Multi-motor cooperative control system and method based on digital twinning

Abstract

The invention relates to the technical field of multi-motor cooperative control, and discloses a multi-motor cooperative control system and method based on digital twinning. The multi-motor cooperative control method based on digital twinning comprises the following steps: acquiring real-time operation data of each motor in a multi-motor system, and generating a physical parameter data set containing time sequence information; based on the physical parameter data set, calculating digital twinborn model parameters of each motor, and generating a motor digital twinborn body; analyzing a topological connection relationship between the motors, and generating a cooperative control constraint condition; and inputting the real-time state data of the motor digital twin and the cooperative control constraint condition into a cooperative optimization algorithm, and outputting a multi-motor cooperative control instruction. According to the method, the problem that accurate digital mapping between a physical motor and a control system is lacked in a traditional method is solved, and the technical problem that a cooperative control strategy cannot be optimized in real time is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of multi-motor cooperative control, and more particularly to a multi-motor cooperative control system and method based on digital twinning. BACKGROUND

[0002] In an intelligent manufacturing line, multiple motors need to be accurately coordinated to complete complex processing tasks, and there are strict requirements for speed matching, torque balance and phase synchronization between motors. Traditional multi-motor control systems mainly rely on pre-set control parameters and simple feedback adjustment, lacking accurate modeling of physical motor states and real-time prediction capabilities.

[0003] Specifically, the traditional multi-motor control has the following technical problems: there is a lack of high-precision digital mapping between physical motors and control systems, which makes the control system unable to accurately grasp the real-time running state of each motor; the cooperative control strategy between multiple motors is mainly based on empirical parameters, and cannot be dynamically optimized according to actual working conditions; when the system topology changes (such as motor failure, new equipment access), the control system is difficult to quickly adapt to the new system structure, resulting in a decrease in cooperative control accuracy and a lag in response. SUMMARY

[0004] The present application provides a multi-motor cooperative control system and method based on digital twinning, which overcomes the problem of lack of accurate digital mapping between physical motors and control systems in traditional methods, and solves the technical problem of real-time optimization of cooperative control strategies.

[0005] The present application provides a multi-motor cooperative control method based on digital twinning, comprising: obtaining real-time running data of each motor in a multi-motor system, generating a physical parameter data set containing time sequence information; based on the physical parameter data set, calculating the digital twinning model parameters of each motor, generating a motor digital twin; analyzing the topological connection relationship between the motors, generating cooperative control constraint conditions; inputting the real-time state data of the motor digital twin and the cooperative control constraint conditions into a cooperative optimization algorithm, outputting multi-motor cooperative control instructions; wherein the cooperative optimization algorithm solves a multi-objective optimization problem containing tracking error, power consumption and cooperative deviation, and generates control parameters that minimize the multi-objective function as the multi-motor cooperative control instructions.

[0006] Further, the physical parameter data set includes the rotational speed, current, temperature and vibration acceleration parameters of each motor, and the parameters are standardized, normalized to a specific interval or standardized to a distribution with a mean of 0 and a variance of 1, eliminating the magnitude difference between different physical quantities.

[0007] Further, the generation of the motor digital twin includes the following sub-steps:

[0008] The parameters of the electromagnetic equation of the motor are fitted by using the least square method, and the resistance, inductance and back electromotive force coefficient of the motor are calculated according to the current-speed data;

[0009] The mechanical characteristic parameters of the motor, including the moment of inertia, friction coefficient and load characteristic, are extracted by using the frequency domain analysis method;

[0010] The thermodynamic model is established based on the temperature-power data, and the parameters such as heat capacity and thermal resistance are calculated;

[0011] The parameters are integrated to form a complete digital twin model of the motor.

[0012] Further, the time-varying characteristics of the motor parameters are learned by using a recurrent neural network, and the historical parameter sequence is input into the recurrent neural network to output a dynamic correction value of the parameter, so as to realize adaptive updating of the digital twin of the motor.

[0013] Further, the cooperative control constraint conditions include:

[0014] A speed synchronization constraint is used to limit the speed deviation between different motors;

[0015] A torque balance constraint is used to ensure that the sum of the torques of the multi-motor system is balanced with the load torque;

[0016] A phase coordination constraint is used to specify the phase difference range between different motors.

[0017] Further, the objective function of the multi-objective optimization problem includes three normalized weighted indicators: the sum of the tracking errors of each motor, the power consumption of each motor, and the cooperative deviation between the motors, and the multi-objective optimization problem is solved by using the gradient descent method.

[0018] Further, the digital twin of the motor is used for forward simulation, the current state and the candidate control strategy are input into the digital twin model of the motor, the system response in the future multiple time steps is predicted, and the control strategy that optimizes the predicted performance index is selected.

[0019] Further, the predicted performance index is defined as a time-weighted multi-objective function, including the weighted sum of the tracking error, power consumption and cooperative deviation in the future multiple time steps, and each indicator is normalized to ensure comparability.

[0020] Further, the topology change event of the motor system is monitored in real time, and when a motor fault or new equipment access is detected, the manifold interpolation algorithm is used to quickly update the topology relationship matrix and recalculate the cooperative control constraint conditions.

[0021] Further, a state deviation of the physical motor and the motor digital twin is calculated, a fault diagnosis result is generated, a fault is determined when the deviation exceeds a threshold, a fault type and position are located through deviation pattern recognition, and a collaborative control strategy is automatically adjusted to compensate for the fault influence.

