Crane working path planning method and system and application thereof

Abstract

The invention discloses a crane working path planning method and system and application thereof. The method comprises the following steps: 1) constructing a crane operation demonstration database; 2) pre-training a convolutional neural network by using a demonstration database, and extracting a mapping relation between environment features and tracks; 3) inputting the current starting point position, the target point position and the environment information obtained by the laser radar into a path generation network, and outputting an initial action sequence of the rotation degree of freedom, the dolly motion degree of freedom and the hook degree of freedom; 4) performing local re-planning on the initial action sequence by adopting a fast random tree algorithm based on physical constraint, wherein the optimization criterion is the weighted sum minimization of the path length, the load swing angle and the action sequence variation; and (5) the crane is controlled to move according to the optimized action sequence, the load swing angle is monitored in real time, and emergency braking is triggered when the swing angle exceeds a threshold value. By means of the method, efficient and safe operation of the crane in the complex environment can be achieved, and good engineering practicability and popularization value are achieved.

Description

Technical Field

[0001] This invention belongs to the field of crane technology, and in particular relates to a crane working path planning method and system and its application. Background Technology

[0002] Cranes, as large-scale engineering equipment, are widely used in construction, port transportation, warehousing and handling, and other scenarios. Their operation typically involves the coordinated movement of multiple degrees of freedom, such as rotation, trolley movement, and hook lifting, to achieve precise transfer of objects from the starting point to the target point.

[0003] In traditional crane operation, path planning and control mainly rely on manual operation or rule-based preset path schemes. This method is not only inefficient, but also prone to problems such as unreasonable trajectories, large load swings, and unsafe operation in complex environments. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a crane working path planning method and system that enables cranes to operate efficiently and safely in complex environments while ensuring the smoothness and stability of path control, thereby meeting the intelligent operation requirements of cranes in complex scenarios.

[0005] To achieve the above-mentioned technical objectives, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a crane working path planning method, comprising:

[0007] 1) Construct a crane operation demonstration database to store the starting point, target point, environmental information, and corresponding operator trajectory data from historical tasks;

[0008] 2) Use a demonstration database to pre-train a convolutional neural network to extract the mapping relationship between environmental features and trajectory;

[0009] 3) Input the current starting point position, target point position, and environmental information acquired by the lidar into the path generation network, and output the initial motion sequence of rotational degrees of freedom, trolley motion degrees of freedom, and hook degrees of freedom;

[0010] 4) A fast random tree algorithm based on physical constraints is used to perform local replanning on the initial action sequence. The optimization criterion is the weighted sum and minimization of path length, load swing angle, and action sequence variation.

[0011] 5) Control the crane movement according to the optimized action sequence and monitor the load swing angle in real time. When the swing angle exceeds the threshold, trigger emergency braking.

[0012] In some embodiments, step 1) specifically includes:

[0013] 1.1) Collect control trajectory data of operators in historical crane operation tasks, including control degree of freedom data of slewing mechanism, trolley traveling mechanism and hook lifting mechanism, and simultaneously record starting point, target point, environmental obstacle distribution information and load parameters;

[0014] 1.2) Tagged historical tasks, marking the obstacle density level, lifting weight level and task complexity level for each task;

[0015] 1.3) The collected trajectory data is processed by nonlinear dimensionality reduction using a variational autoencoder (VAE) to generate a low-dimensional trajectory representation, and an action prototype library is constructed to represent the potential feature distribution of typical operation trajectories.

[0016] 1.4) Store the basic task information, label information, original trajectory data and corresponding low-dimensional latent variable encoding results in the operation demonstration database for subsequent path generation network training and scheduling strategy calls.

[0017] In some embodiments, step 2) specifically includes:

[0018] 2.1) Construct input-output pairs from the data of each task sample, where:

[0019] The input includes the coordinates of the starting point P. s =(x s y s , z s ), target point coordinates

[0020] P g =(x g y g , z g The two-dimensional environmental grid map E(x,y), obstacle density level D, load weight W, and task complexity C obtained by LiDAR are used as structured features and image features after one-hot encoding.

