Intelligent quay crane scaling experiment platform with body and control method

Abstract

The invention relates to the technical field of quay cranes, and provides an intelligent quay crane scaling experiment platform with a body and a control method. The intelligent quay crane scaling experiment platform with the body comprises a hoisting mechanism, a trolley mechanism, a pitching mechanism, a cart mechanism and a control system, and the control system is connected with the hoisting mechanism, the trolley mechanism, the pitching mechanism and the cart mechanism. Load loading and unloading are completed through longitudinal movement of the cart mechanism, arm support angle adjustment of the pitching mechanism, transverse movement of the trolley mechanism and vertical lifting of the hoisting mechanism. The pitching mechanism is provided with a lifting mechanism pulley, a lifting hook is lifted by the lifting mechanism pulley through a rope, the lifting hook is connected with a lifting appliance, and the lifting appliance is used for lifting a load; a power device, a vertical rotating mechanism and a horizontal rotating mechanism are arranged on the lifting appliance, and the power device provides active position regulation and control thrust for the lifting appliance; the vertical rotating mechanism is used for driving the power device to rotate in the vertical direction, and the horizontal rotating mechanism is used for driving the power device to rotate in the horizontal direction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of shore cranes, in particular to a shore crane with a body intelligent scale experiment platform and a control method. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] As an important cargo handling equipment, cranes play an important role in the fields of logistics ports, large engineering infrastructure, etc. At the same time, as a typical under-actuated system, the crane can only control the movement of the load indirectly through rotation or translation, which brings severe challenges to its control research.

[0004] In order to solve the above problems, the existing technology can be roughly divided into two categories: open-loop trajectory planning and closed-loop control. Among them, the open-loop trajectory planning is based on the dynamic characteristics of the crane system, and plans a suitable motion trajectory for it, so as to realize the swing suppression of the load, but its suppression ability to external disturbances is weak, and once the running trajectory deviates from the expected trajectory, the swing of the load will be difficult to suppress, and even will cause more severe swing. Closed-loop control can effectively suppress the swing of the load by measuring the state of the crane in real time through sensors and using intelligent algorithms to calculate and generate the motion trajectory of the crane, and adjusting the running state of the crane in real time, which has attracted widespread attention in recent years.

[0005] However, the existing crane control research focuses on the anti-swing control and precise unloading process of the load transportation process, ignoring the more detailed operation requirements in the working process of the crane. For example, in the hoisting process of large cargos such as bridge box, large fan blades and airplane wings, not only is it required to hoist the cargo without swing, i.e. stable transportation, but also to effectively suppress the swing of the load, i.e. the load is expected to maintain the expected pose during hoisting. At the same time, in the unloading process, the expected pose is the premise to ensure the accurate unloading of the load. For example, in the process of loading and unloading, the container needs to be in the specified pose to smoothly load and unload, otherwise once the pose changes, it is inevitable to cause the collision between the container and the ship body, and even cause a major safety accident. The existing container hoisting shore crane system mainly adopts a mechanical method to suppress the rotation of the hoisted container during hoisting, but this also leads to an increase in system weight and increases maintenance costs.

[0006] In addition, as a special equipment, it is difficult to directly carry out automation and intelligent research experiment on the quay crane, therefore, the commonly used method in the existing research is to build a small experimental platform to carry out preliminary theoretical method verification through the experimental platform, which can be generally divided into quay crane experimental platform based on real-time simulator and scaled experimental platform based on actual quay crane system according to the implementation mode. Among them, the quay crane experimental platform based on the simulator carries out online compilation of the algorithm through the real-time simulator such as Speedgoat, dSPACE and NI, and directly controls the motor through the driver to realize the motion control of the quay crane system, which avoids the secondary development of the program after the simulation of the control method, greatly saves the experimental verification time, but the idea of the quay crane experimental platform based on the simulator is difficult to be directly applied to the actual quay crane system, because the high-cost real-time simulator is difficult to be directly applied. In addition, the scaled model based on the actual quay crane mainly relies on PLC for operation, which has high reliability, but also has the problem of limited computing power, which is difficult to integrate high-level control algorithm. SUMMARY

[0007] In order to solve the technical problems existing in the background art, the present application provides a body-intelligent quay crane scaled experimental platform and a control method, which proposes an active position adjustment method to realize the suppression of load swing by controlling the size and direction of the installation in the hoist power device, with low cost.

[0008] In order to achieve the above purpose, the present application adopts the following technical scheme: The first aspect of the present application provides a body-intelligent quay crane scaled experimental platform.

[0009] A body-intelligent quay crane scaled experimental platform, comprising: a hoisting mechanism, a trolley mechanism, a luffing mechanism, a cart mechanism and a control system, the control system is connected with the hoisting mechanism, the trolley mechanism, the luffing mechanism and the cart mechanism, through the longitudinal movement of the cart mechanism, the arm angle adjustment of the luffing mechanism, the transverse movement of the trolley mechanism and the vertical lifting of the hoisting mechanism, the system completes the loading and unloading of the load; the luffing mechanism is provided with a hoisting mechanism pulley, the hoisting mechanism pulley is hoisted with a hook through a rope, the hook is connected with a spreader, the spreader is used for hoisting the load; the spreader is provided with a power device, a vertical rotation mechanism and a horizontal rotation mechanism, the power device provides active position control thrust for the spreader; the vertical rotation mechanism is used to drive the power device to rotate around the vertical direction, and the horizontal rotation mechanism is used to drive the power device to rotate around the horizontal direction.

