Electric drive rotator for agv fork truck and control method

Abstract

The application provides an electric drive rotator for AGV forklift and a control method, comprising a frame plate, a forklift assembly, a radar, a dynamic inclination sensor and a control system, the control system is configured to receive the cargo contour position data transmitted by the radar, calculate the target rotation amount required by the servo motor to align the forklift assembly to the cargo forklift position, and control the servo motor to drive the forklift assembly to rotate by a corresponding angle; at the same time, the actual rotation angle fed back by the dynamic inclination sensor is received during the rotation process, and the rotation angle is closed-loop corrected. The application realizes electric rotation control, is suitable for AGV forklift and other unmanned driving scenes, avoids the leakage problem of hydraulic components during operation from the source, realizes accurate rotation positioning of the forklift at any angle, and realizes automatic accurate alignment of the forklift through the cooperative work of radar active positioning and dynamic inclination sensor closed-loop control.

Description

Technical Field

[0001] This invention relates to the field of automated material handling technology, specifically to an electric drive rotary device and control method for AGV (Automated Guided Vehicle) forklifts. Background Technology

[0002] With the rapid development of automated logistics technology, AGV forklifts have been widely used in cargo handling and stacking operations in warehousing, manufacturing, and other fields. In special working conditions such as the smelting and waste disposal industries, higher operational requirements are placed on AGV forklifts: they need to be able to automatically pick up high-temperature, dusty, or toxic materials and precisely rotate the forks to dump the materials into designated containers or workstations. During these operations, the precision of the fork rotation angle directly affects the dumping effect, operational efficiency, and operational safety.

[0003] Currently, the vast majority of existing forklift swivels use hydraulic drive to transfer goods, and there are almost no electrically driven swivel products on the market. Hydraulic drive solutions have the following technical drawbacks: (1) Hydraulic leakage risk: Hydraulic drive systems inevitably have hydraulic oil leakage problems, which makes it impossible for rotary products to be used in working conditions that require no pollution risk, such as food processing, precision electronic manufacturing and other scenarios with high cleanliness requirements.

[0004] (2) Low rotation angle control accuracy: The rotation angle of the existing hydraulic rotator relies on the operator's eye to observe the approximate angle, and then adjusts it to the approximate specified position through the forklift control lever. It cannot achieve precise angle control, and it does not have the function of compensating for accumulated operating errors.

[0005] (3) High energy consumption and environmentally unfriendly: The hydraulic system needs to maintain oil pressure continuously, which consumes a lot of energy, and the use and maintenance of hydraulic oil also puts a burden on the environment.

[0006] (4) High risk of manual operation: In high-risk working conditions such as smelting and waste disposal, there is a great safety risk in manually operating the rotary device to dump goods.

[0007] (5) Unable to adapt to unmanned driving scenarios: Existing hydraulic rotary devices cannot achieve automatic control and are difficult to adapt to unmanned driving application scenarios such as AGV forklifts.

[0008] Therefore, there is an urgent need for an electric drive rotary device and control method that can achieve electric drive, precise angle control, automatic error compensation, and is suitable for unmanned driving scenarios of AGV forklifts. Summary of the Invention

[0009] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: An electrically driven rotator for AGV forklifts includes: The frame plate has a built-in speed reduction transmission mechanism for connecting the servo motor; The fork assembly, wherein the reduction gear mechanism is capable of driving the fork assembly to rotate the cargo; The radar, installed on the frame plate, is used to scan and measure the distance of the goods when the AGV forklift arrives at the target goods area, and obtain the outline position data of the goods. A dynamic tilt sensor, integrated into the fork assembly, is used to detect the actual rotation angle of the forks in real time during the rotation of the fork assembly; and The control system is electrically connected to the radar, the dynamic tilt sensor, and the servo driver that drives the servo motor. The control system is configured to: receive cargo contour position data transmitted by the radar; calculate the target rotation amount required for the servo motor to align the fork assembly with the cargo picking position; and control the servo motor to drive the fork assembly to rotate by the corresponding angle; and simultaneously receive the actual rotation angle fed back by the dynamic tilt sensor during the rotation process, and perform closed-loop correction on the rotation angle.

[0010] Preferably, the servo motor is mounted on the frame plate, and its output end is connected to the front base plate through the reduction transmission mechanism. The front base plate is used to mount the fork assembly, and the fork assembly can rotate relative to the frame plate under the drive of the servo motor.

[0011] Preferably, the front base plate is provided with an upper crossbeam and a lower crossbeam, which are used to fix the fork assembly.

[0012] Preferably, the control system performs non-uniform grid division on the scanning coverage area of ​​the radar, establishes a polar coordinate system with the radar location as the origin, divides the scanning coverage area into multiple scanning grid units, and the grid division density decreases with the increase of radial distance, with the grid being denser in areas closer to the radar; at the same time, the grid division density is higher in the preset main detection direction than in other directions, so that the near-range area and the main detection direction obtain higher spatial resolution.

[0013] Preferably, the control system calculates the scanning priority weight for each scanning grid cell. The scanning priority weight is determined based on the historical detection priority obtained by normalizing the number of times the grid cell detected the target in the historical scanning cycle, and the cumulative energy value of all echo signals in the grid cell. The higher the echo energy, the greater the scanning priority weight.

[0014] Preferably, the control system uses duty cycle modulation of the transmitted pulse train to allocate scanning resources. The duty cycle of the transmitted pulse of each scanning grid unit is determined according to the ratio of its scanning priority weight to the average priority weight of all grid units. Grid units with higher priority weights are allocated higher duty cycles, thereby realizing the automatic tilting of scanning resources towards high-value areas.