[0022] The application provides a multi-motor collaborative control system based on digital twinning, comprising:

[0023] A data acquisition module is configured to acquire real-time operation data of each motor in the multi-motor system.

[0024] A digital twin modeling module is configured to generate a motor digital twin based on the data.

[0025] A topology analysis module is configured to analyze the topology connection relationship between the motors.

[0026] A collaborative optimization module is configured to calculate control instructions based on the state of the digital twin and constraint conditions.

[0027] A predictive control module is configured to perform forward-looking simulation.

[0028] A topology adaptation module is configured to handle system topology changes.

[0029] A fault diagnosis module is configured to identify and handle system faults.

[0030] The application has the following advantages:

[0031] The multi-motor collaborative control method based on digital twinning provided by the application overcomes the problem of lack of accurate digital mapping between physical motors and control systems in traditional methods, as a high-fidelity motor digital twin of the physical motor is established to realize real-time mapping of physical parameters to digital models.

[0032] Further, predictive control is achieved through forward-looking simulation of the motor digital twin, which can evaluate the effect of the control strategy in advance; topology changes are quickly adapted through a manifold interpolation algorithm, ensuring the control performance of the system in a dynamic changing environment; fault diagnosis and fault-tolerant control are achieved through virtual-real comparison, improving the reliability of the system. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 is a flowchart of a multi-motor collaborative control method based on digital twinning of the application;

[0034] Figure 2 is a multi-motor system real-time operation parameter monitoring diagram of the application;

[0035] Figure 3 is a motor current-speed characteristic analysis scatter plot of the application;

[0036] Figure 4 is a motor temperature rise thermodynamic model verification diagram of the application;

[0037] Figure 5 is a printer motor system topology connection relationship network diagram of the application;

[0038] Figure 6 is a printer control performance comparison diagram of the application;

[0039] Figure 7 is a fault diagnosis effect evaluation diagram of the application; DETAILED DESCRIPTION

[0040] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that discussions of these implementations are intended to serve as examples only and that changes to elements of the examples discussed can be made in light of the teachings provided herein without departing from the scope of the specification. Various examples can omit, substitute, or add various procedures or components in addition to those described or in lieu thereof. Also, many modifications can be made to the examples described herein, which have been presented for the purposes of example and illustration only.

[0041] In at least one embodiment of the application, a multi-motor cooperative control method based on digital twinning is disclosed, as shown in Figures 1-7 The method comprises the following steps:

[0042] Step 100: Obtain real-time running data of each motor in the multi-motor system, and generate a physical parameter data set.

[0043] Obtain the speed, current, temperature, vibration, etc. of each motor in the multi-motor system, and collect the data at a frequency not less than 1 kHz. Organize the collected data according to the time stamp and motor identification, and generate a physical parameter data set containing time sequence information.

[0044] It should be noted that the data structure of the physical parameter data set is as follows: , wherein represents the parameter vector of the th motor at time , is the speed, is the current, is the temperature, is the vibration acceleration, is the total number of motors, is the sampling time sequence.

[0045] Before applying this physical parameter dataset containing time-series information, data preprocessing is required. This includes standardizing parameters of different dimensions, such as rotational speed, current, temperature, and vibration acceleration, and normalizing each parameter to a specific value. The interval is standardized to a distribution with a mean of 0 and a variance of 1 to eliminate the difference in magnitude between different physical quantities; outliers are detected and filtered out, using a 3D model. Outliers are identified using criteria or local density estimation methods; missing data are filled using time series interpolation to ensure data continuity and integrity.

[0046] Step 200: Based on the physical parameter dataset containing timing information, calculate the digital twin model parameters of each motor and generate a digital twin of the motor.

[0047] The physical parameter dataset containing timing information is input into the motor modeling algorithm. The digital model parameters of each motor, including electromagnetic, mechanical and thermodynamic parameters, are calculated through the fusion of physical modeling and data-driven methods, generating a digital twin of the motor that can reflect the dynamic characteristics of the motor.

[0048] It should be noted that the generation of a digital twin of a motor includes the following sub-steps:

[0049] Step 201: Fit the electromagnetic equation parameters of the motor using the least squares method, and calculate the motor's resistance, inductance, and back electromotive force coefficient based on the current-speed data;

[0050] The input to the aforementioned least squares method is the set of current-speed data pairs collected during motor operation. ,in For the first The current value at each sampling time. For the corresponding rotational speed value, The number of sampling points; the output is the electromagnetic equation parameter vector. ,in For resistance, For inductance, The back electromotive force coefficient; by constructing a system of linear equations And solve The parameter estimates are obtained.

[0051] Furthermore, the process of constructing the linear equation system is as follows: based on the voltage equations of the motor. For each sampling point Constructing a matrix row vectors sum vector elements Rate of change of current Calculated using the central difference method, i.e. where is the sampling interval; angular velocity is obtained by converting the rotational speed, i.e. (when is expressed in r / min); the complete linear equation set is expressed as , To improve the accuracy of parameter estimation, the collected data should cover multiple operating points of the motor, the sampling frequency should be at least 1 kHz, and the sampling time should be no less than 10 seconds.