[0010] Further, the control system comprises a perception module, a control module and an execution module, the control module is connected with the perception module and the execution module respectively, The sensing module includes: a trolley mechanism displacement sensor for monitoring the positional movement of the trolley mechanism; a lifting mechanism encoder for detecting the rotation angle or speed of the lifting mechanism; a vision sensor for recognizing target objects or environmental features through images; a lidar for acquiring high-precision distance and three-dimensional environmental information through laser scanning; a millimeter-wave radar for detecting the speed and position of distant objects; a trolley mechanism encoder for feeding back the real-time speed or position of the trolley motor; a lifting mechanism displacement sensor for measuring the linear displacement of the lifting mechanism; a tension sensor for monitoring the tension change of the lifting rope; and an inclination measuring device for detecting the tilt angle of the load. The control module includes: a controller, a PLC, and a monitoring display terminal; The execution module includes: a lifting mechanism driver, a lifting mechanism motor, a trolley mechanism driver, and a trolley mechanism motor. The lifting mechanism driver is used to control the start, stop, speed, and direction of the lifting mechanism motor; the trolley mechanism driver is used to adjust the motion parameters of the trolley mechanism motor.

[0011] A second aspect of the present invention provides a control method for a scaled-down experimental platform for an embodied intelligent quay bridge.

[0012] A control method for a scaled-down experimental platform for an embodied intelligent quay crane, applied to the scaled-down experimental platform for an embodied intelligent quay crane as described in the first aspect, includes: The initial state of the scaled-down experimental platform of the embodied intelligent quay crane was determined, and a cumulative discount reward, saturation constraint and segmented penalty mechanism was introduced to optimize the anti-sway trajectory of the scaled-down experimental platform of the embodied intelligent quay crane. Based on the length and swing angle of the lifting rope, determine the angular velocity of the load swing; based on the angular velocity and swing angle of the load, determine the load energy; based on the load energy and the work done by the dynamic device on the load, determine the duration of the power device's action, and obtain the magnitude and direction of the force of the power device that suppresses the load swing. Identify the load location and unloading location, and based on the relative spatial relationship between the load location and unloading location, train a network using the PPO algorithm to predict the hoisting path; The load is hoisted according to the hoisting path, and static and dynamic obstacles are identified by LiDAR. The trajectory of dynamic obstacles is predicted, and the hoisting trajectory is adjusted to avoid obstacles.

[0013] Furthermore, the cumulative discount reward is as follows:

[0014]

[0015]

[0016] in, Indicates cumulative discount rewards; It is the discount factor; T This refers to the crane system's operating time; This indicates a reward for the displacement of the large vehicle. Indicates the reward for the underdriven state; These are the load swing angle and angular velocity, respectively.

[0017] Furthermore, the saturation constraint is:

[0018] in, ; Indicates the control quantity.

[0019] Furthermore, the segmented penalty mechanism is as follows: (1) Between 0 and t Between 1 and 2, there is no restriction on the load swing angle; (2) In t 1 and t Between 2, set the trigger condition to the maximum swing angle; if the condition is not met, training will restart. (3) In t 2 and t Between 3, the design trigger condition is the residual swing angle, which satisfies the desired state; in, t 1. t 2. This can be determined based on experience. t 3 represents the total runtime.

[0020] Furthermore, the method for identifying the load location includes: acquiring a load image, using a YOLO model to identify candidate boxes, and combining a loss function to filter the target box where the load is located; Methods for identifying the unloading location include using semantic segmentation algorithms to divide each pixel in the image into the unloading region and other regions.

[0021] Furthermore, the method for predicting the trajectory of dynamic obstacles includes: acquiring three-dimensional point cloud data of the obstacles, obtaining obstacle clusters based on Euclidean clustering, calculating the change in point cloud data between two frames, determining whether it exceeds a set threshold, and if so, identifying the obstacle as dynamic or static; using a Kalman filter to predict the trajectory of the dynamic obstacle and calculating the minimum distance between the hoisting load and the dynamic obstacle.

[0022] Furthermore, in the process of training the network using the PPO algorithm to predict the hoisting path, noise introduced by random action exploration is taken into account.

[0023] Furthermore, the method for optimizing the anti-sway trajectory of the scaled-down experimental platform for the intelligent quay bridge includes: Based on the initial state of the embodied intelligent quay crane scaled-down experimental platform, the old motion network, the new motion network, and the comment network are determined; A novel motion network is used to generate multiple trajectories; The collected states are selected as input to the comment network, and the value estimation function is calculated at each step. A shearing objective function is introduced to update mini-batches across multiple training steps; The new action network parameters are assigned to the old action network, and optimization is performed by cutting the objective function; The Beta strategy is used to determine the control input trajectory.