[0015] Preferably, the control system establishes a trajectory predictor for the tracked target and outputs the predicted position for the next scan time based on the target's historical position sequence; when the radar actually detects the target, it calculates the prediction deviation vector between the actual measured position and the predicted position of the target, and continuously adjusts the scan step size according to the magnitude of the prediction deviation vector.

[0016] Preferably, the control system performs joint time-frequency domain feature mapping processing on the radar echo data, calculates a short-time Fourier transform for each radar echo sampling point, and constructs a three-dimensional time-frequency feature tensor with the transformation result containing three dimensions: time domain dimension, frequency domain dimension, and number of channels. This feature tensor can be directly input into a convolutional neural network for target detection.

[0017] Preferably, the control system constructs a spatiotemporal graph for target temporal correlation. The nodes of the spatiotemporal graph represent targets detected at each time point. The node features include position coordinates, velocity components, and radar cross-section information, and the edges represent possible temporal correlations. A graph attention network is used to iteratively update the node features and pass messages. The importance between nodes is automatically learned through the attention mechanism, and finally the correlation confidence between targets is output.

[0018] Preferably, the control system models the association problem between radar measurements and target tracks as an assignment problem and solves it using an auction algorithm: in the bidding phase, each unassigned measurement calculates its bid based on its association cost with each track; in the allocation phase, each track selects the measurement with the highest bid as a temporary allocation and updates the track price, iterating until all measurements are allocated, with the bidding process for each measurement executed in parallel.

[0019] Preferably, the control system defines the adjustable parameters in the scanning control process as learnable parameter vectors, constructs a deep unfolded network, maps the iterative steps of the traditional scanning control algorithm to the layers of the network and unfolds them, performs inference through forward propagation and calculates the gradient of the loss function with respect to each parameter through backpropagation, thereby achieving adaptive optimization of the parameters; the optimized parameters are fed back to the scanning control loop to correct the prediction deviation threshold, scanning priority weight and scanning step size adjustment range.

[0020] Preferably, the control system is configured to: perform rasterization processing on the polar coordinate point cloud data obtained by the radar scan to extract the cargo outline boundary, calculate the free space asymmetry based on the relative relationship between the cargo outline boundary and the radar center direction, and determine the rotation direction and rotation step size of the servo motor accordingly.

[0021] Preferably, the dynamic tilt sensor and the control system form a closed-loop feedback loop, which is used to feed back the actual rotation angle of the forks to the control system in real time. The control system controls the servo motor to perform closed-loop correction of the rotation angle based on the deviation between the actual rotation angle and the target rotation angle.

[0022] The present invention also provides a method for controlling an electrically driven rotator for an AGV forklift, for use with an electrically driven rotator as described in any of the preceding claims, the control method comprising the following steps: S1: After the AGV forklift arrives at the target cargo area, the control system controls the radar to scan and measure the distance of the cargo to obtain the cargo outline position data. S2: The control system receives the cargo outline position data and calculates the target rotation amount required for the servo motor to align the forks with the cargo picking position. S3: The control system controls the servo motor to achieve the target rotation amount and drives the fork assembly to rotate by the corresponding angle; S4: During the rotation of the fork assembly, the dynamic tilt sensor detects the actual rotation angle of the fork assembly in real time and feeds it back to the control system. The control system performs closed-loop correction of the rotation process based on the actual rotation angle.

[0023] Preferably, in step S1, the control system controls the scanning process of the radar as follows: A polar coordinate system is established with the radar location as the origin. The scanning coverage area is divided into non-uniform grids, with the grid density decreasing as the radial distance increases. The grid density is higher in the preset main detection direction than in other directions. The scanning priority weight is calculated based on the historical detection priority and echo signal energy of each grid cell. The duty cycle of the transmitted pulse train is used to tilt the scanning resources towards high-priority grid cells. At the same time, the trajectory of the tracked target is predicted, and the scanning step size is continuously adjusted according to the magnitude of the prediction deviation vector to achieve dynamic adaptive adjustment of the scanning field of view.

[0024] Preferably, during the radar's scanning and ranging process, radar echo data reflected from the cargo and the environment are received. The control system processes the radar echo data to obtain precise location information of the cargo's outline, specifically including: The radar echo data is subjected to short-time Fourier transform and a three-dimensional time-frequency feature tensor is constructed. This tensor is then input into a convolutional neural network for target detection. A spatiotemporal graph is constructed and a graph attention network is used to perform temporal correlation of the target, outputting the target correlation confidence. The correlation between measurement and trajectory is modeled as an assignment problem and an auction algorithm is used for global matching to obtain the precise location information of the cargo outline.

[0025] Preferably, the control method further includes a parameter adaptive optimization step: The energy sensitivity coefficient, duty cycle modulation index, and gain coefficient in the scanning control process, as well as the attention network parameters and auction algorithm perturbation terms in the echo processing process, are defined as learnable parameter vectors. A deep unfolding network is constructed to perform adaptive parameter optimization. The optimized parameters are fed back to the scanning control strategy and the echo data processing to correct the prediction deviation threshold, scanning priority weight, scanning step size adjustment magnitude, and time sequence correlation weight, forming a deep collaborative closed loop of perception and processing.

[0026] Preferably, calculating the target rotation amount of the servo motor in step S2 specifically includes the following steps: The polar coordinate point cloud obtained by radar scanning is divided into multiple continuous grids at fixed angular intervals, and the minimum distance value in each grid is recorded. Scan the grid sequence from left to right, calculate the distance difference between adjacent grids, and record the left and right boundary angles of the cargo outline when the distance difference exceeds a preset threshold. The free space asymmetry is calculated based on the relative relationship between the left and right boundary angles and the radar center direction. The free space asymmetry is compared with preset upper and lower thresholds to determine the rotation direction and rotation step size of the servo motor. The servo motor is controlled to rotate at a constant speed within a set time according to the rotation step generation timing drive signal.