[0052] Step 202: Extract the mechanical characteristic parameters of the motor, including the moment of inertia, the friction coefficient, and the load characteristic, by the frequency domain analysis method;

[0053] The input of the aforementioned frequency domain analysis method is the time series data of the motor speed response and the corresponding driving torque sequence ; the output is a set of mechanical characteristic parameters where is the moment of inertia, is the friction coefficient, is the load torque; the time domain signal is converted to the frequency domain by fast Fourier transform, and the transfer function is calculated, and the mechanical parameters are extracted according to the amplitude-frequency characteristic and the phase-frequency characteristic of the transfer function.

[0054] Further, the specific steps of mechanical parameter extraction are as follows: first, pre-process the collected speed sequence and the torque sequence , including removing the direct current component and applying the Hanning window to reduce spectral leakage; then calculate the fast Fourier transform of each sequence and ; according to the dynamic equation of the mechanical system , it is converted in the frequency domain as:

[0055] ;

[0056] Calculate the amplitude-frequency characteristic:

[0057] and the phase-frequency characteristic ; by fitting the amplitude-frequency and phase-frequency characteristic curves, calculate the moment of inertia and the friction coefficient ; the load torque is estimated by analyzing the low-frequency response or steady-state error. To improve the accuracy of parameter identification, the motor should be tested under no-load and load conditions respectively, and the excitation signal should contain enough frequency components, and the pseudo-random binary sequence (PRBS) is recommended as the input excitation.

[0058] Step 203: Establishing a thermodynamic model based on temperature-power data, calculating heat capacity, thermal resistance, etc.

[0059] The input of the aforementioned thermodynamic model establishment method is a sequence of temperature-power data pairs during motor operation wherein is the loss power value at the th sampling time, is the corresponding motor temperature value, is the ambient temperature, is the number of sampling points; the output is the thermodynamic parameter wherein is the heat capacity, is the thermal resistance; by solving the heat balance equation:

[0060] ,

[0061] and combining the least squares method to optimize the objective function to obtain the parameter estimate value, wherein is the model predicted temperature value.

[0062] Further, the specific implementation steps of thermodynamic parameter identification are: first, obtain the motor loss power by measuring the difference between the input electric power and the output mechanical power, i.e. wherein and are the motor voltage and current, respectively, and are the output torque and angular velocity, respectively; use high-precision temperature sensors (such as PT100, thermocouple) to measure the temperature of different parts of the motor, focusing on the stator winding, rotor and shell temperature; discretize the heat balance equation to obtain:

[0063] ,

[0064] wherein is the sampling time interval; express the temperature prediction value as a function of the thermodynamic parameters, construct the least squares problem, and solve the parameters and so that is minimized; to accurately identify the parameters, the data collected should include the heating and cooling processes, and the test time should cover at least 3 times the motor thermal time constant (usually 10 minutes to 1 hour). The measured heat capacity typical value range is J / K, thermal resistance typical value range is K / W, the actual value depends on the motor size and cooling conditions.

[0065] Step 204: integrate the above parameters to form a complete motor digital twin model:

[0066] wherein is the resistance, is the inductance, is the back electromotive force coefficient, is the moment of inertia, is the friction coefficient, is the heat capacity, is the thermal resistance. Wherein, the function represents the state update function of the motor digital twin, and is specifically defined as:

[0067]

[0068] The state update function of the motor digital twin obtains the motor speed , current , temperature and torque in the future time interval by solving the motor state equation set. The state equation set includes:

[0069] Electrical equation ,

[0070] Mechanical equation ,

[0071] and thermodynamic equation ,

[0072] wherein is the angular velocity, is the input voltage, is the load torque, is the loss power, is the ambient temperature.

[0073] In the embodiments of the present application, in order to improve the accuracy of the motor digital twin, step 205 is further included on the basis of step 200: learning the time-varying characteristics of the motor parameters by using a recurrent neural network, inputting the historical parameter sequence into the recurrent neural network, outputting the dynamic correction value of the parameter, and realizing adaptive update of the motor digital twin.

[0074] The input layer of the aforementioned recurrent neural network receives the motor historical parameter sequence, which is defined as wherein represents the motor parameter vector at time , and is the time window length; in order to eliminate the dimensional differences between different motor parameters, the parameter sequence is standardized before being input into the neural network, and the transformation where and are the mean and standard deviation of historical data, respectively; the recurrent neural network contains a hidden layer of 128 LSTM units, and the state update equation of each LSTM unit is

[0075]

[0076]

[0077]

[0078]

[0079]

[0080]

[0081] where , , are the forget gate, input gate and output gate, respectively, is the cell state, is the hidden state, and are the learnable parameters, is the sigmoid activation function.

[0082] The output layer of the aforementioned recurrent neural network is a linear mapping layer that maps the hidden state at the last time step to the parameter correction value: where is the correction amount of each parameter, and are the output layer parameters. The output parameter correction value needs to be de-normalized to convert it back to the original physical quantity dimension, i.e. where and are the standard deviation and mean vector of the corresponding parameter in the training data.

[0083] The aforementioned recurrent neural network adopts a supervised learning training mode, with the parameter changes in the historical period as the training data. The loss function during training is defined as:

[0084]

[0085] where is the true parameter value of the th sample, is the predicted value based on the physical model, is the number of samples, This is the correction value for the neural network output. is the regularization coefficient, used to limit the magnitude of the correction; the optimization strategy of the recurrent neural network adopts the Adam optimizer, the initial learning rate is set to 0.001, the learning rate decreases by 0.9 times every 1000 epochs, the batch size is set to 32, and the training iteration is 10000 epochs until the loss function converges.