[0024] Compared with the prior art, the beneficial effects of the present invention are: This invention provides a scaled-down experimental platform for embodied intelligent quay cranes. The main trolley has two states, high and low, and can be easily moved manually. The trolley's pitch mechanism is foldable for convenient transportation. The auxiliary trolley mechanism can move along the main trolley track via rope traction. The lifting mechanism controls the raising and lowering of the hook, spreader, and load through pulleys fixed to the auxiliary trolley. The power unit is fixed to the spreader and can control the load's posture by applying active force to the load. Furthermore, this invention builds an embodied intelligent system based on the ROS system, which can achieve automatic load identification, automatic grasping, automatic anti-sway transportation, obstacle identification and automatic obstacle avoidance, active posture control, and unloading position identification and precise unloading. The invention's strong perception capabilities, powerful computing platform, and ingenious structure provide a good platform for research on embodied intelligent algorithms for quay crane systems.

[0025] This invention provides a control method for a scaled-down experimental platform of an embodied intelligent quay crane. Compared to the quay crane system's method of preventing swaying by adjusting the trolley, carriage, and lifting mechanism, the active posture adjustment function proposed in this invention can suppress load swaying by controlling the size and direction of the power unit installed on the spreader. This method offers significant advantages such as fast response speed, high control precision, and good stability. In particular, when the lifting rope is long, it can no longer be simply simplified as rigid; its flexibility cannot be ignored. In this case, the effectiveness of suppressing load swaying through the movement of the trolley and carriage will be significantly reduced. Based on the active suppression method proposed in this invention, the flexibility of the lifting rope can be ignored, allowing direct control of the load. This significantly improves load stability, facilitates precise unloading, and ultimately enhances the working efficiency and operational safety of the quay crane, helping enterprises reduce costs and increase efficiency. Attached Figure Description

[0026] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0027] Figure 1 This is a schematic diagram of the overall structure of the scaled-down experimental platform for the embodied intelligent quay bridge shown in an embodiment of the present invention; Figure 2 This is a schematic diagram of the trolley mechanism in the scaled-down experimental platform of the embodied intelligent quay crane shown in an embodiment of the present invention; Figure 3 This is a schematic diagram of the lifting mechanism in the scaled-down experimental platform of the embodied intelligent quay bridge shown in an embodiment of the present invention; Figure 4 This is a schematic diagram of the trolley mechanism in the scaled-down experimental platform of the embodied intelligent quay bridge shown in an embodiment of the present invention; Figure 5 This is a schematic diagram of the lifting mechanism in the scaled-down experimental platform of the embodied intelligent quay bridge shown in an embodiment of the present invention; Figure 6 This is a schematic diagram of the active load device in the scaled-down experimental platform of the embodied intelligent quay bridge shown in an embodiment of the present invention; Figure 7 This is a block diagram of the control system in the scaled-down experimental platform of the embodied intelligent quay bridge shown in the embodiment of the present invention; Figure 8 This is a schematic diagram of the lifting mechanism in the scaled-down experimental platform of the embodied intelligent quay bridge shown in an embodiment of the present invention; The components include: 1. Lifting mechanism; 2. Trolley mechanism; 3. Pitching mechanism; 4. Main trolley mechanism; 5. Main trolley high support mechanism; 6. Main trolley high wheel; 7. Main trolley low support mechanism; 8. Main trolley low wheel; 9. Lifting mechanism left pulley; 15. Hook; 16. Lifting device; 17. Load; 18. Trolley mechanism left pulley; 19. Trolley mechanism drum; 20. Trolley mechanism motor and reducer; 21. Trolley mechanism right pulley; 22. Trolley mechanism slide; 23. Trolley mechanism guide wheel; 24. Trolley mechanism wheel; 25. Lifting mechanism contraction nut and bolt; 26. Lifting mechanism opening and closing mechanism; 27. Lifting mechanism fixed support; 28. Lifting mechanism support support. 9. Power unit; 30. Vertical rotation mechanism; 31. Horizontal rotation mechanism; 32. Trolley mechanism displacement sensor; 33. Lifting mechanism encoder; 34. Vision sensor; 35. LiDAR; 36. Millimeter-wave radar; 37. Trolley mechanism encoder; 38. Lifting mechanism displacement sensor; 39. Tension sensor; 40. Inclination measuring device; 41. Controller; 42. PLC; 43. Monitoring display terminal; 44. Lifting mechanism driver; 45. Lifting mechanism motor; 46. Trolley mechanism driver; 47. Trolley mechanism motor; 48. Upper pulley of the first lifting mechanism; 49. Upper pulley of the second lifting mechanism; 50. Lower pulley of the lifting mechanism. Detailed Implementation

[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0029] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0030] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0031] Figure 1 This is a schematic diagram of the overall structure of the scaled-down experimental platform for the embodied intelligent quay bridge, as shown in an embodiment of the present invention; see reference. Figure 1 The intelligent quay crane scaled-down experimental platform includes: a lifting mechanism 1, a trolley mechanism 2, a pitching mechanism 3, and a trolley mechanism 4. The trolley mechanism 4 is connected to the pitching mechanism 3. The pitching mechanism 3 is equipped with the lifting mechanism 1 and the trolley mechanism 2. The trolley mechanism 2 is connected to the lifting mechanism 1 and drives the lifting mechanism to move.