[0027] Preferably, the closed-loop correction in step S5 specifically includes: The dynamic tilt sensor continuously detects the actual rotation angle of the forks throughout the entire rotation process of the fork assembly and provides real-time feedback. The control system compares the actual rotation angle with the target rotation angle. When there is a deviation, it calculates the compensation rotation amount and controls the servo motor to make fine adjustments until the actual rotation angle matches the target rotation angle.

[0028] As can be seen from the above technical solutions, the present invention has the following beneficial effects: (1) The present invention uses electric drive instead of hydraulic drive, which avoids leakage problems in the operation of hydraulic components from the source, so that the rotary product can work under working conditions that require no pollution risk, making it more green and environmentally friendly; the electric drive rotary operates smoothly and can save about 25% energy compared with hydraulic drive, and the load adjustment is reliable. (2) This invention achieves precise rotation positioning of the forks at any angle through the collaborative working mode of radar active positioning and dynamic tilt sensor closed-loop control. At the same time, the control system has the ability to compensate for accumulated operating errors, which effectively improves the control accuracy of the rotation angle. (3) The present invention overcomes the problems of difficult visibility and high risk of manual rotation when handling goods by traditional forklifts through the collaborative working mechanism of radar scanning, control system calculation and servo motor execution. It has high positioning accuracy and fast adjustment response speed. (4) This invention is suitable for unmanned driving scenarios such as AGV forklifts in the dark, and can realize automatic picking, stacking, transfer and dumping operations, which greatly improves the industry's production efficiency; (5) The present invention employs a series of advanced signal processing and intelligent control algorithms, such as non-uniform grid division, duty cycle modulation, prediction deviation dynamic adjustment, time-frequency domain joint feature mapping, graph neural network temporal correlation, auction algorithm global matching, and deep unfolding network parameter adaptation, so that radar data processing and scanning control are deeply coordinated to form an intelligent scheduling closed loop, which significantly improves the accuracy and real-time performance of cargo contour recognition. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the overall structure of the electrically driven rotator according to an embodiment of the present invention. Figure 1 ; Figure 2 This is a schematic diagram of the overall structure of the electrically driven rotator according to an embodiment of the present invention. Figure 2 ; Figure 3 This is a side view of the electrically driven rotator according to an embodiment of the present invention; Figure 4 for Figure 3 AA section view; Figure 5 This is a block diagram illustrating the control principle of the servo driver in an embodiment of the present invention. Figure 6 This is a schematic flowchart of an electrically driven rotator control method provided in an embodiment of the present invention.

[0030] In the diagram: 1. Frame plate; 2. Servo motor; 3. Reduction transmission mechanism; 4. Front bottom plate; 5. Fork assembly; 6. Radar; 7. Dynamic tilt sensor; 8. Control system; 9. Servo driver; 10. Upper crossbeam; 11. Lower crossbeam; 12. Rear bottom plate; 13. Slewing bearing assembly. Detailed Implementation

[0031] A preferred embodiment of the present invention will now be described in detail with reference to the accompanying drawings.

[0032] like Figure 1-4 As shown, this embodiment provides an electric drive rotator for AGV forklifts, including a frame plate 1, a servo motor 2, a reduction transmission mechanism 3, a front bottom plate 4, a fork assembly 5, a radar 6, a dynamic tilt sensor 7, a control system 8, and a servo driver 9.

[0033] The frame plate 1 serves as the main support structure for the rotator and houses a reduction gear transmission mechanism 3 for connecting the servo motor 2. The servo motor 2 is mounted on the frame plate 1, and its output end is connected to the front base plate 4 via the reduction gear transmission mechanism 3. The front base plate 4 is equipped with an upper crossbeam 10 and a lower crossbeam 11, which are used to fix the fork assembly 5. The fork assembly 5 can rotate relative to the frame plate 1 under the drive of the servo motor 2, along with the front base plate 4.

[0034] Radar 6 is mounted on the frame plate 1 and is used to scan and measure the distance of the goods when the AGV forklift arrives at the target goods area to obtain the goods' outline position data. Dynamic tilt sensor 7 is integrated into the fork assembly 5 and is used to detect the actual rotation angle of the forks in real time during the rotation of the fork assembly 5. In this embodiment, radar 6 is mounted on the upper end of the frame plate 1, and dynamic tilt sensor 7 is mounted on the front end of the forks.

[0035] The control system 8 is electrically connected to the radar 6, the dynamic tilt sensor 7, and the servo driver 9, respectively. The servo driver 9 is used to drive the servo motor 2, such as... Figure 5 As shown. The control system 8 and servo driver 9 are built into the protective box and mounted on the frame plate via the rear base plate 12.

[0036] The control system 8 is configured to: receive cargo contour position data transmitted by radar 6, calculate the target rotation amount required for servo motor 2 to align fork assembly 5 with cargo picking position, and control servo motor 2 to drive fork assembly 5 to rotate by the corresponding angle; at the same time, receive the actual rotation angle fed back by dynamic tilt sensor 7 during the rotation process, and perform closed-loop correction on the rotation angle.