[0086] Furthermore, the training data acquisition method for recurrent neural networks is as follows: sample data is collected through specially designed experimental conditions, including step response, ramp response, and random excitation response conditions; true parameter values... Parameters are obtained through offline parameter identification methods, including Extended Kalman Filter (EKF), Particle Filter (PF), or Maximum Likelihood Estimation (MLE). The training dataset contains at least 1000 samples, including parameter variation sequences under different working conditions and environmental conditions. Each sample contains a length of [length missing]. The parameter sequence within the time window, Typical values ​​are 20-100 time steps; training, validation, and test sets are divided in a 7:1:2 ratio; regularization coefficients... The range of values ​​is The optimal value is determined through cross-validation; to prevent overfitting, in addition to L2 regularization, Dropout technology is also applied with a Dropout rate of 0.2; the relative error of parameter prediction on the test set after training should be less than 5%, otherwise the model structure needs to be adjusted or training data needs to be recollected.

[0087] Step 300: Analyze the topological connection relationship between motors and generate cooperative control constraints.

[0088] Based on the physical connection method and process requirements of the multi-motor system, the coupling relationship between the motors is analyzed, including mechanical coupling, electrical coupling and control coupling, and a set of constraints describing the coordination requirements between the motors is generated.

[0089] It should be noted that the constraints of coordinated control include:

[0090] Speed ​​synchronization constraints ,

[0091] Torque balance constraint ,

[0092] Phase Coordination Constraints ,

[0093] in For speed deviation threshold, For the first The output torque of the motor The total load torque, For the first Phase of the tekt motor, For motor and The expected phase difference between them This is the phase deviation threshold.

[0094] Furthermore, the speed deviation threshold The value is determined based on process requirements and motor type. For precision machining applications, the range is 0.1% to 1% of the rated speed. ,in The rated speed of the motor; phase. The range of values ​​is Desired phase difference Depending on the process requirements and the motor's relative position in the system, its value range is... Phase deviation threshold For high-precision synchronous control systems, the degree of change is generally no more than 5 degrees. .

[0095] Step 400: Input the digital twin state and cooperative control constraints into the cooperative optimization algorithm, and output multi-motor cooperative control commands.

[0096] The real-time status data of each motor digital twin and the constraints generated in step 300 are input into the collaborative optimization algorithm. By solving the multi-objective optimization problem, the optimal control parameters that satisfy the collaborative constraints are calculated, and control commands for each motor are generated.

[0097] It should be noted that the objective function of the collaborative optimization algorithm is:

[0098]

[0099] To ensure the comparability of different physical quantity terms in the objective function, the tracking error is... Power consumption and cooperational bias Normalization is performed: , , ,in , and These represent the maximum expected value or historical maximum value for each item. The objective function is rewritten as:

[0100]

[0101] in For the first The control input of the motor, To track errors, For power consumption, for the motor and a cooperative deviation, and are weight coefficients. The multi-objective optimization problem is solved by gradient descent method to obtain the optimal control instructions of each motor.

[0102] Further, the tracking error is defined as the difference between the actual speed and the operating target speed, i.e. where is the target speed, which is set according to the production task; the time dimension in the objective function is represented by a discrete time series, and each time step corresponds to a target function value:

[0103]

[0104] The control period is (typically 1ms to 100ms, depending on the system dynamic characteristics and control requirements), and the control input generated in each control period remains unchanged, i.e. .

[0105] Further, the cooperative deviation is defined as:

[0106]

[0107] where and are the weight coefficients of speed deviation and phase deviation, respectively, taking values of and ; the weight coefficients and are dynamically adjusted according to the application scenario and control target, the value range of , the value range of , for high energy efficiency requirements, increase the value of , for high cooperative accuracy requirements, increase the value of .

[0108] The input of the aforementioned gradient descent method is the initial control parameters and the objective function ; the output is the optimized control parameters ; the iterative update formula is:

[0109]

[0110] where is the learning rate, For the gradient of the objective function, iterate until or reach the maximum number of iterations. The control parameters obtained after optimization need to be converted into actual executable control instructions, the conversion process includes: parameter legality verification, ensuring that the control parameters are within the rated range of each motor; instruction quantization, converting continuous value control parameters into discrete control signals; timing arrangement, generating a control instruction sequence with precise timestamps; control protocol conversion, formatting the instructions into the communication protocol format supported by the specific motor driver.

[0111] Further, the gradient calculation method is: the gradient of the objective function with respect to the control parameters is calculated by numerical differentiation method:

[0112]

[0113] Where each component is calculated by the central difference formula:

[0114]

[0115] is the unit vector in the th dimension, is the small perturbation, the typical value is ; the learning rate is set to 0.01 initially, and when the objective function value decreases for 3 consecutive iterations, it is multiplied by 1.1 to increase, and when the objective function value increases, it is multiplied by 0.5 to decrease; the convergence threshold is set to , and the maximum number of iterations is set to 100; to prevent local optimal solution, a multi-start strategy is used, starting optimization from different initial points, and selecting the solution with the smallest final objective function value; parameter legality verification ensures , where and are the minimum and maximum values of the th motor control input, usually ±20% of the rated value.

