordinates of the points refers to matching the timestamps of point cloud data obtained by two adjacent scans according to the fixed scanning frequency of the laser contour sensor, and completing the coordinate alignment under the condition that the time difference is not greater than the preset sampling period threshold. The method of judging discrete points and removing outliers based on the continuity of spatial location refers to calculating the average Euclidean distance between any point in the mapped point cloud data and a preset number of adjacent points. When the average Euclidean distance exceeds a preset spatial distance threshold, the corresponding point is judged as an outlier and removed.

6. The method for aluminum molten casting using a robotic arm based on multi-sensor fusion according to claim 4, characterized in that, The specific steps for S3 are as follows: S301: Call the centroid coordinates of the flow section and the center coordinates of the gate, perform a difference operation on the two sets of spatial coordinates in the corresponding coordinate axis directions, perform a direction consistency retrieval on the difference results and construct a vector component combination, perform amplitude merging and direction identification on the vector components, and generate a centroid spatial deviation vector. S302: Based on the centroid space deviation vector, call the principal axis vector of the flow section and the gate normal vector, match the direction components of the two vectors one by one, perform dot product operation on the matched components, and perform ratio conversion in combination with the vector magnitude. Based on the angle definition relationship, perform mapping processing to obtain the characteristic value of the included angle of the axis. S303: Based on the characteristic value of the included angle of the axis and the centroid spatial deviation vector, call the cross-sectional contour area of ​​the flow stream, perform a ratio calculation on the current cross-sectional area and the reference cross-sectional area, and classify and mark the ratio results according to the interval division rules to establish a set of flow stream state features.

7. The method for aluminum molten casting using a robotic arm based on multi-sensor fusion according to claim 6, characterized in that, The specific steps of S4 are as follows: S401: Extract the spatial deviation vector and the angle between the axis from the set of flow state features, perform fuzzy membership mapping on the spatial deviation vector components and the angle between the axis respectively, substitute the mapping results into the proportional term and differential term rule table to perform rule matching, and generate attitude deviation control quantity; S402: Based on the attitude deviation control quantity, perform deviation accumulation operation between the control quantity and the current state parameter in the attitude dimension to construct the integral parameter sequence, and map the proportional quantity, integral quantity, and differential quantity to the corresponding matrix row and column positions according to the attitude dimension to establish the end attitude correction matrix. S403: Call the end attitude correction matrix to obtain the cross-sectional shrinkage rate parameter, construct it as a change amplitude, perform the association mapping with the attitude correction component in the end attitude correction matrix, and perform interval determination and discrete encoding on the result, calculate the valve opening adjustment value, and generate the casting coordinated adjustment command.

8. The method for aluminum molten casting using a robotic arm based on multi-sensor fusion according to claim 7, characterized in that, The spatial deviation vector component and the angle between the axis are respectively subjected to fuzzy membership mapping, which means that the segment is divided into at least three consecutive membership segments according to a preset angle interval, and the membership degree of the membership segment changes in a monotonic linear mapping manner.

9. The method for aluminum molten casting using a robotic arm based on multi-sensor fusion according to claim 7, characterized in that, The specific steps of S5 are as follows: S501: Extract the end attitude correction matrix from the casting coordination adjustment command, calculate the end attitude angular velocity sequence based on the time derivative of adjacent attitude matrices, use the inverse kinematics algorithm of the robotic arm to convert the end angular velocity into joint angular velocity command, perform sign consistency check and interval determination, and generate joint angular velocity correction pose sequence. S502: Based on the joint angular velocity correction pose sequence, extract the valve opening adjustment value flow velocity vector parameter, convert the joint angular velocity into the end linear velocity through the kinematic forward solution, establish the spatial correspondence between the flow velocity vector and the end motion velocity, perform amplitude normalization mapping, and obtain the aluminum liquid flow coordinated adjustment configuration. S503: Based on the aluminum liquid flow rate coordinated adjustment configuration, call the attitude change component in the joint angular velocity correction pose sequence, perform state combination calculation of flow velocity vector parameter and attitude change component, perform interval judgment and state identifier mapping according to the state transition rules of the mold entry process, and establish a stable pouring state of aluminum liquid entering the mold.

10. The method for aluminum molten casting using a robotic arm based on multi-sensor fusion according to claim 9, characterized in that, During the execution of symbol consistency verification and interval determination, a predetermined threshold range is set for the joint angular velocity command. When the amplitude of the joint angular velocity changes within the threshold range, it is determined to be a valid adjustment range. When it exceeds the threshold range, the execution of the joint angular velocity correction command is paused and an exception handling mechanism is triggered.

Images