[0021] The output is the trajectory latent variable code Z = f under the given input conditions. VAE (T(t)), where

[0022] T(t) represents the expert trajectory, f VAE This is a trained variational autoencoder;

[0023] 2.2) Construct a convolutional neural network F(·) to map the above inputs to the trajectory latent variable outputs, satisfying the relationship: Z = F(E(x, y), P s P g (D, W, C);

[0024] 2.3) The mean square error loss function L=||ZZ is adopted. gt || 2 Supervised training is performed on the network, where Z... gt The target value for expert trajectory coding;

[0025] 2.4) After the network training is completed, the neural network can quickly generate the initial trajectory latent variables of the three-degree-of-freedom control action under the condition of any starting point, target point and environment map data, which can be used as the initialization input of the path generation module.

[0026] In some embodiments, step 3) specifically includes:

[0027] 3.1) Construct the task input vector, including the starting point position P. s =(x s y s , z s ), Target point location P g =(x g y g , z g ), a two-dimensional environmental grid map generated by lidar.

[0028] E(x, y), and structured task label information L = {D, W, C}, where D is the obstacle density level, W is the lifting weight level, and C is the task complexity level;

[0029] 3.2) The input vector is fed into a path generation neural network model consisting of a lightweight convolutional network and gated recurrent units, and the output is a time series of three-degree-of-freedom control actions:

[0030] A(t)=[θ rot (t), x car (t), h ook [(t)], t∈[0,T]

[0031] Where, θ rot (t) is the rotation angle, x car (t) represents the position of the trolley, h hook (t) represents the hook height;

[0032] 3.3) The action sequence output by the path generation neural network satisfies the following physical constraints:

[0033] Limitations on the range of rotation angle, the range of movement of the trolley, and the lifting height of the hook.

[0034] In some embodiments, step 4) specifically includes:

[0035] 4.1) Based on the initial action sequence output in step 3), new sampling nodes are generated in the action space using a constraint-based fast random tree algorithm;

[0036] Among them, the constraint-based fast random tree algorithm uses the initial action sequence as the guiding direction to perform action space sampling, in (θ rot x car h ook Candidate nodes are generated in a three-dimensional space, and then the following physical constraints are applied to each sampled node for determination:

[0037] Spatial reachability constraint: Determine whether the load trajectory crosses the obstacle zone; if so, discard the node.

[0038] Action continuity constraint: Determine if it is satisfied If not, then discard the node; δ max This is the preset rate of change threshold.

[0039] Load dynamic stability constraint: Determine the load swing angle θ swing Does (t) satisfy |θ swing (t)|≤θ th If not, then discard the node; θ th 4.2) The path cost function is a weighted combination of path length, load swing angle change, and action sequence change. The optimization objective is as follows:

[0040]

[0041] Where α, β, and γ are weighting coefficients, and PathLength is a path length function of A(t), based on the rotation angle θ in A(t). rot (t), the position of the car x car (t), hook height h hook (t) is obtained;

[0042] 4.3) Minimize the weighted sum J of path length, load swing angle change, and action sequence change to generate the endpoint node that satisfies the objective function optimally;

[0043] 4.4) After generating the optimal endpoint node that satisfies the objective function, the path sequence is extracted using the backtracking method of the random tree algorithm, and interpolation and trajectory smoothing are performed to finally generate the optimized action sequence:

[0044]

[0045] In some embodiments, the interpolation and trajectory smoothing processing in step 4.4) is specifically as follows:

[0046] a) The discrete trajectory node sequence generated by the fast random tree algorithm Each node A continuous-time trajectory function is constructed using cubic spline interpolation.

[0047] A(t) = a i (tt i ) 3 +b i (tt i ) 3 +c i (tt i ) 3 +d i

[0048] And it satisfies positional continuity and first derivative continuity to generate in the interval [t1,t] n The initial continuous trajectory function A(t) is defined on the [theory / parameter].

[0049] b) Based on the interpolated trajectory, further construct the trajectory smoothing optimization objective function:

[0050]

[0051] Wherein, λ1 and λ2 are smoothing adjustment weight parameters;

[0052] c) Adjust the control points in the trajectory interpolation function using optimization methods to minimize the objective function and generate a smoothed motion trajectory sequence A. s (t), where the trajectory of each degree of freedom component is continuous at least on the second derivative; and the load swing angles corresponding to all trajectory points satisfy the following condition.

[0053] |θ swing (t)||≤θ th .