[0116] Further, to ensure the consistency of control and time constraints, the time constraints of the gradient descent method are supplemented: the optimization process requires to be completed within the predetermined system reaction time , the typical value range is , and when running on a real-time system, the calculation buffer time of each control period is set to ; the actual execution time of optimization iteration must satisfy This means that optimization calculations are completed within the effective calculation time of each control cycle to ensure that control commands can be sent to the motor driver before the start of the next control cycle.

[0117] In this embodiment of the application, in order to achieve predictive control, step 401 is further included in addition to step 400: performing forward simulation using a motor digital twin, inputting the current state and candidate control strategies into the motor digital twin model to predict the future. The system response at each time step is used to select the control strategy that optimizes the predictive performance index.

[0118] The aforementioned predictive performance metric is defined as a time-weighted multi-objective function:

[0119]

[0120] To ensure the comparability of different physical quantities in the predicted performance indicators, each indicator is normalized:

[0121] Tracking error ,

[0122] Power consumption ,

[0123] Coordination bias ,

[0124] in For reference error value, This refers to the rated power of the motor. This represents the maximum permissible collaborative deviation.

[0125] Furthermore, the time step in the predictive performance metrics This represents the discrete time interval predicted by the model, and its value is related to the control period. Equal or Integer multiples of, typically ranging from ; At the current time point, Indicates the future number There are 1 prediction time points, and the entire prediction time domain is 1. Predicting the length of the time domain Typically, this is 1-10 times the system's main dynamic time constant to ensure the effectiveness of the prediction. At each current time point... Predicted performance indicators The control decisions at this moment have an impact on the future. The influence of each time step enables the control system to anticipate potential error accumulation and changes in system state.

[0126] in As a time discount factor, For the first tracking error of the motor at the next time step, corresponding power consumption, cooperative deviation between the motor and , and , , and are weight coefficients. By sampling multiple candidate control strategies in the control space , the prediction performance indicators of each strategy at the current time are calculated, and the control strategy that minimizes is selected as the final control decision at the current time . Further, the time discount factor

[0127] is 0.9, used to balance short-term and long-term control effects; the weight coefficients , and correspond to the weights of the tracking error, power consumption and cooperative deviation, respectively, and their value ranges are , , .

[0128] and satisfy the normalization condition ; the prediction step number is determined according to the dynamic characteristics of the system, and generally has a value range of , a smaller value for a system with fast response speed and a larger value for a system with slow response speed.

[0129] In the embodiments of the present application, in order to cope with the dynamic changes of the system topology, step 301 is further included on the basis of step 300: real-time monitoring of the topology change event of the motor system, when detecting motor failure or new device access, using the manifold interpolation algorithm to quickly update the topology relationship matrix and recalculate the cooperative control constraint condition.

[0130] The aforementioned manifold interpolation algorithm includes the following steps:

[0131] Step 301-1: Obtain the adjacency matrix before the topology change and the target topology structure after the change , wherein the element of the adjacency matrix represents the connection strength between the motor and , and the information of the newly added or deleted nodes is contained;

[0132] ​​Step 301-2: Map the adjacency matrix to the manifold space by eigenvalue decomposition to get the eigenvector matrix and the eigenvalue matrix , construct the manifold coordinates ;

[0133] Step 301-3: Calculate the interpolation path in the manifold space according to the topology change type. If a node is deleted, remove the corresponding row and column and re-orthogonalize. If a node is added, minimize the energy function:

[0134] Calculate the manifold coordinates of the new node , where is the neighbor set of the new node;

[0135] Further, when a node is added, the energy function minimization uses the gradient descent method, and the gradient calculation formula is:

[0136] ,

[0137] The iterative update formula is , where is the step size, with a value range of , and the initial value is set as the average of the neighbor node coordinates; the iteration termination condition is or the maximum number of iterations (usually 100 times) is reached; when a node is deleted, remove the corresponding row and column, and then orthogonalize the remaining eigenvector matrix by QR decomposition method to ensure that the modified matrix still satisfies the orthogonality requirement . After adding or deleting a node, the manifold coordinates need to be updated, and the adjacency matrix is reconstructed through inverse mapping .

[0138] Step 301-4: Generate intermediate topology sequences along the interpolation path , where , and ensure the positive definiteness and connectivity of each intermediate matrix through spectral analysis;

[0139] The aforementioned spectral analysis method includes calculating the eigenvalues of each intermediate matrix , verifying whether it satisfies the positive definiteness condition and the connectivity condition (the second smallest eigenvalue of the Laplacian matrix is greater than zero, indicating that the graph is connected); if not, modify (where is a small positive number, is or adjust the interpolation weight to regenerate the intermediate matrix until the condition is met.

[0140] Further, the positive definiteness correction parameter is in the range of , which is dynamically adjusted according to the condition number of the matrix, the larger the condition number, the closer the value to the upper limit; the connectivity determination threshold is set to , to ensure that the connectivity of the graph has sufficient numerical stability; when the number of intermediate points of the interpolation path is , the value sequence of the path parameter is , , the value range of depends on the complexity of the topological change.

[0141] Step 301-5: output the updated topological relationship matrix and the corresponding collaborative control constraint condition update parameter .

