Construction method of crown block motion characteristic digital twinborn body

Abstract

The invention belongs to the field of digital twinning, and provides a construction method of a crown block motion characteristic digital twinning body, which comprises the following steps of: establishing a Lagrange classical analytical model according to a relation between crown block kinematics and dynamics, and acquiring crown block motion state data in a real working condition environment in real time; through finite element analysis (FEA) and an on-line sensor technology, crown block operation data are collected, a three-dimensional model is established, and simulation is carried out, so that dynamic characteristics such as displacement, speed and swing angle are obtained. And 3, carrying out standardization processing on the collected data, extracting typical motion characteristics by adopting clustering analysis, and calculating the average speed, displacement and acceleration of each cluster. And finally, constructing an equivalent model based on a radial basis function (RBF) neural network, and optimizing system parameters in combination with a particle swarm optimization (PSO) algorithm. The method has strong adaptability and global search capability, can effectively process a complex nonlinear system, and provides a new idea for real-time monitoring and optimization of crown block motion characteristics.

Description

Technical Field

[0001] The present invention belongs to the technical field of digital twins, and specifically relates to a method for constructing a digital twin of overhead crane motion characteristics. Background Art

[0002] Overhead cranes, also known as bridge cranes, are mainly used for lifting and transportation of large engineering projects. They are responsible for the transportation and weighing of large goods, such as containers. In order to meet the requirements of domestic enterprises for overhead crane operation efficiency and operation accuracy, innovation of overhead crane algorithms has become an imperative issue. However, building a digital twin of overhead crane lifting and transportation engineering is the core content of overhead crane algorithm innovation. At present, digital twin technology for overhead cranes is rarely used, and the use of digital twin technology can greatly save the actual industrial trial and error costs and improve the efficiency of algorithm innovation. Therefore, the research on digital twin technology provides new ideas and tools for improving the operation efficiency and operation accuracy of overhead cranes. Through digital twin technology, more intelligent and efficient overhead crane operation and management can be achieved, thereby promoting the development and progress of the industry. Summary of the invention

[0003] According to a first aspect of the present invention, the present invention claims a method for constructing a digital twin of the motion characteristics of an overhead crane, comprising:

[0004] S1. Establish the Lagrangian motion equation of the overhead crane hoisting system:

[0005]

[0006] Where m1 is the mass of the trolley, m2 is the mass of the load, l is the length of the wire rope, θ is the load swing angle, F is the input control force of the trolley, and F l is the axial tension of the wire rope, μ is the friction coefficient, g is the acceleration due to gravity, is the first-order differential of the car displacement x, is the second-order differential of the car displacement x, is the first-order differential of θ, is the second-order differential of θ, is the first-order differential of l, is the second-order differential of l.

[0007] In the process of building a digital twin, the Lagrangian equations of motion can be used as a benchmark for model validation. When the system collects actual operating data, this data can be used to verify the accuracy and effectiveness of the Lagrangian model. By comparing actual data with model predictions based on the Lagrangian equations, the digital twin can be continuously optimized to ensure that it accurately reflects the behavior of the actual system.

[0008] The overhead crane system is an under-actuated system, the trolley driving force F is used as the system input, and the output is the displacement and the swing angle of the hanging object. The state variables of the system are That is, the state equation of the overhead crane is:

[0009]

[0010] in is the four-dimensional state matrix of the system, X1 is the displacement x of the trolley, X2 is the first-order differential of x, X3 is the load swing angle θ, and X4 is the first-order differential of θ.

[0011] Furthermore, the state equation of the overhead crane is rewritten as:

[0012]

[0013] Where F is the control force of the system; f1(x) and q1(x) are the dynamic parts of the system.

[0014] f1(x), q1(x) expressions:

[0015]

[0016] S2. Obtaining overhead crane operation data through finite element analysis (FEA) and online methods

[0017] Specifically, the online method is to collect the mass of the crane and its trolley, the length of the wire rope, the swing angle, the speed and the displacement of the crane through different sensors when the crane is working, so as to ensure the real-time and accuracy of the data and provide basic data for the digital twin.