[0032] Figure 2 This is a schematic diagram of the trolley mechanism in the scaled-down experimental platform of the embodied intelligent quay crane shown in an embodiment of the present invention; see reference. Figure 2 The trolley mechanism includes: a high support mechanism 5, a high wheel 6, a low support mechanism 7, and a low wheel 8. The high support mechanism 5 and the high wheel 6 are mounted on the trolley support frame. The high support mechanism 5 is connected to the high wheel 6. The trolley support frame is mounted on the low support mechanism 7, and the low support mechanism 7 is mounted on the low wheel 8.

[0033] like Figure 1 , Figure 2 As shown, considering the inconvenience of transporting a scaled-down experimental platform for an embodied intelligent quay crane due to its considerable length, the extension of the pitching mechanism in the experimental platform is designed to be foldable. It can be folded downwards during transport, and the experimental configuration is as follows: Figure 1 As shown, this facilitates the movement of the trolley mechanism 2. Compared to the actual movement requirements of a quay crane, the movement requirements of this experimental platform are relatively low. Therefore, the trolley mechanism 4 adopts a manual movement method and is designed with a detachable movement mechanism. It moves in both high and low states through the high wheels 6 and the low wheels 8 of the trolley. At the same time, it is fixed by the high support mechanism 5 and the low support mechanism 7 of the trolley. This satisfies both the adjustable lifting height and the rapid movement requirements of the rapid experimental platform.

[0034] Figure 3This is a schematic diagram of the lifting mechanism in the scaled-down experimental platform of the embodied intelligent quay bridge, as shown in an embodiment of the present invention; see reference. Figure 3 The lifting mechanism includes: two parallel rails with slidable lifting mechanism pulleys 13, the lifting mechanism pulleys 13 being equipped with hooks 15 via ropes, the hooks 15 being connected to lifting devices 16, and the lifting devices 16 being used to lift loads 17.

[0035] Figure 4 This is a schematic diagram of the trolley mechanism in the scaled-down experimental platform for an embodied intelligent quay crane, as shown in an embodiment of the present invention; see reference. Figure 4 The trolley mechanism includes: a left pulley 18, a drum 19, a motor and reducer 20, a right pulley 21, a slide block 22, a guide wheel 23, and a wheel 24. The left and right pulleys 18 and 21 are located at opposite ends of the track. They are connected to the drum 19, motor and reducer 20, slide block 22, guide wheel 23, and wheel 24 via ropes. The slide block 22 connects the wheel 24 and guide wheel 23. The guide wheel 23 guides the trajectory, while the wheel 24 provides driving force and support. The direction of the rope can be changed using the left and right pulleys 14.

[0036] Figure 5 This is a schematic diagram of the lifting mechanism in the scaled-down experimental platform of the intelligent quay bridge according to an embodiment of the present invention; see reference. Figure 5 The lifting mechanism is used to connect the trolley support frame and the trolley low wheel 8. The lifting mechanism includes: lifting mechanism contraction nut and bolt 25, lifting mechanism opening and closing mechanism 26, lifting mechanism fixed support 27 and lifting mechanism support 28. The lifting mechanism opening and closing mechanism 26 is fixed by the lifting mechanism contraction nut and bolt 25. The lifting mechanism opening and closing mechanism 26 is connected to the trolley support frame through the lifting mechanism fixed support 27. The lifting mechanism opening and closing mechanism 26 is connected to the trolley low wheel 8 through the lifting mechanism support 28.

[0037] Figure 6 This is a schematic diagram of the active load device in the scaled-down experimental platform of the embodied intelligent quay bridge shown in an embodiment of the present invention; see reference. Figure 6The spreader 16 includes an active load device, which comprises a power unit 29, a vertical rotation mechanism 30, and a horizontal rotation mechanism 31. Active posture control utilizes the power unit 29 mounted on the spreader 16, based on fluid dynamics principles, to suppress the swaying of the load 17 by providing a reverse thrust. Furthermore, unlike methods that use a rotary motor to prevent torsion, this invention adjusts the direction of the power from the power unit 29 through the vertical and horizontal rotation mechanisms, thereby enabling adjustment of the spreader 16's rotation direction. The proposed intelligent quay crane platform possesses intelligent lifting capabilities. Based on automatic anti-sway and active posture control functions, it achieves precise gripping and stacking of the load 17 through active adjustment of the load 17's posture.

[0038] like Figure 3 , Figure 4 , Figure 6 As shown, the active position control system includes a power unit 29, a vertical rotation mechanism 30, and a horizontal rotation mechanism 31. The power unit 29 can be a propeller mechanism (such as a propeller drive mechanism), a high-pressure pneumatic device, etc., which provides active position control thrust to the lifting device 16 through the rapid rotation of the propeller or the injection of high-pressure gas. Meanwhile, considering that the lifting device 16 is connected to the trolley mechanism slide 22 via a traction rope, the movement of the lifting device 16 is similar to the swinging motion of a pendulum. Furthermore, to change the thrust direction of the power unit 29, a vertically arranged vertical rotation mechanism 30 and a horizontal rotation mechanism 31 are designed, which can drive the power unit 29 to rotate in the vertical and horizontal directions. The vertical direction is the lifting direction of the load 17, and the horizontal direction is along the long side of the lifting device 16.