[0054] In some embodiments, step 5) specifically includes:

[0055] 5.1) Optimize the motion trajectory sequence

[0056] The input is sent to the crane control system, which drives the movement of the slewing mechanism, trolley mechanism and hook mechanism step by step according to the control cycle;

[0057] 5.2) The swing angle is collected in real time by an inertial measurement unit installed on the load or hook.

[0058] θ swing (t);

[0059] 5.3) Determine whether the swing angle meets the following safety threshold conditions:

[0060] |θ swing (t)|>θ th

[0061] If the absolute value of the swing angle exceeds the set threshold θ th If the system issues an emergency braking command, it will immediately stop the operation of all drive mechanisms of the crane.

[0062] In some embodiments, braking actions include trolley braking to a stop, slewing locking, and hook braking, and optionally triggering an audible and visual alarm to prompt on-site personnel to take action.

[0063] In a second aspect, the present invention provides a crane work path planning system that applies the above-described crane work path planning method, the system comprising:

[0064] The demonstration database module is used to store operation demonstration data from historical crane operation tasks. The data includes the starting point, target point, environmental information, load weight, task complexity label, and corresponding operation trajectory data. The demonstration database module further includes a trajectory dimensionality reduction unit, which performs low-dimensional encoding on the trajectory data based on a variational autoencoder to build an action prototype library.

[0065] The path generation module includes a pre-trained lightweight neural network, which receives the starting point, target point, and environmental information of the current task and outputs the initial motion trajectory sequence of three degrees of freedom: rotation, trolley, and hook. The path generation module integrates an environmental perception submodule, which receives an environmental grid map provided by a lidar or visual sensor.

[0066] The local path optimization module is used to perform local replanning of the initial motion trajectory based on the fast random tree (RRT) algorithm and combined with physical constraints. The optimization process uses the weighted sum of path length, load swing angle change and motion sequence change as the optimization objective, and performs interpolation and trajectory smoothing on the optimized trajectory.

[0067] The execution control module is used to convert the smoothed trajectory sequence into control signals to drive the crane's slewing mechanism, trolley mechanism and hook mechanism to perform operations according to the planned path;

[0068] The load monitoring module includes an inertial measurement unit, which is used to collect the swing angle state of the load in real time and compare the collected swing angle information with the set angle threshold.

[0069] The safety control module is used when the load swing angle exceeds a preset angle threshold θ. th Emergency braking is triggered when the situation arises; braking actions include stopping the trolley, locking the slewing mechanism, and braking the hook. An audible and visual alarm can also be triggered to alert on-site personnel to take appropriate measures.

[0070] In a third aspect, the present invention also provides a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the above-described crane work path planning method.

[0071] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art:

[0072] This invention provides a crane work path planning method that integrates expert demonstration data, neural network path generation, physical constraint optimization, and safety control mechanisms to achieve intelligent planning and stable execution of automatic crane operation paths. First, this invention constructs an operation demonstration database and introduces a variational autoencoder to reduce the dimensionality of the trajectory, forming a transferable low-dimensional motion prototype library and enhancing the ability to express prior knowledge for path generation. Second, a pre-trained neural network is used to establish a mapping relationship between environmental features and the trajectory, enabling the rapid generation of a three-degree-of-freedom initial motion sequence in dynamic environments. Furthermore, a fast random tree algorithm based on physical constraints is introduced to locally optimize the initial trajectory, using a weighted sum of path length, load swing angle, and motion sequence variation as the objective function to ensure the generated trajectory is executable, continuous, and stable.

[0073] Furthermore, this invention uses an inertial measurement unit to monitor the load swing angle in real time. If the swing angle exceeds a set threshold, emergency braking is automatically triggered, significantly improving operational safety. The overall method combines intelligence, stability, and safety, making it suitable for the automated control of various types of cranes and possessing good engineering practicality and promotional value. Attached Figure Description

[0074] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0075] Figure 1 This is a simplified implementation flowchart of the crane working path planning method of the present invention;

[0076] Figure 2 This is a schematic diagram of a system implementation structure for the crane working path planning method corresponding to the present invention.