[0037] Working principle: The electric drive rotator of this invention is powered by a 48V DC power supply. The control system 8 controls the operation of the electric reducer assembly (i.e., the combination of servo motor 2 and reduction transmission mechanism 3) through a PLC program. The reduction transmission mechanism is connected to the slewing bearing assembly 13 through its internal gears. The slewing bearing assembly 13 is connected to the front base plate 4. The upper and lower crossbeams are bolted to the front base plate 4. The fork assembly 5 is externally mounted on the upper and lower crossbeams. The control system 8 realizes the functions of rotating the forks at any angle and resetting the operation, so as to realize the cargo handling function of the electric drive rotator under high standard environmental conditions.

[0038] The reduction transmission mechanism 3 includes a worm gear pair and a gear pair, where the transmission ratio between the worm and the worm gear is 30:1, and the transmission ratio between the gear and the slewing bearing assembly 13 is 11:98, for a total transmission ratio of 267.2. That is, the output of the servo motor 2 undergoes a first-stage reduction through the worm gear pair, and then drives the slewing bearing assembly 13 through the gear pair, thus achieving the rotation of the fork assembly 5. When calculating the target rotation amount of the servo motor 2, the control system 8 needs to multiply the target rotation angle required for the fork assembly 5 by the total transmission ratio of 267.2 to obtain the actual rotation angle required by the servo motor 2.

[0039] like Figure 6 As shown, this embodiment describes in detail the control method of the electrically driven rotator, which includes the following steps: S1: After the AGV forklift arrives at the target cargo area, the control system controls the radar to scan and measure the distance of the cargo to obtain the cargo outline position data.

[0040] After the AGV forklift reaches the approximate area of ​​the target goods via the navigation and positioning system, the control system 8 activates radar 6 to scan and measure the distance to the goods and the surrounding environment. The control system 8 performs a series of intelligent controls on the scanning process of radar 6, including non-uniform division and resource allocation of the scanning coverage area, dynamic adjustment of the scanning field of view based on prediction bias, and time-frequency domain joint feature mapping, graph neural network temporal correlation, auction algorithm global matching, and adaptive optimization of depth unfolding network parameters on the radar echo data, ultimately obtaining the precise location information of the goods outline (polar coordinate point cloud data).

[0041] This embodiment describes in detail the complete process of the scanning control of the radar 6 and the processing of echo data by the control system 8 in step S1.

[0042] S11: Non-uniform division and resource allocation of the scan coverage area The control system 8 establishes a polar coordinate system with the location of radar 6 as the origin, and divides the scanning coverage area into non-uniform grids. The grid density decreases as the radial distance increases, and the grid is denser in areas closer to the radar. At the same time, the grid density is higher in the preset main detection direction (i.e. the direction in which the forks usually face), so that the near-field area and the main detection direction can obtain higher spatial resolution.

[0043] The mesh density function is determined using formula (1): (1) in, This is the radial distance, specifically the straight-line distance between the target point and the radar. The scanning angle is specifically the deflection angle of the target point relative to the radar normal (or central axis). The center focusing angle is specifically the center angle value corresponding to the location with the highest mesh density. The base grid density is specifically the lowest grid density (reference value) within the entire scan area. The encryption amplitude coefficient is used to control the maximum increase in density of hotspot areas. α≥0. When α=3, the density of the hotspot center is 4 times the basic density (i.e., 1+α). This is the distance attenuation coefficient, which specifically controls how quickly the mesh refinement effect decreases with distance. The larger the value, the more concentrated the encryption is in the vicinity; The smaller the value, the farther the encryption can extend; γ is the angular concentration coefficient, specifically controlling the degree of angular concentration of the mesh encryption. The larger γ is, the more concentrated the encryption area is. The closer the beam is to the edge (the narrower the beam), the smaller the γ value, and the wider the encryption angle range (the wider the beam).

[0044] Based on this, the control system 8 calculates the scan priority weight for each scan grid cell. The scan priority weight w_ij for each scan grid cell (i,j) is calculated using formula (2): (2) in, The historical weight is obtained by normalizing the number of times the target was detected in the past N scan cycles for this grid cell; This refers to the local energy, specifically the feature intensity within the grid (i,j). γ is the global average energy, specifically the average energy of all grid cells; γ is the weighted gain coefficient, specifically controlling the gain sensitivity of energy to the weights. This is the upper limit of the weight, specifically a cutoff value to prevent excessive weight growth.

[0045] Furthermore, the control system 8 uses duty cycle modulation of the transmitted pulse train to achieve scanning resource allocation. The duty cycle of the transmitted pulse for the (i,j)th grid cell is... Determine according to formula (3): (3) in, w_avg is the baseline grid size, specifically the standard grid size under the average weight; w_avg is the global average weight, specifically the average of all weights. The current grid weight is calculated using a weighting formula. The scaling index controls the strength of the weights' influence on the size. This is the minimum grid size, specifically the finest grid limit (encryption limit). This represents the maximum grid size, specifically the coarsest grid constraint (sparse lower bound).

[0046] By using non-uniform grid division and duty cycle modulation, scanning resources are automatically tilted towards high-value areas without the need for mechanical speed adjustment, which is a form of soft resource reallocation.

[0047] S12: Dynamic adjustment of the scanning horizon based on prediction bias The control system 8 establishes a trajectory predictor for the tracked target and outputs the target state prediction position vector for the next scan time based on the target's historical position sequence. Once the radar actually detects the target, it records the target's measured position vector. Calculate the prediction deviation vector e according to formula (4): (4) The magnitude of the prediction deviation vector, ‖d‖, reflects the degree of deviation between the actual target motion and the prediction model. Based on the magnitude of the prediction deviation, the adaptive scanning step size is continuously adjusted according to formula (5). : (5) in, η is the reference scan step size; η is the step size adjustment gain; d represents the error tolerance threshold used for normalization; d is the prediction deviation modulus. The adjustment logic is as follows: the larger ‖d‖ is (the stronger the target's mobility), the larger the scan step size, expanding the coverage area to prevent the target from being lost; the smaller ‖d‖ is (the target's movement is stable), the smaller the scan step size, performing a fine scan around the predicted position. This method changes the scanning field of view from discrete jumps to continuous adjustment, avoiding scanning blind spots caused by step size switching.