[0039] Figure 7 This is a block diagram of the control system in the scaled-down experimental platform for the embodied intelligent quay bridge shown in an embodiment of the present invention; see reference. Figure 7In this embodiment, the intelligent quay crane scaled-down experimental platform includes a control system. The control system comprises a sensing module, a control module, and an execution module. The control module is connected to both the sensing module and the execution module. The sensing module includes a trolley mechanism displacement sensor 32, a lifting mechanism encoder 33, a vision sensor 34, a lidar 35, a millimeter-wave radar 36, a trolley mechanism encoder 37, a lifting mechanism displacement sensor 38, a tension sensor 39, and an inclination measurement device 40. The trolley mechanism displacement sensor 32 monitors the positional movement of the trolley mechanism; the lifting mechanism encoder 33 detects the position of the lifting mechanism. The system includes a rotation angle or speed sensor; a vision sensor 34 for identifying target objects or environmental features through images; a lidar 35 for acquiring high-precision distance and three-dimensional environmental information through laser scanning; a millimeter-wave radar 36 for detecting the speed and position of distant objects, adapting to complex weather conditions; a trolley mechanism encoder 37 for feeding back the real-time speed or position of the trolley motor; a lifting mechanism displacement sensor 38 for measuring the linear displacement of the lifting mechanism; a tension sensor 39 for monitoring changes in the tension of the lifting rope; and an inclination measuring device 40 for detecting the tilt angle of the load. The control module includes a controller 41, a PLC 42, and a monitoring display terminal 43. The controller 41 is the core computing unit that runs control algorithms (such as PID and motion planning); the PLC 42 coordinates the timing and safety interlocks of each actuator through logic control; and the monitoring display terminal 43 is a human-machine interface that displays data, alarms, and operation commands in real time. The execution module includes: a lifting mechanism driver 44, a lifting mechanism motor 45, a trolley mechanism driver 46, and a trolley mechanism motor 47. The lifting mechanism driver 44 is used to control the start, stop, speed, and direction of the lifting mechanism motor 45; the lifting mechanism motor 45 is used to drive the lifting mechanism (such as a winch) to achieve lifting and lowering movements; the trolley mechanism driver 46 is used to adjust the motion parameters of the trolley mechanism motor 47; and the trolley mechanism motor 47 is used to drive the trolley to move along the track and adjust its horizontal position.

[0040] This invention relates to intelligent hoisting based on a robot operating system (ROS), integrating the aforementioned functional modules. Sensor signals are connected to the ROS via nodes, and the operation of the trolley mechanism, hoisting mechanism, and power unit is controlled via nodes, thereby achieving automatic load identification, automatic grasping, automatic anti-sway transportation, automatic obstacle avoidance, active posture control, and precise unloading.

[0041] Figure 8 This is a schematic diagram of the lifting mechanism in the scaled-down experimental platform of the embodied intelligent quay bridge, as shown in an embodiment of the present invention; see reference. Figure 8The lifting mechanism includes: a first lifting mechanism upper pulley 48, a second lifting mechanism upper pulley 49, and a lifting mechanism lower pulley 50. The trolley mechanism 2 is fixed to eight first lifting mechanism upper pulleys 48 and eight second lifting mechanism upper pulleys 49, with one of the first lifting mechanism upper pulleys 48 and two lifting mechanism upper pulleys 49 corresponding to one lifting mechanism lower pulley 50. A traction rope enters from one first lifting mechanism upper pulley 48, passes around one lifting mechanism lower pulley 50, and then passes through the second lifting mechanism upper pulley 49 paired with that first lifting mechanism upper pulley 48, thereby achieving traction of one lifting mechanism lower pulley 50. The traction method for the other three lower pulleys is similar, only the rope direction is different.

[0042] like Figure 1 , Figure 3 , Figure 4 , Figure 8 As shown, the scaled-down experimental platform for the embodied intelligent quay crane proposed in this invention includes: a lifting mechanism 1, a trolley mechanism 2, a pitching mechanism 3, and a trolley mechanism 4. Specifically, the trolley achieves translational movement through rope traction. One end of the traction rope is fixed to the drum 19 of the trolley mechanism, and the other end is fixed to the side of the trolley. Directional changes are achieved through the left pulley 18 and the right pulley 21 of the trolley mechanism, which are respectively fixed to both ends of the trolley mechanism's running track. The lifting mechanism 1 controls the vertical lifting and lowering of the load 17 through rope traction. Four lifting mechanism pulleys 50 are fixed on the lifting device 16.

[0043] The embodied intelligent quay crane platform proposed in this embodiment also has an automatic anti-sway function. Based on the quay crane dynamics model, it uses deep reinforcement learning to suppress the swaying of the load 17 by controlling the operation of the trolley mechanism 2 and the lifting mechanism 1.

[0044] Another embodiment of the present invention provides a control method for a scaled-down experimental platform for an intelligent quay bridge, enabling the aforementioned scaled-down experimental platform for an intelligent quay bridge to have an automatic anti-sway function.

[0045] Environment definition: Define the state of the quay crane platform as follows ,in, These are the displacement and velocity of the trolley, respectively. These represent the load swing angle and angular velocity, respectively. The initial state is... Where x0>0 indicates that the initial displacement is greater than zero, thus ensuring the normal training of the PPO algorithm. Because if x Setting 0=0 might result in a large reward near the origin during training, which is not desirable. Also, choosing... To control the quantity, it is denoted as a The sampling time is 0.01 s. Without loss of generality, the acceleration and deceleration processes are symmetrical, so only the trajectory planning of the acceleration process is considered.