[0077] Figure 3 This is a schematic diagram of a computer-readable storage medium provided in an embodiment of the present invention. Detailed Implementation

[0078] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0079] Example 1

[0080] See attached document Figure 1 As shown, this embodiment provides a crane working path planning method, including:

[0081] 1) Construct a crane operation demonstration database to store the starting point, target point, environmental information (including obstacle location, density distribution, etc., acquired by LiDAR), and corresponding operator trajectory data from historical tasks; specifically including:

[0082] 1.1) Collect control trajectory data of operators in historical crane operation tasks, including control degree of freedom data of slewing mechanism, trolley traveling mechanism and hook lifting mechanism, and simultaneously record starting point, target point, environmental obstacle distribution information and load parameters;

[0083] 1.2) Tagged historical tasks, marking the obstacle density level, lifting weight level and task complexity level for each task;

[0084] In this embodiment, the obstacle density level is divided into sparse, medium, and dense based on the number of obstacles per unit area; the lifting weight level is divided according to the load weight, for example: light <1 ton, medium 1-3 tons, heavy >3 tons; the task complexity level is evaluated based on the weighted sum of path curvature, number of obstacles, and change in swing angle, and the larger the value, the higher the task complexity level.

[0085] 1.3) Since the original operation trajectory is usually high in dimensionality and noise, it is difficult to use it directly for training. In this embodiment, a variational autoencoder (VAE) is used to perform nonlinear dimensionality reduction on the collected trajectory data to generate a low-dimensional trajectory representation, and an action prototype library is constructed to represent the potential feature distribution of typical operation trajectories.

[0086] 1.4) Store the basic task information (including: starting point location, target point location, obstacle information, load weight, etc.), label information, original trajectory data and corresponding low-dimensional latent variable encoding results in the operation demonstration database for subsequent path generation network training and scheduling strategy calls.

[0087] 2) Pre-train a convolutional neural network using a demonstration database to extract the mapping relationship between environmental features and trajectory; specifically including:

[0088] 2.1) Construct input-output pairs from the data of each task sample, where:

[0089] The input includes the coordinates of the starting point P. s =(x s y s Z s ), Target point coordinates P g =(x g y g , z g The two-dimensional environment grid map E(x,y) obtained by the lidar (wherein, the environment grid map extracted by the lidar is encoded into a two-dimensional image form, resulting in the two-dimensional environment grid map E(x,y)), obstacle density level D, load weight W, and task complexity C, where D, W, and C are one-hot encoded and then used as structured features and image features spliced ​​together.

[0090] The output is the trajectory latent variable code Z = f under the given input conditions. VAE (T(t)), where T(t) represents the expert trajectory, f VAE This is a trained variational autoencoder;

[0091] 2.2) Construct a convolutional neural network F(·); map the above inputs to the trajectory latent variable outputs, i.e., satisfy the relationship: Z=F(E(x,y), P s P g (D, W, C);

[0092] The network structure mainly includes the following modules:

[0093] Convolutional layers: extract features such as obstacles and reachable areas of the path in the environmental image; batch normalization and activation layers: improve training stability; fully connected layers: combine structured information (starting point, target point, label encoding, etc.); output layer: predict the corresponding latent variables of the action trajectory.

[0094] Furthermore, to improve the model's deployment capability on resource-constrained devices, lightweight network structures (such as MobileNet, ShuffleNet, etc.) can be used for environment-action mapping modeling, and model compression can be performed through pruning and quantization.

[0095] 2.3) The mean square error loss function L=||ZZ is adopted. gt || 2 Supervised training is performed on the network, where Z... gt The target value for expert trajectory coding;

[0096] 2.4) After the network training is completed, the neural network can quickly generate the initial trajectory latent variables of the three-degree-of-freedom control action under the condition of any starting point, target point and environment map data, which can be used as the initialization input of the path generation module.

[0097] 3) Input the current starting point position, target point position, and environmental information acquired by the LiDAR into the lightweight path generation network, and output the initial motion sequence of rotational degrees of freedom, trolley motion degrees of freedom, and hook degrees of freedom; specifically including: 3.1) Constructing the task input vector, including the starting point position

[0098] P s =(x s y s , z s ), Target point location P g =(x g y g , z g The two-dimensional environmental grid map E(x, y) generated by lidar and the structured task label information.