[0048] S13: Joint Time-Frequency Domain Feature Mapping of Radar Echo Data During the scanning and ranging process of the radar 6, radar echo data reflected from the cargo and the environment are received. The control system 8 processes the radar echo data to obtain precise position information of the cargo outline.

[0049] First, the short-time Fourier transform (STFT) is calculated for each radar echo sampling point, as shown in formula (6): (6) Where x[n] is the discrete sampling sequence of radar echo; w[n] is the window function; k is the time frame index (center position of the window function); ω is the angular frequency (continuous frequency variable); This is the Fourier transform kernel function.

[0050] The STFT results are constructed as a three-dimensional time-frequency feature tensor. Its dimensions are N_t×N_f×N_c, where N_t is the time domain dimension, N_f is the frequency domain dimension, and N_c is the number of channels. Tensor elements are represented by formula (7): (7) In the formula: i is the time domain index; j is the frequency domain index; k is the channel index.

[0051] This three-dimensional data cube can be directly input into a convolutional neural network for target detection, transforming the one-dimensional time-domain signal of radar echoes into a two-dimensional time-frequency representation. The CNN then automatically learns the time-frequency features of the target, replacing the traditional manual rule design method.

[0052] S14: Target Temporal Correlation Based on Graph Neural Network Control System 8 Constructs Spatiotemporal Graph G = (V, ) Perform target temporal correlation, where the node set V represents the targets detected at each time step, and the node features include position coordinates, velocity components, and radar cross-section (RCS) information, and the edge set Indicates possible temporal correlations.

[0053] A graph attention network (GAT) is used to iteratively update node features and pass messages. The attention mechanism is calculated according to formula (8): (8) in, The attention coefficient is calculated using a learnable mechanism; W is a learnable weight matrix. For activation functions; Let v be the set of neighboring nodes of node v. Through iterative message passing in the graph network, the final output edge features are mapped to a correlation confidence level between 0 and 1 using the sigmoid function. If the confidence level exceeds a preset threshold, the association is considered successful.

[0054] S15: Global Association Matching Based on Auction Algorithm The control system 8 models the association between measurement and trajectory as an assignment problem, with the objective function shown in formula (9): (9) Constraints: Σ_ix_ij=1, each track can be assigned at most one measurement; Σ_jx_ij=1, each measurement can be assigned at most one track; ∈{0,1}, which are binary decision variables.

[0055] An auction algorithm is used to solve the assignment problem. During the bidding phase, each unassigned measurement i calculates its bid for track j according to formula (10): (10) in, v_j is the bid of unassigned measurement i for track j; v_j is the current value of track j; c_ij is the association cost between measurement i and track j; ε is a minimal random perturbation constant used to break the tie of equal bids and avoid iteration stagnation.

[0056] During the allocation phase, each track j selects the measurement with the highest bid as a temporary allocation and updates its price according to formula (11): (11) Auction algorithms have good parallelization potential, and the bidding process for each metric can be executed simultaneously on parallel computing hardware such as GPUs, making them particularly suitable for large-scale correlation problems.

[0057] S16: Parameter Adaptation Based on Deep Unfolded Networks The control system 8 also defines the weighted gain coefficient γ, scaling exponent β_D, step size adjustment gain η in the scanning control process, as well as the attention network parameters and the disturbance term ε of the auction algorithm in the echo processing process as learnable parameter vectors, and constructs a deep unfolding network for parameter adaptive optimization.

[0058] The loss function is shown in equation (12): (12) in, Estimate the target position in frame t; The target's true location in frame t (from the labeled data); λ represents the number of false alarms in frame t; λ is the balance coefficient, with a value ranging from 0.1 to 0.5.

[0059] The learnable parameters are updated according to formula (13) using backpropagation and gradient descent algorithms: (13) Where μ is the learning rate; This represents the gradient of the loss function with respect to the parameter θ. The algorithm can automatically adjust the parameters according to the actual data distribution, adapting to different scenarios without the need for manual parameter tuning.

[0060] S17: Feedback Closed-Loop Control The adaptively updated parameters from step 3.6 are fed back to steps 3.1 to 3.5, forming a complete "perception-decision-control" feedback loop. Specifically: the correlation confidence and tracking accuracy indicators output from step 3.6 are used to correct the prediction deviation threshold d_th in step 3.2; the optimized weighted gain coefficient γ from step 3.6 is used to update the scan priority weight calculation in step 3.1; and the optimized step size adjustment gain η from step 3.6 is used to update the scan step size adjustment amplitude in step 3.2. This achieves deep collaboration between radar data processing and scan control, forming an intelligent scheduling loop, and ultimately outputting the precise location information of the cargo outline (polar coordinate point cloud data) as input for step S2.

[0061] S2: The control system receives the cargo outline position data and calculates the target rotation amount required for the servo motor to align the forks with the cargo picking position.

[0062] The control system 8 receives the cargo contour position data (polar coordinate point cloud) obtained in the first step, performs rasterization processing on the point cloud data, extracts the cargo contour boundary, calculates the free space asymmetry based on the relative relationship between the cargo contour boundary and the radar center direction, and determines the rotation direction and rotation step size of the servo motor 2 accordingly.