[0142] In the embodiments of the present application, in order to realize fault diagnosis and fault-tolerant control, the following steps are further included:

[0143] Step 500: calculate the state deviation between the physical motor and the motor digital twin, and generate a fault diagnosis result.

[0144] Compare the actual measured motor state with the motor digital twin predicted state, and when the deviation exceeds the threshold, determine that there is a fault, locate the fault type and position through deviation pattern recognition, and automatically adjust the collaborative control strategy to compensate for the fault impact.

[0145] The aforementioned deviation pattern recognition method is realized based on feature vector extraction and fault pattern classification, and specifically includes: constructing a state deviation vector , where each component is the speed deviation, current deviation, temperature deviation and torque deviation; due to the different dimensions and orders of magnitude of different physical quantities, the deviation vector is normalized, and is used, where , , and are the nominal speed, current, temperature and torque values of the motor, respectively;

[0146] Further, the fault determination threshold is set to: the speed deviation threshold , the current deviation threshold , the temperature deviation threshold , and the torque deviation threshold ; when any deviation component exceeds the corresponding threshold and the duration exceeds the preset determination time , the value range of which is , the fault diagnosis process is triggered according to the application scenario (determined by the application scenario). Calculate the statistical characteristics of the standardized deviation vector , wherein is the mean, is the standard deviation, is the skewness, is the kurtosis, is the autocorrelation coefficient; using a support vector machine classifier Map the feature vector to a predefined set of fault categories , wherein represents the class of fault mode (such as open circuit fault, short circuit fault, sensor fault, etc.), and the decision function of the classifier is:

[0147]

[0148] , wherein is the kernel function, and are model parameters, is the training sample label, is the number of support vectors.

[0149] Further, the training method of the support vector machine classifier is: first, construct a fault sample library, simulate various typical faults by experimental methods, including motor open circuit fault, short circuit fault, bearing fault, sensor fault, etc. Collect at least 100 sets of state deviation data samples for each type of fault;

[0150] The kernel function uses a radial basis function (RBF):

[0151] ,

[0152] , wherein is the kernel parameter, and the optimal value is determined by grid search, with a search range of ; the value range of the penalty parameter is , also determined by grid search; cross-validation is used to evaluate the performance of the model during training, with k set to 5; the accuracy of the trained classifier should be no less than 95%. The output fault type code of the support vector machine is converted into specific fault diagnosis results and corresponding fault control strategy parameters through a fault mapping table, including fault level, compensation coefficient and control mode selection, etc.; the fault mapping table is constructed by expert experience and historical fault case library, including the identification features of typical fault modes, severity rating (1-5 levels) and recommended compensation strategy.

[0153] Based on the steps of the above-mentioned multi-motor cooperative control method based on digital twinning, a multi-motor cooperative control system based on digital twinning is proposed, comprising:

[0154] a data acquisition module for acquiring real-time operating data of each motor in the multi-motor system, including speed, current, temperature, and vibration acceleration parameters, and performing standardization processing to generate a physical parameter dataset containing time sequence information;

[0155] a digital twin modeling module for fitting electromagnetic equation parameters using the least squares method based on the physical parameter dataset, extracting mechanical characteristic parameters through frequency domain analysis, establishing a thermodynamic model based on temperature-power data, and learning parameter time-varying characteristics using a recurrent neural network to generate a motor digital twin with adaptive updating capability;

[0156] a topology analysis module for analyzing the topological connection relationship between motors to generate collaborative control constraint conditions including speed synchronization constraints, torque balance constraints, and phase coordination constraints;

[0157] a collaborative optimization module for solving a multi-objective optimization problem containing tracking errors, power consumption, and collaborative deviations based on the digital twin state and constraint conditions, and generating control parameters that minimize the multi-objective function as multi-motor collaborative control instructions through gradient descent method;

[0158] a predictive control module for using the motor digital twin to perform forward-looking simulation to predict the system response at multiple future time steps and select the optimal control strategy based on a time-weighted multi-objective function;

[0159] a topology adaptation module for real-time monitoring of topology change events in the motor system, and quickly updating the topology relationship matrix using manifold interpolation algorithm when detecting motor failure or new device access;

[0160] a fault diagnosis module for calculating the state deviation between the physical motor and the motor digital twin, determining a fault when the deviation exceeds a threshold, locating the fault type and position through deviation pattern recognition, and automatically adjusting the collaborative control strategy to compensate for the fault impact.

[0161] The following is a real application scenario of the present application: multi-motor collaborative control in a multi-color printing mechanical system. In this embodiment, a large multi-color offset printing machine is used as the application scenario. The printing system is equipped with 6 servo motors, which drive different printing cylinders and paper conveying mechanisms to realize continuous multi-color register printing. During printing, each motor must work in strict coordination to ensure accurate color registration and smooth paper conveying.

[0162] The technical requirements of the printing system are as follows: the synchronization accuracy of each printing cylinder motor speed is not less than 0.01%; the phase synchronization error is not more than 0.05 degrees; the motor torque fluctuation under load disturbance is not more than 2% of the rated torque; when the system structure changes dynamically (such as adding or removing a printing unit), the control system needs to complete reconstruction and maintain stable operation within 100ms.