[0018] Finite element analysis uses 3D structural data and loading mass data to build a 3D model digital twin, and uses finite element simulation software to set the inherent properties of the crane material parameters, friction coefficient, etc. Subsequently, by establishing a physical field that simulates the real working environment, including wind impedance field, solid mechanics field, and motor response delay, finite element simulation is performed with the crane speed as input to obtain the crane displacement, speed, and swing angle data, and the online collected data is integrated with the finite element simulation data.

[0019] S3. Collect and organize the motion data of the crane. Assume that there is a data set D containing the following fields as the input of the crane digital twin:

[0020] D={(v1,x1,θ1,α1),(v2,x2,θ2,α2),...,(v n ,x n ,θ n ,α n )}

[0021] Among them, v is the velocity data, x is the displacement data, θ is the swing angle data, and α is the acceleration data.

[0022] Before clustering, it is usually necessary to standardize the data to eliminate the dimensional effects between different features. The method used is Z-score standardization:

[0023]

[0024] Where x′ is the standardized value, x is the original data value, μ is the mean, and σ is the standard deviation

[0025] To cluster data, randomly select j initial center points and assign each data point to the nearest center. The formula is:

[0026]

[0027] where d i is the data point, c j is the cluster center.

[0028] New cluster centers are calculated and the assignment and update steps are repeated until convergence.

[0029]

[0030] Clustering can be completed by analyzing the characteristics of each cluster to fuse information. For example, the average speed, displacement, angle, and acceleration of each cluster can be calculated to obtain the typical motion characteristics represented by each cluster:

[0031]

[0032] S4. Build a digital twin model of the overhead crane based on RBF neural network.

[0033] The RBF neural network includes an input layer, an intermediate layer and an output layer. The intermediate layer mainly calculates the radial basis function value of the Euclidean distance between the input quantity x and the sample quantity c, which has good normalization ability. It approximates any nonlinear function through a combination of multiple radial basis functions, and the output layer makes a linear combination of them.

[0034] The radial basis function is:

[0035]

[0036] Among them, h j is the radial basis function; x is the input layer vector, x = [x1, x2, x3] T ;c j is the neural network center vector c j =[c j1 , c j2 , cj3 ] T , j = 1, 2, ..., 6; is the width of the Gaussian basis function.

[0037] The output layer model is:

[0038]

[0039] Among them, y m (k) is the output value obtained after k training times; ω j is the weight value of the jth unit;

[0040] Construct the digital twin of the overhead crane and determine the input dimension of the RBF network. For each dimensional input, there is a corresponding adaptive idealized weight vector. The unknown items of the bridge crane dynamics can be expressed as follows:

[0041]

[0042] Where W T , Y T , P T , V T are weight matrices in the neural network, which are adjusted through training to adapt to the actual dynamics of the system. h1(x) is the output of the radial basis function, which is usually the input data after nonlinear transformation, capturing the complex nonlinear characteristics of the system. ε1, ε2, ε3, ε4 are error terms, representing the difference or noise between the model and the actual system.

[0043] The Particle Swarm Optimization (PSO) method is used to optimize the digital twin model of the overhead crane. The system parameters (i.e. the variables to be optimized) are represented as particles. Each particle has a position and velocity, and these particles are iteratively updated based on experience and global experience.

[0044] Randomly initialize the positions of a group of particles, that is, the initial parameter values ​​of each particle and initial velocity v i , set the historical optimal position of each particle and the global optimal position g best .

[0045] Define the objective function (i.e. the error function to be minimized), by which the error size of the parameter combination θ of each particle is evaluated:

[0046]

[0047] in,

[0048]

[0049] fNN (x i )=W T h1(x i ),

[0050] q NN (x i )=Y T h1(x i )

[0051] Each particle is based on its current position, historical optimal position and the global optimal position g best To update speed:

[0052]

[0053] Update the particle's position and look for a new solution in the parameter space, i.e., the system parameters:

[0054] θ i (t+1)=θ i (t)+v i (t+1)

[0055] If particle i is at the new position θ i If the fitness of (t+1) is better than its historical best position, update If the new position of particle i is better than the global optimum of all particles, update g best , iterate until the termination condition is reached (such as the fitness reaches a certain threshold or the maximum number of iterations is reached).