[0099] L = {D, W, C}, where D is the obstacle density level, W is the lifting weight level, and C is the task complexity level;

[0100] 3.2) The input vector is fed into a path generation neural network model consisting of a lightweight convolutional network and gated recurrent units, and the output is a time series of three-degree-of-freedom control actions:

[0101] A(t)=[θ rot (t), x car (t), h hook [(t)], t∈[0,T]

[0102] Where, θ rot (t) is the rotation angle, x car (t) represents the position of the trolley, h hook (t) represents the hook height;

[0103] Furthermore, in this embodiment, to ensure the continuity and smoothness of the action output, the output action trajectory adopts a time window sliding modeling method. The output time step can be set according to the actual operation step size. For example, each moment represents a control cycle of 0.2 seconds, and a total of T=50 steps are output to form a 10-second execution plan.

[0104] 3.3) In this embodiment, the path generation neural network is a lightweight model, preferably with a MobileNet or ShuffleNet backbone structure, and its output action sequence satisfies the following physical constraints:

[0105] Rotation angle range limitation:

[0106]

[0107] Vehicle movement range limitations:

[0108]

[0109] Hook lifting height limit:

[0110]

[0111] in,

[0112] This indicates the minimum and maximum values ​​for the crane's slewing angle, trolley travel distance, and hook lifting height.

[0113] The output initial action sequence is used for subsequent local replanning by a physical constraint-based path optimization algorithm and serves as a reference input for controlling the crane's movement.

[0114] 4) A fast random tree algorithm based on physical constraints is used to locally reprogram the initial action sequence. The optimization criterion is the weighted minimization of path length, load swing angle, and changes in the action sequence; specifically including:

[0115] 4.1) Based on the initial action sequence output in step 3), new sampling nodes are generated in the action space using the constraint-based fast random tree algorithm (RRT).

[0116] Among them, the constraint-based fast random tree algorithm uses the initial action sequence as the guiding direction to perform action space sampling, in (θ rot x car h hook Candidate nodes are generated in a three-dimensional space, and then the following physical constraints are applied to each sampled node for determination:

[0117] Spatial reachability constraint: Determine whether the load trajectory crosses the obstacle zone; if so, discard the node.

[0118] Action continuity constraint: Determine if it is satisfied If not, then discard the node; δ max This is the preset rate of change threshold.

[0119] Load dynamic stability constraint: Determine the load swing angle θ swing Does (t) satisfy |θ swing (t)|≤θ th If not, then discard the node; θ th The set angle threshold.

[0120] 4.2) The path cost function is a weighted combination of path length, load swing angle change, and action sequence change. The optimization objective is as follows:

[0121]

[0122] Where α, β, and γ are weighting coefficients, and PathLength is a path length function of A(t), based on the rotation angle θ in A(t). rot (t), the position of the car x car (t), hook height h hook (t) is obtained; specifically, based on the rotation angle, trolley position, hook height and combined with the structural parameter data of the crane, the path length function PathLength with respect to A(t) is obtained;

[0123] 4.3) Minimize the weighted sum J of path length, load swing angle change, and action sequence change to generate the endpoint node that satisfies the objective function optimally;

[0124] 4.4) After generating the optimal endpoint node that satisfies the objective function, the path sequence is extracted using the backtracking method of the random tree algorithm, and interpolation and trajectory smoothing are performed to finally generate the optimized action sequence:

[0125]

[0126] In this embodiment, the interpolation and trajectory smoothing processing methods are as follows:

[0127] a) The discrete trajectory node sequence generated by the fast random tree algorithm Each node A continuous-time trajectory function is constructed using cubic spline interpolation.

[0128] A(t) = a i (tt i ) 3 +b i (tt i ) 3 +c i (tt i ) 3 +d i

[0129] And it satisfies positional continuity and first derivative continuity to generate in the interval [t1, t2]. n The initial continuous trajectory function A(t) is defined on the [theory / parameter].

[0130] b) Based on the interpolated trajectory, further construct the trajectory smoothing optimization objective function:

[0131]

[0132] Where λ1 and λ2 are smoothing adjustment weight parameters; the first penalty term in the above formula

[0133] Used to suppress oscillations; second penalty item Used to suppress mutations;

[0134] c) Adjust the control points in the trajectory interpolation function using optimization methods (such as gradient descent, quasi-Newton method, etc.) to minimize the objective function and generate a smoothed motion trajectory sequence A. s (t), where the trajectory of each degree of freedom component is continuous at least on the second derivative; and the load swing angle corresponding to all trajectory points satisfies the safety threshold constraint: |θ swing (t)|≤θ th .