[0063] like Figure 4 As shown, this embodiment describes in detail the specific process by which the control system 8 calculates the target rotation amount of the servo motor 2 based on the cargo outline position data obtained by the radar 6 in step S2 (corresponding to claim 18).

[0064] In this embodiment, the fork rotation alignment before picking up the pallet is no longer a one-time calculation of the target angle, but a visual servo closed-loop process based on real-time radar point clouds. This process iterates continuously through the following sub-steps, directly utilizing the asymmetry of the pallet contour to drive the fork movement until the forks are precisely parallel to the fork holes.

[0065] S21: Rasterization Processing The polar coordinate point cloud obtained by radar scanning (i.e., the cargo outline position data output in step S1) is divided into N continuous grids at fixed angular intervals Δθ. Each grid i records the minimum distance value R_min(i) within that angular interval.

[0066] S22: Boundary Extraction Scan the raster sequence from left to right and calculate the distance difference between adjacent raster cells. .when Exceeding the preset threshold Record the left boundary angle. and right boundary angle These correspond to the grid angle values ​​before and after the distance jump.

[0067] S23: Asymmetry Calculation Obtain the radar angle corresponding to the current orientation of the servo motor. (i.e., radar zero-degree direction), calculate the free space asymmetry D:

[0068] If the denominator is less than or equal to 0, then let D = 1.

[0069] S24: Determine the rotation step size Compare D with the preset upper threshold. Lower threshold D_L comparison: If D> Let the direction of rotation be positive, and the step size S = S0 × (D-1); If D < Let the rotation direction be negative, and the step size be S = S0 × (1 - D); Otherwise, step size S=0.

[0070] S0 is the basic step size (unit: degrees).

[0071] S25: Timing output generation and motor drive The timing drive signal is generated based on the rotation step size S: an analog voltage signal with a fixed amplitude V_fix is ​​generated, and a timer with a timing period of T=k×|S| is started at the same time, where k is a pre-calibrated time-angle coefficient (unit: seconds / degree).

[0072] The analog voltage signal and the timer enable signal are input together to the servo driver 9 operating in speed mode, so that the servo motor 2 rotates at a constant speed within time T, thereby realizing the angle change corresponding to the step size S.

[0073] Since the total transmission ratio of the reduction gear mechanism 3 is 267.2, the control system 8 multiplies the target step size S by the total transmission ratio 267.2 when calculating the final rotation angle to obtain the actual rotation angle required by the servo motor 2. That is:

[0074] Where θ_motor is the actual rotation angle of servo motor 2, and S is the angle that the fork assembly 5 needs to rotate.

[0075] The above sub-steps S21 to S25 are continuously iterated until the free space asymmetry D falls within the range of [D_L, D_H] (i.e., S=0), indicating that the forks have been precisely aligned with the position of the goods being picked up. At this point, the rotation is stopped, and the alignment operation before picking up the goods is completed.

[0076] S3: The control system controls the servo motor to achieve the target rotation amount and drives the fork assembly to rotate by the corresponding angle.

[0077] The control system 8 sends the target rotation amount calculated in the second step to the servo driver 9, which drives the servo motor 2 to rotate according to the target rotation amount. The output end of the servo motor 2 drives the front base plate 4 and the fork assembly 5 to rotate to the target angle through the reduction transmission mechanism 3 (total transmission ratio 267.2), so that the forks are aligned with the cargo picking position.

[0078] S4: During the rotation of the fork assembly, the dynamic tilt sensor detects the actual rotation angle of the fork assembly in real time and feeds it back to the control system. The control system performs closed-loop correction of the rotation process based on the actual rotation angle.

[0079] Throughout the rotation of the fork assembly 5, the dynamic tilt sensor 7 continuously detects the actual rotation angle of the forks and feeds it back to the control system 8. The control system 8 compares the actual rotation angle with the target rotation angle. When a deviation exists, it calculates the compensation rotation amount and controls the servo motor 2 to make fine adjustments until the actual rotation angle matches the target rotation angle.

[0080] The control system 8 sends the target rotation amount (the motor rotation angle θ_motor converted by the transmission ratio 267.2) calculated in step S2 to the servo driver 9. The servo driver 9 drives the servo motor 2 to rotate according to the target rotation amount. The output end of the servo motor 2 drives the front base plate 4 and the fork assembly 5 to rotate through the reduction transmission mechanism 3, so that the forks are aligned with the cargo picking position.

[0081] Throughout the rotation of the fork assembly 5, the dynamic tilt sensor 7 continuously detects the actual rotation angle θ_actual of the forks and feeds it back to the control system 8 in real time.

[0082] The control system 8 compares the actual rotation angle θ_actual with the target rotation angle θ_target, and performs closed-loop correction as follows when a deviation exists: First, calculate the angle deviation. .

[0083] If | | Greater than the preset allowable error threshold Then the control system 8 calculates the compensation rotation amount:

[0084] in, : Output angle compensation amount; : Proportional control coefficient; Integral control coefficient; Differential control coefficient :Angle deviation; : The integral term of the angular deviation; / dt: The differential term of the angular deviation.

[0085] Control system 8 based on compensation rotation amount The servo motor 2 is controlled to perform fine-tuning rotation until the deviation between the actual rotation angle and the target rotation angle is less than the allowable error threshold, thereby achieving precise closed-loop control of the fork rotation angle.

[0086] The control method further includes: dumping operation control. After the fork assembly 5 picks up the goods, the control system 8 executes the tilting operation control according to the preset tilting area position and tilting command.