[0163] Step 100 instance: Obtain real-time running data of the motor of the printing system;

[0164] In the actual running printing machine, a high-precision sensor network is installed, including Hall sensors, current sensors, PT100 temperature sensors and three-axis acceleration sensors on each motor. The data acquisition system is set to a sampling frequency of 2 kHz, which is transmitted in real time to the edge computing unit through the EtherCAT field bus. As shown in Table 1:

[0165] Table 1: Data segment of original physical parameters of each motor of the printing machine (t=10.000s to t=10.002s)

[0166]

[0167] In the data preprocessing process, the system standardizes the collected raw data. The rotational speed data is divided by the rated rotational speed of 1200 rpm for normalization, the current data is normalized according to the rated current of each motor (M1-M4: 10A, M5-M6: 5A), and the temperature data is normalized by Standardization, and the vibration acceleration data is divided by 1g for normalization. At the same time, 3 Criterion is used to detect outliers, for example, the current 13.75A collected by M2 motor at t=12.157s is identified as an outlier and corrected by time series interpolation.

[0168] Step 200 instance: Generate a digital twin of the motor of the printing machine;

[0169] Model the digital twin for each motor. Take the main motor M1 of the printing cylinder as an example, the electromagnetic parameters are obtained by least squares fitting: , , .

[0170] Through the frequency domain analysis method, the mechanical parameters of M1 motor are calculated: rotational inertia , friction coefficient , nominal load torque .

[0171] During the 30-minute printing operation, the temperature rise data of M1 motor is recorded, and the thermodynamic parameters are obtained by fitting the thermodynamic model: heat capacity , thermal resistance . As shown in Table 2:

[0172] Table 2: Parameters of digital twin model of six motors of the printing machine

[0173]

[0174] To further improve the accuracy of the motor digital twin, a recurrent neural network is applied to learn the time-varying characteristics. In the actual operation of the printing system, as the printing material changes (such as the paper thickness changing from 80 g / m² to 200 g / m²), the recurrent neural network automatically adjusts the moment of inertia parameters of the motor M1, and outputs the correction value which is applied to the digital twin model.

[0175] Step 300 instance: analyze the topological connection relationship of the printing machine motor;

[0176] According to the physical structure of the printing machine, the topological connection relationship between the motors is established. M1-M4 are the main drive motors of the four-color printing unit, which need to be strictly synchronized; M5-M6 are the drive motors of the paper feeding and collecting part, which have matching requirements with the printing unit in terms of paper feeding speed. As shown in Table 3:

[0177] Table 3: Topological connection relationship matrix of printing machine motor system

[0178]

[0179] Based on the topological relationship, the cooperative control constraint conditions are generated: speed synchronization constraint between printing unit motors M1-M4 (i.e. 0.01% of the rated speed), phase synchronization constraint (i.e. 0.05 degrees); speed matching constraint between paper feeding motor M5 and printing unit where , is the transmission ratio coefficient; speed matching constraint between paper collecting motor M6 and printing unit where , .

[0180] Step 400 instance: implement cooperative optimization control based on digital twin;

[0181] When the printing machine system is running, cooperative optimization control is realized based on the constructed motor digital twin. Taking the paper thickness change scenario as an example, when the printing machine switches from 80 g / m² paper to 200 g / m² paper, the system automatically adjusts the control strategy.

[0182] First, the control algorithm solves the cooperative optimization objective function by gradient descent method to obtain the optimal control parameters that satisfy the constraint conditions. Then, in order to realize predictive control, the motor digital twin is used for forward-looking simulation to simulate the response of each motor in the future 10 time steps (total 10 ms) under the condition of paper load change. The system samples multiple candidate control strategies in the control space, including strategy A, strategy B and strategy C, and calculates the prediction performance indicators of each strategy. As shown in Table 4:

[0183] Table 4: Results of Collaborative Optimization Control under Thick Paper Switching Conditions in Printing Presses

[0184]

[0185] By evaluating predictive performance metrics The system selects to make The minimum control strategy is chosen as the final control decision at the current moment. In this example, the weighted objective function value of strategy C is 0.0425, which is significantly better than strategy A (0.0537) and strategy B (0.0683), and therefore it is selected as the optimal control strategy. Specific implementation parameters for strategy C: adjusting the input voltage increments of motors M1-M4. The M5 motor's speed-up time is delayed by 5ms, while the M6 ​​motor's speed-up time is advanced by 3ms, ensuring smooth and stable paper transfer.

[0186] Example of step 301: Addressing topology changes in a printing system;

[0187] In actual production, when switching from four-color printing to two-color printing, the system needs to remove motors M3 and M4 from the topology. Using a manifold interpolation algorithm, the system quickly reconstructs the topology matrix, as shown in Table 5.

[0188] Table 5: Changes in the motor topology matrix during manifold interpolation (intermediate state at t=0.5)

[0189]

[0190] The system completes topology reconstruction within 82ms and recalculates constraints: the speed synchronization constraints between M1 and M2 remain unchanged, but the speed matching relationships of M5 and M6 are only associated with M1 and M2. The updated speed matching constraints are as follows: and ,in .

[0191] Example of Step 500: Fault Diagnosis and Fault-Tolerant Control Implementation;

[0192] Taking the M2 motor bearing failure during printing as an example, the system detects the anomaly by comparing the state deviation between the physical motor and the digital twin. After 60 minutes of normal printing operation, the vibration acceleration of the M2 motor begins to increase, and the deviation vector... Vibration acceleration components If the value exceeds the preset threshold of 0.4g, the fault diagnosis process is triggered.

