Multi-motor cooperative control method

Abstract

The invention discloses a multi-motor cooperative control method, belongs to the technical field of electrical engineering and automation thereof, and aims at overcoming the defects of the traditional technology, and realizing the precise estimation of the rotating speed of a motor at the next moment and the assistance of subsequent regulation and control by means of a multi-layer feedforward neural network and the training of multi-working-condition operation data of the motor. And a sliding mode variable structure control system is constructed, related synchronization errors are defined, corresponding functions and reaching laws are designed, and it is ensured that the system can rapidly and stably respond and correct deviation. And an online parameter adaptive adjustment mechanism is set, and parameters are flexibly adjusted according to system fluctuation by using a genetic algorithm. In actual operation, the method is remarkable in effect. Multi-motor synchronization errors can be effectively reduced, and stable linkage is guaranteed; for complex working conditions and motor parameter fluctuation, error diffusion can be inhibited, and robustness can be enhanced; manual frequent debugging is not needed, the adaptability is good, the high requirement of industrial production for multi-motor cooperation is met, and the operation and maintenance cost can be reduced.

Description

Technical Field

[0001] The present invention relates to the field of electrical engineering and automation technology, and in particular to a multi-motor coordinated control method. Background Art

[0002] In many fields such as modern industrial production, automation equipment and intelligent robots, multi-motor collaborative control systems play a vital role and are widely used in assembly line operations, multi-axis linkage machinery, precision instrument drives and other scenarios, aiming to ensure that multiple motors can operate accurately and stably in synchronization and achieve efficient operation of equipment. However, traditional multi-motor collaborative control technology has many limitations and is difficult to meet the increasingly stringent industrial needs.

[0003] Traditional control methods mostly rely on simple coupling control architectures, such as the common cross-coupling and deviation coupling control methods. When faced with complex working conditions, synchronization accuracy is difficult to guarantee. When the motor encounters frequent load changes, such as a sudden increase or decrease in the weight of the workpiece to be processed on an automated production line, the traditional method lacks an accurate speed prediction mechanism and cannot adjust the motor output in advance, causing the speed synchronization error to increase sharply, seriously affecting the stability of the equipment linkage, and the product processing accuracy is greatly reduced.

[0004] Poor robustness is also a major shortcoming of traditional technology. After the motor runs for a long time, due to the wear and aging of the components, the motor's own parameters, such as inductance, resistance, and moment of inertia, will change; or the power grid fluctuates, causing the input voltage and current to be unstable. At this time, traditional control methods are difficult to quickly adapt to changes in motor parameters because they do not set dynamic adjustment strategies. The system is very likely to fall into a state of loss of control, and synchronization errors continue to accumulate, even causing equipment failures and shutdowns, causing huge losses to production.

[0005] Furthermore, traditional multi-motor collaborative control technology has poor adaptability. The industrial production process is complex and changeable, and different production stages have different requirements for motor speed and torque; even in the same production link, as equipment ages and environmental temperature and humidity change, working conditions are also changing dynamically. However, traditional methods often preset fixed control parameters, and manual debugging is time-consuming and labor-intensive, and it is difficult to adjust in real time according to working conditions, which not only increases equipment debugging and maintenance costs, but also frequently reduces production efficiency due to parameter mismatch.

[0006] In summary, the existing multi-motor collaborative control technology is plagued by problems such as low synchronization accuracy, weak robustness, and insufficient adaptability, which seriously restricts the development of industrial automation, high-end manufacturing and other fields. Summary of the invention

[0007] The main purpose of the present invention is to provide a multi-motor coordinated control method, which can effectively solve the problems mentioned in the background technology.