[0056] After PSO optimization, the key parameters of the system are identified by PSO method. Minimize the output error between the analytical model and the neural network model, and the optimal global position g best This is the optimal parameter estimate obtained:

[0057]

[0058] After PSO optimization, the key parameters of the system can be identified to minimize the output error between the analytical model and the neural network model. The optimal global position is the optimal parameter estimate obtained. This PSO-based method can handle high-dimensional parameter space and has the ability to search globally. It is particularly suitable for complex nonlinear systems, thereby minimizing the error between the model output and the actual system and improving the prediction accuracy of the digital twin.

[0059] The beneficial effects of the present invention are that it has strong adaptability and global search capability, can effectively process complex nonlinear systems, and provides a new idea for real-time monitoring and optimization of overhead crane motion characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flow chart of the method for constructing a digital twin model of an overhead crane according to the present invention;

[0061] Figure 2 This is the implementation method of the digital twin of the overhead travelling crane under the hardware carrier of the present invention; DETAILED DESCRIPTION

[0062] The specific implementation of the present invention is described below in conjunction with the accompanying drawings. It is described in the form of examples. It should be pointed out that the examples described below are intended to facilitate the understanding of the present invention and do not have any limiting effect on it.

[0063] Figure 1 The flowchart of the method for constructing a digital twin model of a crane in an embodiment of the present invention includes: first, the Lagrangian motion equation of the crane hoisting system is established. The state equation of the system is derived by analyzing the components of the system, including factors such as the trolley mass, the load mass, the length of the wire rope, and external disturbances. This process can accurately describe the dynamic characteristics of the system. A variety of sensors are used to collect operating data online, including the displacement, speed, acceleration, and swing angle of the trolley, to ensure that the acquired data is real-time and accurate. In order to eliminate the influence of different dimensions on data analysis, the collected data is subjected to Z-score standardization. This step allows different parameters to be compared under the same standard, which is convenient for subsequent analysis. Next, the typical motion characteristics of the system under different working conditions are extracted through cluster analysis technology, such as the key parameters of the acceleration stage, the uniform speed stage, and the deceleration stage. These characteristics can help identify the motion laws of the system under different working conditions and provide an important basis for the establishment of the model. Subsequently, a motion equivalent model based on a radial basis function (RBF) neural network is constructed. Due to its powerful nonlinear mapping ability, the RBF neural network can effectively fit the dynamic behavior of complex systems. During the model training process, the extracted motion features are used as input, and supervised learning is performed using historical operation data to improve the prediction accuracy of the model. In order to further improve the performance of the model, the particle swarm optimization (PSO) algorithm is used to globally optimize the parameters of the neural network. The PSO algorithm finds the optimal solution by simulating the flight of particles in multidimensional space, thereby minimizing the error between the analytical model and the output of the neural network. This optimization process not only improves the accuracy of the model, but also enhances its adaptability to complex nonlinear systems. Finally, through the method of this embodiment, high-precision motion control of the overhead crane hoisting system is achieved, significantly improving operating efficiency and safety. This method is applicable to various industrial automation fields, especially in complex nonlinear systems that require high-precision control, and has broad application prospects.

[0064] Figure 2This is an embodiment of the present invention: This embodiment uses hardware to optimize the digital twin model of the overhead crane control system. The system consists of three parts: a personal computer (PC), a programmable logic controller (PLC) and a digital twin hardware carrier embedded with a machine learning algorithm. The PC sends the actual overhead crane operation status training set (including target position, real-time position, swing angle parameters, etc.), PLC initialization status parameters and overhead crane loading data parameters to the PLC S-1200 controller via the Labview host computer in the TCP / IP protocol. The PLC S-1200 controller uses the internal DP function block to pass the training set to the digital twin hardware module through the PROFIBUS-DP protocol. The module includes but is not limited to using control chips such as MCU, MPU, FPGA or application-specific integrated circuits (Application Specific Integrated Circuit, ASIC) as hardware carriers, thereby predicting the overhead crane position curve. The predicted curve is returned to the PC via the same communication method for system status display, displacement curve fitting and error calculation. At the same time, the PC sends the learning rate and learning weight to the digital twin hardware module through the serial port to achieve real-time model update. Through this iterative process, the optimized overhead crane digital twin model is finally obtained.