[0087] The specific process is as follows: (1) The control system 8 calculates the target rotation angle θ_dump required for the fork assembly 5 to rotate from the current fork pick position to the tilt position; (2) The control system 8 controls the servo motor 2 to drive the forks and goods to rotate to the tilting position; (3) During the tilting process, the dynamic tilt sensor 7 monitors the tilt angle of the forks in real time; (4) When the fork tilt angle reaches the preset tilt angle At that time, control system 8 controls the forks to stop rotating; (5) After the forks maintain the tilting angle for a certain period of time, the goods will automatically tilt under the action of gravity; (6) After the tipping is completed, the control system 8 controls the forks to automatically reset to the horizontal position, waiting for the next forklift operation.

[0088] By using pre-set tilting areas and related commands, along with radar 6, dynamic tilt sensor 7, and control system 8, this embodiment achieves automatic picking, placing, and tilting operations of the electric rotator. This process overcomes the problems of limited visibility during traditional forklift handling and the high risk of manual rotation. It features high positioning accuracy, fast adjustment response, and a compact and reliable structure, greatly simplifying the picking, stacking, and transfer of goods in industries such as smelting and waste processing.

[0089] The above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. An electrically driven rotary device for AGV forklifts, characterized in that, include: The frame plate has a built-in speed reduction transmission mechanism for connecting the servo motor; The fork assembly, wherein the reduction gear mechanism is capable of driving the fork assembly to rotate the cargo; The radar, mounted on the frame plate, is used to scan and measure the distance of the goods when the AGV forklift arrives at the target goods area, and to obtain the outline position data of the goods. A dynamic tilt sensor, integrated on the fork assembly, is used to detect the actual rotation angle of the forks in real time during the rotation of the fork assembly; as well as The control system is electrically connected to the radar, the dynamic tilt sensor, and the servo driver that drives the servo motor. The control system is configured to: receive cargo contour position data transmitted by the radar; calculate the target rotation amount required for the servo motor to align the fork assembly with the cargo picking position; and control the servo motor to drive the fork assembly to rotate by the corresponding angle; and simultaneously receive the actual rotation angle fed back by the dynamic tilt sensor during the rotation process, and perform closed-loop correction on the rotation angle.

2. The electrically driven rotary device according to claim 1, characterized in that, The servo motor is mounted on the frame plate, and its output end is connected to the front base plate through the reduction transmission mechanism. The front base plate is used to mount the fork assembly, and the fork assembly can rotate relative to the frame plate under the drive of the servo motor.

3. The electrically driven rotary device according to claim 2, characterized in that, The front base plate is provided with an upper crossbeam and a lower crossbeam, which are used to fix the fork assembly.

4. The electrically driven rotary device according to claim 1, characterized in that, The control system performs non-uniform grid division on the scanning coverage area of ​​the radar. A polar coordinate system is established with the radar location as the origin, and the scanning coverage area is divided into multiple scanning grid units. The grid division density decreases with the increase of radial distance, and the grid is denser in areas closer to the radar. At the same time, the grid division density is higher in the preset main detection direction than in other directions, so that the near-range area and the main detection direction can obtain higher spatial resolution.

5. The electrically driven rotary device according to claim 1, characterized in that, The control system calculates the scanning priority weight for each scanning grid cell. The scanning priority weight is determined based on the historical detection priority obtained by normalizing the number of times the target was detected in the grid cell during the historical scanning cycle, and the cumulative energy value of all echo signals in the grid cell. The higher the echo energy, the greater the scanning priority weight.

6. The electrically driven rotary device according to claim 4 or 5, characterized in that, The control system uses duty cycle modulation of the transmitted pulse train to allocate scanning resources. The duty cycle of the transmitted pulse of each scanning grid unit is determined according to the ratio of its scanning priority weight to the average priority weight of all grid units. Grid units with higher priority weights are allocated higher duty cycles, thereby automatically tilting scanning resources towards high-value areas.

7. The electrically driven rotary device according to claim 1, characterized in that, The control system establishes a trajectory predictor for the tracked target and outputs the predicted position for the next scan time based on the target's historical position sequence. When the radar actually detects the target, it calculates the prediction deviation vector between the actual measured position and the predicted position of the target and continuously adjusts the scan step size according to the magnitude of the prediction deviation vector.

8. The electrically driven rotary device according to claim 1, characterized in that, The control system performs joint time-frequency domain feature mapping processing on the radar echo data, calculates a short-time Fourier transform for each radar echo sampling point, and constructs a three-dimensional time-frequency feature tensor with the transformation result containing three dimensions: time domain dimension, frequency domain dimension, and number of channels. This feature tensor can be directly input into a convolutional neural network for target detection.

9. The electrically driven rotator according to claim 1, characterized in that, The control system constructs a spatiotemporal graph to perform target temporal correlation. The nodes of the spatiotemporal graph represent the targets detected at each time moment. The node features include position coordinates, velocity components, and radar cross-section information, and the edges represent possible temporal correlations. A graph attention network is used to iteratively update the node features and pass messages. The importance between nodes is automatically learned through the attention mechanism, and finally the correlation confidence between targets is output.

10. The electrically driven rotator according to claim 1, characterized in that, The control system models the correlation between radar measurements and target trajectories as an assignment problem, and solves it using an auction algorithm, specifically: During the bidding phase, each unallocated measurement calculates its bid based on its associated cost with each track; during the allocation phase, each track selects the measurement with the highest bid as a temporary allocation and updates the track price, iterating until all measurements are allocated, with the bidding process for each measurement executed in parallel.