[0135] The above solution can significantly reduce load sway caused by path irregularities or oscillations while maintaining path feasibility, thereby improving the stability, path robustness and safety of crane operation.

[0136] 5) Control the crane's movement according to the optimized motion sequence, and monitor the load swing angle in real time through an inertial measurement unit. Trigger emergency braking when the swing angle exceeds a threshold; specifically including:

[0137] 5.1) Optimize the motion trajectory sequence

[0138] The input is sent to the crane control system, which drives the movement of the slewing mechanism, trolley mechanism and hook mechanism step by step according to the control cycle;

[0139] 5.2) The swing angle is collected in real time by an inertial measurement unit installed on the load or hook.

[0140] θ swing (t);

[0141] 5.3) Determine whether the swing angle meets the following safety threshold conditions:

[0142] |θ swing (t)|>θ th

[0143] Where, threshold θ th It can be determined according to the load weight W and the hook height h. hook Factors such as these are dynamically set.

[0144] If the absolute value of the swing angle exceeds the set threshold θ thIf the system issues an emergency braking command, it will immediately stop the operation of all drive mechanisms of the crane. The braking actions include stopping the trolley, locking the slewing mechanism, and braking the hook. An audible and visual alarm can also be triggered to alert on-site personnel to take appropriate action.

[0145] Example 2

[0146] See attached document Figure 2 As shown, this embodiment of the invention provides a crane work path planning system, which applies the crane work path planning method in Embodiment 1 above. The system includes:

[0147] The demonstration database module is used to store operation demonstration data from historical crane operation tasks. The data includes starting point, target point, environmental information, load weight, task complexity label, and corresponding operation trajectory data. The demonstration database module further includes a trajectory dimensionality reduction unit, which performs low-dimensional encoding on the trajectory data based on a variational autoencoder to build an action prototype library.

[0148] The path generation module includes a pre-trained lightweight neural network, which receives the starting point, target point, and environmental information of the current task and outputs the initial motion trajectory sequence of the three degrees of freedom of rotation, trolley, and hook. The path generation module integrates an environmental perception submodule, which receives the environmental grid map provided by LiDAR or visual sensor.

[0149] The local path optimization module is used to perform local replanning of the initial action trajectory based on the fast random tree (RRT) algorithm and combined with physical constraints. The optimization process uses the weighted sum of path length, load swing angle change and action sequence change as the optimization objective, and performs interpolation and trajectory smoothing on the optimized trajectory to ensure that the trajectory is continuous, executable and physically stable.

[0150] The execution control module is used to convert the smoothed trajectory sequence into control signals, which drive the crane's slewing mechanism, trolley mechanism and hook mechanism to perform operations according to the planned path.

[0151] The load monitoring module includes an inertial measurement unit (IMU) for real-time acquisition of the load's swing angle state and comparison of the acquired swing angle information with a set angle threshold.

[0152] The safety control module is used to prevent the load from swinging beyond a preset angle threshold θ. th Emergency braking is triggered when the situation arises; braking actions include stopping the trolley, locking the slewing mechanism, and braking the hook. An audible and visual alarm can also be triggered to alert on-site personnel to take appropriate measures.

[0153] Example 3

[0154] See attached document Figure 3As shown, based on the same inventive concept, this embodiment provides a computer-readable storage medium storing computer instructions that, when executed on a computer, cause the computer to perform the crane work path planning method as described in Embodiment 1.

[0155] In specific implementation, computer-readable storage media include: Universal Serial Bus flash drive (USB), portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and other storage media that can store program code.