[0046] In deep reinforcement learning, the quality of actions is evaluated using a reward function. The proposed control policy is optimal when the total reward is maximized. For quay crane systems, this invention proposes the following reward function. Specifically, the reward function for the trolley displacement is first determined, as follows:

[0047] Furthermore, to avoid excessive rewards due to drastic changes in the underactuated state, the underactuated state reward function is designed as follows:

[0048] Furthermore, the cumulative discount reward can be obtained as follows:

[0049] in, It is a discount factor used to reduce variance, and T is the crane system running time.

[0050] Next, consider For each step, the following saturation constraints are designed:

[0051] Subsequently, in order to satisfy x, , To address the constraints, a penalty mechanism is introduced. Specifically, for the load, its state constraints satisfy... Based on this, a trigger condition is designed so that training will restart when the state constraints are not met.

[0052] Meanwhile, to ensure convergence and improve training speed, the following segmented penalty mechanism is proposed based on the time series: (1) Between 0 and t Within a range of 1, there is no restriction on the load swing angle. If the swing angle is restricted, the crane system may remain in the initial position; (2) In t 1 and t Between 2, set the trigger condition to the maximum swing angle. If the condition is not met, training will restart; (3) In t 2 and t Between 3, the design trigger condition is the residual swing angle, which satisfies the desired state.

[0053] in, t 1. t 2. This can be determined based on experience. t 3 represents the total runtime.

[0054] Specifically, the proximal policy optimization (PPO) algorithm consists of three parts: an old action network, a new action network, and a comment network. For the quay crane experimental system, the output of the action network is... a t The input to the action network can be selected using a probability distribution function, and the input to the action network is defined as... The trigonometric functions are used to ensure that the input value of the action network is at its maximum when the desired angle is 0.

[0055] During training, the new action network first interacts with the established environment, generating multiple trajectories. Based on this, the collected states are selected as input to the comment network to compute the value estimation function for each step. Furthermore, to reduce variance while maintaining an acceptable level of bias, a generalized advantage estimator (GAE) is defined as the advantage function, with the expression:

[0056] in, This represents how much more cumulative reward choosing an action in the current state will bring compared to the average value of all possible actions in that state; V is an approximation function. It is about The temporal-difference (TD) residuals of the value function V can be obtained through... calculate, Each at collects N trajectory segments. This represents the instantaneous reward at step t; This is to make a trade-off between bias and variance.

[0057] Next, a shearing objective function is introduced to achieve mini-batch updates across multiple training steps, thus addressing the difficulty in determining the step size in the policy gradient algorithm. Subsequently, the new action network parameters are assigned to the old action network, and optimization is performed using the following shearing objective function:

[0058] in, The terminology is defined as a truncation function, which can limit the magnitude of policy updates and avoid training instability due to excessive policy changes. In other words, when optimizing a new policy, the probability ratio of the new and old policies is constrained within a certain range, thereby balancing policy improvement and stability. It is the expected empirical average estimate. Indicates the strategy parameters, Define the ratio between the old and new strategies. This represents the current strategy, i.e., the new strategy. This indicates the old strategy used in the last sampling. Used to determine the shear function The hyperparameters are used to limit the shearing probability ratio of the policy loss function in order to reduce... and The gap between them.

[0059] Specifically, by cutting Limiting it to always be in or The exact value depends on whether A is negative or positive.

[0060] Subsequently, to satisfy the control constraints of the AC asynchronous motor, the control input trajectory is determined using the following Beta strategy:

[0061] in, , These are parameters that define the boundaries. , Depend on Decide, Represents the Gamma function. Indicates the maximum acceleration. a This represents acceleration. The Beta policy can map actions to the desired acceleration space [0,1].

[0062] Furthermore, the comment network parameters are optimized using the mean-squared error (MSE) regression method.

[0063] in, yes Quantity, .

[0064] Based on the above description, an anti-sway trajectory planning method for a PPO-based quay crane experimental system can be obtained. Specifically, considering the noise introduced by random action exploration during the learning process, the following rolling mean filter is designed:

[0065] In the formula, , Represents a data point.

[0066] The intelligent quay crane scaled-down experimental platform features active posture control: Compared to quay crane systems that rely on adjusting the trolley, crane mechanism, and lifting system to prevent swaying, the active posture control function proposed in this invention suppresses load swaying by controlling the size and direction of the power unit installed on the spreader. Specifically, the power unit is divided into two groups, one moving in the direction of the trolley's movement and the other in the opposite direction. Simultaneously, the tilt angle measuring device can measure the rope's sway angle in real time. The lifting mechanism can measure the length of the vertical lifting rope in real time. Then the angular velocity of the load swing is:

[0067] In the formula, g is the gravitational acceleration constant.

[0068] Furthermore, the load swing angle is At that time, the load energy equation is:

[0069] In the formula, m represents the load and the mass of the lifting device and power unit.