[0193] The system calculates the statistical characteristics of the standardized deviation vector to obtain the eigenvector. The feature vector is identified as a bearing inner ring fault (fault type) using a support vector machine classifier. ), with a failure severity rating of 2 (minor failure, can continue operation but requires close monitoring).

[0194] The system automatically applies a fault-tolerant control strategy: reduce M2 motor load by 10%, reduce the coupling strength between M1-M2-M3-M4, temporarily relax the speed synchronization constraint of M2 to Adjust the control parameters of other motors to compensate for the performance reduction of M2. Through these measures, the system can still maintain normal production under bearing failure conditions, and the printing register accuracy is only reduced by 12% until the scheduled maintenance period to replace the bearing.

[0195] It can be understood that the data preprocessing method known to those skilled in the art includes data cleaning, data conversion, data reduction, wherein the data conversion includes type conversion and normalization and standardization, although the dimension and type of the data are ignored in the description of the foregoing embodiments, the data preprocessing is the technical knowledge known to those skilled in the art and the preprocessing step of data processing, therefore, the known data preprocessing step is not described independently in the foregoing content.

[0196] The above describes the embodiments of the present application, but the embodiments are not limited to the specific embodiments described above, the specific embodiments described above are only illustrative, but not restrictive, and those skilled in the art can also make more forms of equivalent embodiments under the inspiration of the embodiments, which all belong to the protection of the embodiments.

Claims

1. A multi-motor cooperative control method based on digital twinning, characterized in that, The method comprises the following steps: obtaining real-time operation data of each motor in a multi-motor system, and generating a physical parameter data set containing time sequence information; based on the physical parameter data set, calculating the digital twin model parameters of each motor, and generating a motor digital twin; analyzing the topological connection relationship between the motors, and generating a cooperative control constraint condition; inputting the real-time state data of the motor digital twin and the cooperative control constraint condition into a cooperative optimization algorithm, and outputting a multi-motor cooperative control instruction; wherein the cooperative optimization algorithm solves a multi-objective optimization problem containing tracking error, power consumption and cooperative deviation, and generates control parameters that minimize the multi-objective function as the multi-motor cooperative control instruction.

2. The multi-motor cooperative control method according to claim 1, characterized in that, The physical parameter data set includes the speed, current, temperature and vibration acceleration parameters of each motor, and the parameters are standardized, normalized to a specific interval or standardized to a distribution with a mean of 0 and a variance of 1, eliminating the magnitude difference between different physical quantities.

3. The multi-motor cooperative control method according to claim 1, characterized by, The generation of the motor digital twin includes the following sub-steps: fitting the electromagnetic equation parameters of the motor using the least squares method, calculating the resistance, inductance and back electromotive force coefficient of the motor according to the current-speed data; extracting the mechanical characteristic parameters of the motor through a frequency domain analysis method, including the moment of inertia, friction coefficient and load characteristics; establishing a thermodynamic model based on temperature-power data to calculate parameters such as heat capacity and thermal resistance; integrating the parameters to form a complete motor digital twin model.

4. The multi-motor cooperative control method according to claim 3, characterized in that, It also includes learning the time-varying characteristics of motor parameters using a recurrent neural network, inputting historical parameter sequences into the recurrent neural network, and outputting dynamic correction values of the parameters to realize adaptive updating of the motor digital twin.

5. The method of claim 1, wherein, The cooperative control constraint condition includes: speed synchronization constraint for limiting the speed deviation between different motors; torque balance constraint for ensuring that the sum of the torques of the multi-motor system and the load torque are balanced; phase coordination constraint for specifying the phase difference range between different motors.

6. The multi-motor cooperative control method according to claim 1, wherein The objective function of the multi-objective optimization problem includes three normalized weighted indicators: the sum of the squares of the tracking errors of each motor, the power consumption of each motor, and the cooperative deviation between the motors, and the multi-objective optimization problem is solved by gradient descent method.

7. The multi-motor cooperative control method of claim 1, wherein It also includes using the motor digital twin to perform forward-looking simulation, inputting the current state and candidate control strategy into the motor digital twin model, predicting the system response at multiple future time steps, and selecting the control strategy that optimizes the predicted performance indicator.

8. The multi-motor cooperative control method according to claim 7, characterized by, The predicted performance indicator is defined as a time-weighted multi-objective function, including the weighted sum of tracking errors, power consumption and cooperative deviation at multiple future time steps, and each indicator is normalized to ensure comparability.

9. The method of claim 1, wherein, It also includes real-time monitoring of topological change events of the motor system, and when a motor failure or new device access is detected, a manifold interpolation algorithm is used to quickly update the topological relationship matrix and recalculate the cooperative control constraint condition.

10. A multi-motor cooperative control system based on digital twinning, configured to perform the multi-motor cooperative control method of any one of claims 1-9, wherein It comprises: a data acquisition module for obtaining real-time operation data of each motor in a multi-motor system; a digital twin modeling module for generating a motor digital twin based on the data; a topological analysis module for analyzing the topological connection relationship between the motors; a co-optimization module for computing control instructions based on the digital twin state and the constraints; a predictive control module for performing forward simulation; a topology adaptation module for handling system topology changes; a fault diagnosis module for identifying and handling system faults.

Images