[0156] The device embodiments described above are merely illustrative. The units / modules described as separate components may or may not be physically separate. The components shown as units / modules may or may not be physical units / modules; that is, they may be located in one place or distributed across multiple network units / modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0157] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0158] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for planning the working path of a crane, characterized in that, include: 1) Construct a crane operation demonstration database to store the starting point, target point, environmental information, and corresponding operator trajectory data from historical tasks; 2) Use a demonstration database to pre-train a convolutional neural network to extract the mapping relationship between environmental features and trajectory; 3) Input the current starting point position, target point position, and environmental information acquired by the lidar into the path generation network, and output the initial motion sequence of rotational degrees of freedom, trolley motion degrees of freedom, and hook degrees of freedom; 4) A fast random tree algorithm based on physical constraints is used to perform local replanning on the initial action sequence. The optimization criterion is the weighted sum and minimization of path length, load swing angle, and action sequence variation. 5) Control the crane movement according to the optimized action sequence and monitor the load swing angle in real time. When the swing angle exceeds the threshold, trigger emergency braking.

2. The crane working path planning method according to claim 1, characterized in that, Step 1) specifically includes: 1.1) Collect control trajectory data of operators in historical crane operation tasks, including control degree of freedom data of slewing mechanism, trolley traveling mechanism and hook lifting mechanism, and simultaneously record starting point, target point, environmental obstacle distribution information and load parameters; 1.2) Tagged historical tasks, marking the obstacle density level, lifting weight level and task complexity level for each task; 1.3) The collected trajectory data is processed by nonlinear dimensionality reduction using a variational autoencoder to generate a low-dimensional trajectory representation, and an action prototype library is constructed to represent the potential feature distribution of typical operation trajectories. 1.4) Store the basic task information, label information, original trajectory data and corresponding low-dimensional latent variable encoding results in the operation demonstration database for subsequent path generation network training and scheduling strategy calls.

3. The crane working path planning method according to claim 1, characterized in that, Step 2) specifically includes: 2.1) Construct input-output pairs from the data of each task sample, where: The input includes a starting point coordinate P s = (X s , y s , Z s ), a target point coordinate P g = (x g , y g , z g ), a two-dimensional environment grid map E(x, y) acquired by a laser radar, an obstacle density level D, a load weight W, and a task complexity C, wherein D, W, and C are used as structured features and image features after one-hot encoding. The output is the trajectory latent variable encoding Z = f VAE (T(t)), where T(t) denotes the expert trajectory, f VAE is a trained variational autoencoder; 2.2) Construct a convolutional neural network F(·) to map the above inputs to the trajectory latent variable outputs, satisfying the relationship: Z = F(E(x, y), P s P g (D, W, C); 2.3) The mean square error loss function L=||ZZ is adopted. gt || 2 Supervised training is performed on the network, where Z... gt The target value for expert trajectory coding; 2.4) After the network training is completed, the neural network can quickly generate the initial trajectory latent variables of the three-degree-of-freedom control action given any starting point, target point and environmental map data.

4. The crane working path planning method according to claim 1, characterized in that, Step 3) specifically includes: 3.1) Construct the task input vector, including the starting point position P. s =(x s y s , z s ), Target point location P g =(x g y g , z g ), a two-dimensional environmental grid map generated by lidar. E(x, y), and structured task label information L = {D, W, C}, where D is the obstacle density level, W is the lifting weight level, and C is the task complexity level; 3.2) The input vector is fed into a path generation neural network model consisting of a lightweight convolutional network and gated recurrent units, and the output is a time series of three-degree-of-freedom control actions: A(t)=[θ rot (t),x car (t),h hook (t)],t∈[0,T] Where, θ rot (t) is the rotation angle, x car (t) represents the position of the car. h hook (t) represents the hook height; 3.3) The action sequence output by the path generation neural network satisfies the following physical constraints: Limitations on the range of rotation angle, the range of movement of the trolley, and the lifting height of the hook.

5. The crane working path planning method according to claim 1, characterized in that, Step 4) specifically includes: 4.1) Based on the initial action sequence output in step 3), new sampling nodes are generated in the action space using a constraint-based fast random tree algorithm; Among them, the constraint-based fast random tree algorithm uses the initial action sequence as the guiding direction to perform action space sampling, in (θ rot， x car h hook Candidate nodes are generated in a three-dimensional space, and then the following physical constraints are applied to each sampled node for determination: Spatial reachability constraint: Determine whether the load trajectory crosses the obstacle zone; if so, discard the node. Action continuity constraint: Determine if it is satisfied If not, then discard the node; Load dynamic stability constraint: Determine the load swing angle θ swing Does (t) satisfy? |θ swing (t)|≤θ th If not, then discard the node; 4.2) The path cost function is a weighted combination of path length, load swing angle change, and action sequence change. The optimization objective is as follows: Where α, β, and γ are weighting coefficients, and PathLength is a path length function of A(t), based on the rotation angle θ in A(t). rot (t), position of the car x car (t), hook height h hook (t) is obtained; 4.3) Minimize the weighted sum of path length, load swing angle change, and action sequence change. J generates the endpoint node that satisfies the optimal objective function; 4.4) After generating the optimal endpoint node that satisfies the objective function, the path sequence is extracted using the backtracking method of the random tree algorithm, and interpolation and trajectory smoothing are performed to finally generate the optimized action sequence:

6. The crane working path planning method according to claim 5, characterized in that, The interpolation and trajectory smoothing processing methods in step 4.4) are as follows: a) The discrete trajectory node sequence generated by the fast random tree algorithm Each node A continuous-time trajectory function is constructed using cubic spline interpolation. A(t)=a i (t-t i ) 3 +b i (t-t i ) 3 +c i (t-t i ) 3 +d i And it satisfies positional continuity and first derivative continuity to generate in the interval [t1, t2]. n The initial continuous trajectory function A(t) is defined on the [theory / parameter]. b) Based on the interpolated trajectory, further construct the trajectory smoothing optimization objective function: Wherein, λ1 and λ2 are smoothing adjustment weight parameters; c) Adjust the control points in the trajectory interpolation function using optimization methods to minimize the objective function and generate a smoothed motion trajectory sequence A. s (t), where the trajectory of each degree of freedom component is continuous at least on the second derivative; and the load swing angles corresponding to all trajectory points satisfy the following condition. |θ swing (t)|≤θ th 。 7. The crane working path planning method according to claim 5, characterized in that, Step 5) specifically includes: 5.1) Optimize the motion trajectory sequence The input is sent to the crane control system, which drives the movement of the slewing mechanism, trolley mechanism and hook mechanism step by step according to the control cycle; 5.2) The swing angle θ is collected in real time by an inertial measurement unit installed on the load or hook. swing (t); 5.3) Determine whether the swing angle meets the following safety threshold conditions: |θ swing (t)||>θ th If the absolute value of the swing angle exceeds the set threshold θ th If the system issues an emergency braking command, it will immediately stop the operation of all drive mechanisms of the crane.

8. The crane working path planning method according to claim 7, characterized in that, The braking actions include trolley braking to a stop, slewing locking, and hook braking. An audible and visual alarm can also be triggered to alert on-site personnel to take appropriate measures.

9. A crane working path planning system, characterized in that, The system, employing the path planning method as described in any one of claims 1 to 8, comprises: The demonstration database module is used to store operation demonstration data from historical crane operation tasks. The data includes the starting point, target point, environmental information, load weight, task complexity label, and corresponding operation trajectory data. The demonstration database module further includes a trajectory dimensionality reduction unit, which performs low-dimensional encoding on the trajectory data based on a variational autoencoder to build an action prototype library. The path generation module includes a pre-trained lightweight neural network, which receives the starting point, target point, and environmental information of the current task and outputs the initial motion trajectory sequence of three degrees of freedom: rotation, trolley, and hook. The path generation module integrates an environmental perception submodule, which receives an environmental grid map provided by a lidar or visual sensor. The local path optimization module is used to perform local replanning of the initial action trajectory based on the fast random tree algorithm and combined with physical constraints. The optimization process uses the weighted sum of path length, load swing angle change and action sequence change as the optimization objective, and performs interpolation and trajectory smoothing on the optimized trajectory. The execution control module is used to convert the smoothed trajectory sequence into control signals to drive the crane's slewing mechanism, trolley mechanism and hook mechanism to perform operations according to the planned path; The load monitoring module includes an inertial measurement unit, which is used to collect the swing angle state of the load in real time and compare the collected swing angle information with the set angle threshold. The safety control module is used when the load swing angle exceeds a preset angle threshold θ. th Emergency braking is triggered when the situation arises; braking actions include stopping the trolley, locking the slewing mechanism, and braking the hook. An audible and visual alarm can also be triggered to alert on-site personnel to take appropriate measures.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the crane work path planning method as described in any one of claims 1 to 8.

Images