Claims

1. A method for constructing a digital twin of overhead crane motion characteristics, characterized in that: The following steps are involved: S1. Establish the Lagrangian motion equation of the overhead crane hoisting system: Where m1 is the mass of the trolley, m2 is the mass of the load, l is the length of the wire rope, θ is the load swing angle, F is the input control force of the trolley, and F l is the axial tension of the wire rope, μ is the friction coefficient, g is the acceleration due to gravity, is the first-order differential of the car displacement x, is the second-order differential of the car displacement x, is the first-order differential of θ, is the second-order differential of θ, is the first-order differential of l, is the second-order differential of l; The driving force F of the trolley is taken as the input of the system, and the output is the displacement and the swing angle of the hanging object. Define the state variables of the system The state equation of the overhead crane is: Among them, X1 is the displacement x of the trolley, X2 is the first-order differential of x, X3 is the load swing angle θ, and X4 is the first-order differential of θ; The state equation of the overhead crane is rewritten as: The expressions of f1(x), q1(x), f2(x), q2(x) are: S2. Obtaining operation data through finite element analysis and online methods, specifically: collecting the mass of the overhead crane and trolley, the length of the wire rope, the swing angle, the speed and the displacement of the crane through sensors; S3, perform Z-score standardization on the collected data, and extract the typical motion characteristics of the system under different working conditions through cluster analysis technology; S4. Construct a digital twin model of an overhead crane based on an RBF neural network; the RBF neural network includes an input layer, an intermediate layer, and an output layer, the intermediate layer is used to calculate the radial basis function value of the Euclidean distance between the input x and the sample c, and approximate any nonlinear function through a combination of multiple radial basis functions, and the output layer makes a linear combination of them; The radial basis function is: Among them, h j is the radial basis function; x is the input layer vector, x = [x1, x2, x3] T ;c j is the neural network center vector c j =[c j1 , c j2 , c j3 ] T ,j=1,2,...,6; is the width of the Gaussian basis function; The output layer model is: Among them, y m (k) is the output value obtained after k training times; ω j is the weight value of the jth unit; The method of constructing the digital twin of the overhead crane is to determine the input dimension of the RBF network. For each dimension of input, there is a corresponding adaptive idealized weight vector. The unknown term of the bridge crane dynamics is calculated by the following formula: Where W T , Y T , P T , V T are weight matrices in the neural network, which are adjusted through training to adapt to the actual dynamics of the system; h1(x) is the output of the radial basis function, which is the input data after nonlinear transformation and is used to capture the complex nonlinear characteristics of the system; ε1, ε2, ε3, ε4 are error terms, representing the difference or noise between the model and the actual system; Finally, the overhead crane digital twin model is optimized by particle swarm optimization method to construct a digital twin.

2. The method for constructing a digital twin of the motion characteristics of an overhead crane according to claim 1, characterized in that: The specific method of optimizing the digital twin model of the overhead crane using the particle swarm optimization method is as follows: Randomly initialize the positions of a group of particles, that is, the initial parameter values ​​of each particle and initial velocity v i , set the historical optimal position of each particle and the global optimal position g best ; Define the objective function, through which the error of the parameter combination θ of each particle is evaluated as: in, f NN (x i )=W T h1(x i ), q NN (x i )=Y T h1(x i ) Each particle is based on its current position, historical optimal position and the global optimal position g best To update the speed: Update the particle's position and look for a new solution in the parameter space, namely the system parameters: θ i (t+1)=θ i (t)+v i (t+1) If particle i is at the new position θ i If the fitness of (t+1) is better than its historical best position, update If the new position of particle i is better than the global optimum of all particles, update g best , iterate until the termination condition is reached; After optimization by particle swarm optimization, the key parameters of the system are identified by particle swarm optimization method. Minimize the output error between the analytical model and the neural network model, and the optimal global position g best This is the optimal parameter estimate: After optimization by particle swarm optimization method, the key parameters of the system are identified to minimize the output error between the analytical model and the neural network model.

Images