11. The electrically driven rotator according to claim 1, characterized in that, The control system defines the adjustable parameters in the scanning control process as learnable parameter vectors, constructs a deep unfolded network, and maps the iterative steps of the traditional scanning control algorithm to the layers of the network and unfolds them. It performs inference through forward propagation and calculates the gradient of the loss function with respect to each parameter through backpropagation to achieve adaptive optimization of the parameters. The optimized parameters are fed back to the scanning control loop to correct the prediction deviation threshold, scanning priority weight, and scanning step size adjustment range.

12. The electrically driven rotary device according to claim 1, characterized in that, The control system is configured to: perform rasterization processing on the polar coordinate point cloud data obtained by the radar scan to extract the cargo outline boundary, calculate the free space asymmetry based on the relative relationship between the cargo outline boundary and the radar center direction, and determine the rotation direction and rotation step size of the servo motor accordingly.

13. The electrically driven rotary device according to claim 1, characterized in that, The dynamic tilt sensor and the control system form a closed-loop feedback loop, which is used to feed back the actual rotation angle of the forks to the control system in real time. The control system controls the servo motor to perform closed-loop correction of the rotation angle based on the deviation between the actual rotation angle and the target rotation angle.

14. A method for controlling an electrically driven rotator for an AGV forklift, used in the electrically driven rotator as described in any one of claims 1-13, characterized in that, The control method includes the following steps: S1: After the AGV forklift arrives at the target cargo area, the control system controls the radar to scan and measure the distance of the cargo to obtain the cargo outline position data. S2: The control system receives the cargo outline position data and calculates the target rotation amount required for the servo motor to align the forks with the cargo picking position. S3: The control system controls the servo motor to achieve the target rotation amount and drives the fork assembly to rotate by the corresponding angle; S4: During the rotation of the fork assembly, the dynamic tilt sensor detects the actual rotation angle of the fork assembly in real time and feeds it back to the control system. The control system performs closed-loop correction of the rotation process based on the actual rotation angle.

15. The electric drive rotator control method according to claim 14, characterized in that, In step S1, the control system controls the scanning process of the radar as follows: A polar coordinate system is established with the radar location as the origin. The scanning coverage area is divided into non-uniform grids, with the grid density decreasing as the radial distance increases. The grid density is higher in the preset main detection direction than in other directions. The scanning priority weight is calculated based on the historical detection priority and echo signal energy of each grid cell. The duty cycle of the transmitted pulse train is used to tilt the scanning resources towards high-priority grid cells. At the same time, the trajectory of the tracked target is predicted, and the scanning step size is continuously adjusted according to the magnitude of the prediction deviation vector to achieve dynamic adaptive adjustment of the scanning field of view.

16. The electric drive rotator control method according to claim 15, characterized in that, During the radar's scanning and ranging process, radar echo data reflected from the cargo and the environment are received. The control system processes the radar echo data to obtain precise location information of the cargo's outline, specifically including: The radar echo data is subjected to short-time Fourier transform and a three-dimensional time-frequency feature tensor is constructed. This tensor is then input into a convolutional neural network for target detection. A spatiotemporal graph is constructed and a graph attention network is used to perform temporal correlation of the target, outputting the target correlation confidence. The correlation between measurement and trajectory is modeled as an assignment problem and an auction algorithm is used for global matching to obtain the precise location information of the cargo outline.

17. The electric drive rotator control method according to claim 15 or 16, characterized in that, The control method further includes a parameter adaptive optimization step: The energy sensitivity coefficient, duty cycle modulation index, and gain coefficient in the scanning control process, as well as the attention network parameters and auction algorithm perturbation terms in the echo processing process, are defined as learnable parameter vectors. A deep unfolding network is constructed to perform adaptive parameter optimization. The optimized parameters are fed back to the scanning control strategy and the echo data processing to correct the prediction deviation threshold, scanning priority weight, scanning step size adjustment magnitude, and time sequence correlation weight, forming a deep collaborative closed loop of perception and processing.

18. The electric drive rotator control method according to claim 14, characterized in that, Step S2 involves calculating the target rotation of the servo motor, specifically including the following steps: The polar coordinate point cloud obtained by radar scanning is divided into multiple continuous grids at fixed angular intervals, and the minimum distance value in each grid is recorded. Scan the grid sequence from left to right, calculate the distance difference between adjacent grids, and record the left and right boundary angles of the cargo outline when the distance difference exceeds a preset threshold. The free space asymmetry is calculated based on the relative relationship between the left and right boundary angles and the radar center direction. The free space asymmetry is compared with preset upper and lower thresholds to determine the rotation direction and rotation step size of the servo motor. The servo motor is controlled to rotate at a constant speed within a set time period based on the rotation step size timing drive signal.

19. The electric drive rotator control method according to claim 14, characterized in that, The closed-loop correction described in step S5 specifically includes: The dynamic tilt sensor continuously detects the actual rotation angle of the forks throughout the entire rotation process of the fork assembly and provides real-time feedback. The control system compares the actual rotation angle with the target rotation angle. When there is a deviation, it calculates the compensation rotation amount and controls the servo motor to make fine adjustments until the actual rotation angle matches the target rotation angle.

20. The electric drive rotator control method according to claim 14, characterized in that, The control method further includes: Tilting operation steps: After the fork assembly picks up the goods, the control system calculates the target rotation angle required for the fork assembly to rotate from the picking position to the tilting position according to the preset tilting area position and tilting command, and controls the servo motor to drive the forks and goods to rotate to the tilting position; during the tilting process, the dynamic tilt angle sensor monitors the tilt angle of the forks in real time, and controls the forks to stop rotating when the preset tilting angle is reached, thus completing the automatic tilting operation of the goods.

Images