[0070] Assuming the forces acting on the two power units on one side are constant. If the time of action is t, then the work done by the dynamic device on the load is:

[0071] Based on the law of conservation of energy, using the load energy equation and From the formula, the duration of action of the power unit can be obtained as follows:

[0072] The intelligent hoisting system proposed in this invention achieves precise hoisting and highly dynamic obstacle avoidance through visual sensors, lidar, and millimeter-wave radar. The visual sensors are used to accurately identify the shape of the load and the unloading position. The lidar is used for real-time monitoring of static and dynamic obstacles and works in conjunction with the visual sensors to achieve real-time monitoring of the load's pose. The millimeter-wave radar is used for obstacle identification and tracking of high-speed moving loads and obstacles in adverse weather conditions (such as heavy fog, rain, and snow).

[0073] Load recognition based on deep learning: First, load images are acquired under different viewpoints, lighting conditions, weather conditions, etc., and the category, location, and bounding box of each load are labeled. After normalization processing, a dataset of loads and unloading locations is established. Then, the object detection problem is transformed into a regression problem using the You Only Look Once (YOLO) model, thereby achieving automatic load recognition.

[0074] The loss function is defined as follows:

[0075]

[0076]

[0077]

[0078] in, This represents the bounding box coordinate regression loss (responsible for predicting the location and size of the load); This represents the load category classification loss (responsible for identifying the specific type of load, such as "wooden box", "metal barrel", etc.). This represents the target confidence loss (responsible for determining whether the area within the bounding box is a load, and the accuracy of the localization). These are real category labels; It is the predicted class probability; These are the coordinates and dimensions of the actual bounding box; It predicts the coordinates and dimensions of the bounding box; , This is the scaling factor; It is the true confidence level. It represents the confidence level of the prediction.

[0079] Once a load is identified, the network can classify it based on features such as its number, size, and color. Subsequently, to further improve the accuracy of load identification, non-maximum suppression is used to remove highly overlapping boxes, retaining only the most representative bounding boxes to ensure that each load is identified only once.

[0080] Let the set of candidate boxes be The corresponding confidence level is Then, according to The candidate box with the best output score is used to label the load, where , Represents different bounding boxes.

[0081] Meanwhile, the detection network in the initial candidate boxes Based on this, and using a smoothed L1 loss function, a target bias is obtained through regression learning, thereby adjusting the initial candidate boxes to the true boxes. The regression value is The formula for calculating the prediction box is: .

[0082] Deep learning-based unloading location recognition works as follows: First, semantic segmentation algorithms are used to divide each pixel in the image into unloading areas and other areas. Specifically, the U-Net network is used to process spatial information and generate classification results at the pixel level. Then, a convolutional neural network (CNN) is used to extract image features, and unloading area recognition is achieved based on pixel-by-pixel prediction. Furthermore, based on the relative spatial relationship between the load and the unloading location, the PPO algorithm is used to train the network to predict the hoisting path, ensuring that the load accurately reaches the target location during unloading and avoiding collisions with obstacles.

[0083] Static and Dynamic Obstacle Recognition: This invention employs lidar to identify static and dynamic obstacles in the quay crane's working area, thereby enabling proactive obstacle avoidance planning during load transportation. The lidar measures the distance to surrounding objects by emitting laser pulses and receiving reflected light signals, generating high-precision 3D point cloud data, including obstacle clusters obtained based on Euclidean clustering. Two frames of point cloud data are calculated. Change In the formula, To set a threshold, q , p These represent the point cloud of the current frame, respectively. Points in the middle and the point cloud of the previous frame Points in the cloud are considered. If the point cloud time series does not change over time, it is considered a static obstacle; if the point cloud time series changes, it is considered a dynamic obstacle. Subsequently, the Kalman filter method is used to predict the trajectory of the dynamic obstacle, and the calculation formula is:

[0084] in, This represents the dynamic obstacle state vector at the current time t. These represent the state matrix, control input matrix, and process noise covariance matrix, respectively. To control the input amount, The covariance matrix represents the uncertainty of the current state.

[0085] Furthermore, the minimum distance between the hoisting load and the obstacle is:

[0086] in, To generate obstacle point clouds in real time for LiDAR, The predicted location for loading the load is then determined. Subsequently, considering both static obstacle maps and dynamic obstacle trajectories, the hoisting trajectory is dynamically adjusted to achieve safe obstacle avoidance.

[0087] Obstacle recognition and high-speed moving load and obstacle tracking in adverse weather conditions (such as heavy fog, rain, etc.): Based on the good penetration of millimeter wave signals in fog, haze, rain, and snow, obstacle distance detection can be achieved in low-visibility environments. ,speed and angle recognition, among which These represent echo beat frequency, speed of light, frequency modulation period, and bandwidth, respectively. These represent the wavelength and carrier frequency of the millimeter-wave signal, respectively. The state equation for the tracked target is further constructed, and the continuous trajectory of the target can be obtained using an extended Kalman filter. Obstacle avoidance planning is triggered when the tracked target enters a safe area.

[0088] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A physical intelligent shore-bridge scaled experiment platform, characterized in that, The crane mechanism, the trolley mechanism, the luffing mechanism, the crab mechanism, and the control system are connected, and the load is loaded and unloaded through the longitudinal movement of the crab mechanism, the adjustment of the arm angle of the luffing mechanism, the transverse movement of the trolley mechanism, and the vertical lifting of the crane mechanism; the luffing mechanism is provided with a crane mechanism pulley, the crane mechanism pulley is hoisted with a hook through a rope, the hook is connected with a sling, and the sling is used for hoisting the load; the sling is provided with a power device, a vertical rotating mechanism, and a horizontal rotating mechanism, the power device provides a driving position control thrust for the sling, the vertical rotating mechanism is used for driving the power device to rotate around the vertical direction, and the horizontal rotating mechanism is used for driving the power device to rotate around the horizontal direction. The control system comprises a sensing module, a control module, and an execution module, the control module is connected with the sensing module and the execution module, 2. The body-aware shore crane scale experiment platform according to claim 1, wherein, The sensing module comprises a trolley mechanism displacement sensor for monitoring the position movement of the trolley mechanism, a crane mechanism encoder for detecting the rotation angle or speed of the crane mechanism, a visual sensor for identifying target objects or environmental features through image recognition, a laser radar for obtaining high-precision distance and three-dimensional environmental information through laser scanning, a millimeter wave radar for detecting the speed and position of a long-distance object, a trolley mechanism encoder for feeding back the real-time rotating speed or position of the trolley motor, a crane mechanism displacement sensor for measuring the linear displacement of the crane mechanism, a tension sensor for monitoring the tension change of the hoisting rope, and an inclination measuring device for detecting the inclination angle of the load; The control module comprises a controller, a PLC, and a monitoring display end; The execution module comprises a crane mechanism driver, a crane mechanism motor, a trolley mechanism driver, and a trolley mechanism motor, the crane mechanism driver is used for controlling the start-stop, rotating speed, and rotating direction of the crane mechanism motor, and the trolley mechanism driver is used for adjusting the motion parameters of the trolley mechanism motor. The application is applied to the body-integrated intelligent shore crane scale experiment platform in any one of claims 1-2, comprising:

3. A control method for a physical intelligent shore-bridge scale experiment platform, characterized in that, An initial state of the body-integrated intelligent shore crane scale experiment platform is determined, a cumulative discount reward, a saturation constraint, and a segmented penalty item mechanism are introduced, and a swing prevention trajectory of the body-integrated intelligent shore crane scale experiment platform is optimized; A load swing angular velocity is determined according to a hoisting rope length and a hoisting rope swing angle, a load energy is determined based on the load swing angular velocity and the load swing angle, a dynamic device acting time is determined based on the load energy and work of the dynamic device on the load, and a dynamic device acting force size and direction for suppressing load swing are obtained; A load position and an unloading position are identified, a network is trained to predict a hoisting path based on a relative spatial relationship between the load position and the unloading position by using a PPO algorithm, and the load is hoisted according to the hoisting path. The cumulative discount reward is: The saturation constraint is:

4. The body-aware control method for a scaled-down container yard platform according to claim 3, wherein, The segmented penalty item mechanism is: wherein, represents the accumulated discount reward; is a discount factor; T is the crane system operating time; represents the trolley displacement reward; represents the underdrive state reward; are the load swing angle and angular velocity, respectively.

5. The body-aware control method for a scaled-down container yard experiment platform according to claim 3, wherein, ​ wherein ; denotes a control variable.

6. The body-aware control method for a scaled-down container yard experiment platform according to claim 3, wherein, ​ (1) between 0 and t 1, without limiting the load swing angle; (2) In t 1 and t 2, set the trigger condition as the maximum swing angle; if the condition is not met, the training will restart; (3) In t 2 and t 3, the trigger condition is designed as the residual swing angle, which meets the expected state. wherein, t 1, t 2may be determined empirically, t 3is the total run time.

7. The body-aware control method for a scaled-down container yard experiment platform according to claim 3, wherein, The method for identifying the load position comprises: acquiring a load image, identifying a candidate box by using a YOLO model, and screening a target box where the load is located in combination with a loss function; The method for identifying the unloading position comprises: using a semantic segmentation algorithm to divide each pixel in the image into an unloading area and other areas.

8. The body-aware control method for a scaled-down container yard platform according to claim 3, wherein, The method for predicting the trajectory of the dynamic obstacle comprises: acquiring three-dimensional point cloud data of the obstacle, obtaining an obstacle clustering block based on Euclidean clustering, calculating a change amount of two frames of point clouds, judging whether the change amount is greater than a set threshold, if yes, the obstacle is a dynamic obstacle, otherwise, the obstacle is a static obstacle; using a Kalman filtering method to predict the trajectory of the dynamic obstacle, and calculating a minimum distance between the hoisted load and the dynamic obstacle.

9. The body-aware control method for a scaled-down container yard platform according to claim 3, wherein, In the process of training the network to predict the hoisting path by the PPO algorithm, the noise introduced by the random action exploration is considered.

10. The body-aware control method for a scaled-down container yard platform according to claim 3, wherein, The method for optimizing the anti-sway trajectory of the embodied intelligent shore crane scale experiment platform comprises: Based on the initial state of the embodied intelligent shore crane scale experiment platform, an old action network, a new action network and a comment network are determined; The new action network is used to generate multiple trajectories; The state is collected as the input of the comment network, and the value estimation function of each step is calculated; A clipping objective function is introduced to update the small batches of multiple training steps; The new action network parameters are assigned to the old action network, and the old action network is optimized through the clipping objective function; The Beta strategy is used to determine the control input trajectory.

Images