[0008] To achieve the above object, the technical solution adopted by the present invention is:

[0009] The multi-motor coordinated control method comprises the following steps:

[0010] S1. Each motor is equipped with a high-precision speed sensor and torque sensor to collect the motor speed n in real time. i (t) and torque T i (t) information, where i = 1, 2, ..., n, n is the total number of motors, and the sampling period is set to T s ; Build a motor parameter database and enter the rated speed of each motor Rated torque Moment of inertia J i ;

[0011] S2, select a multi-layer feedforward neural network to calculate the real-time speed, torque and speed change rate Δn of the motor at the previous k sampling moments i (tj), where j = 0, 1, ..., k-1, as the input layer node information, the output layer node is set to the motor speed prediction value at the next moment The ReLU activation function and back propagation algorithm are used to train the neural network using the motor's operating data under steady-state, acceleration, deceleration, and load mutation conditions. The training goal is to minimize the sum of squares of the error between the predicted speed and the actual speed. Where N is the number of sampling points corresponding to the total duration of the training data;

[0012] S3. For multi-motor systems, the speed synchronization error is defined as Where i≠j; torque synchronization error Integration to build a comprehensive synchronization error vector Design of synovial surface function Where L is the dimension of the comprehensive synchronization error vector, c l is a positive constant; using the exponential approach law Where η, ∈ are positive parameters, and sign(·) is the sign function.

[0013] Preferably, the method further comprises a real-time control step:

[0014] ① Each sampling period T s Perform data acquisition and preprocessing operations. Each motor sensor collects speed and torque data, removes noise interference through filtering, and calculates the speed change rate. The processed data is normalized to the [0,1] interval using the linear normalization method. The linear normalization formula is: Where x is the original data, x min 、x maxare the minimum and maximum values ​​of the sample data of the original data, respectively, to adapt to the neural network input requirements;

[0015] ② Input the normalized data into the trained neural network to obtain the predicted speed value of each motor at the next moment Then reverse normalization is performed to obtain the predicted value of the actual physical quantity; based on the real-time and predicted speed and torque data, the comprehensive synchronization error vector e(t) is calculated and substituted into the sliding film surface function to obtain s(e(t));

[0016] ③ Calculate the control input increment Δu according to the exponential reaching law i (t), combined with the motor's current control input u i (t), update the control quantity Make the control quantity act on the motor drive circuit, adjust the motor voltage and current, and change the motor speed and torque.

[0017] Preferably, an online parameter adaptive adjustment step is provided:

[0018] ①Introduction of synchronization error integral index Control energy consumption indicators As a performance evaluation indicator;

[0019] ② Every certain time mT s , where m is a positive integer. Based on the current performance indicators, the genetic algorithm is used to fine-tune the neural network weights, the synovial variable structure parameters η, ∈, c l ; The crossover and mutation probabilities of the genetic algorithm are adaptively adjusted according to the real-time fluctuation level of the system.

[0020] Preferably, the number of hidden layer nodes of the multi-layer feedforward neural network is determined based on empirical formulas and multiple experiments to adapt to the nonlinear mapping requirements of the motor operating state.

[0021] Preferably, the motor parameter database can be updated in real time. When the motor undergoes maintenance, replacement of key components or long-term operation causing parameter changes, new parameters are entered to ensure that the control algorithm operates according to accurate motor characteristics.

[0022] Preferably, the filtering process uses a combination of low-pass filtering and Kalman filtering to remove high-frequency noise and system random noise.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] With the help of a multi-layer feedforward neural network, the real-time speed, torque and speed change rate of the motor are used as input information, and the operating data of the motor under multiple working conditions such as steady state, acceleration, deceleration, and load mutation are collected for training, so as to achieve accurate prediction of the motor speed at the next moment, thereby providing effective support for subsequent regulation; a synovial variable structure control system is constructed, and the synovial surface function and exponential reaching law are designed by clearly defining the speed and torque synchronization errors to ensure that the system can respond quickly and stably to correct the operating deviation when the motor operation deviates; an online parameter adaptive adjustment mechanism is established, the synchronization error integral is introduced, the energy consumption index is controlled, and the genetic algorithm is regularly used to flexibly adjust the neural network weights and synovial key parameters in combination with the real-time fluctuations of the system. Thanks to these designs, this method relies on precise prediction and timely regulation in actual operation, which can effectively reduce the synchronization error between multiple motors and ensure smooth linkage of equipment; the robustness is enhanced. When encountering complex working conditions and motor parameter fluctuations, with the help of sliding film and adaptive mechanisms, error diffusion can be quickly suppressed to maintain stable synchronization of the motor; it has good adaptability and does not require frequent manual debugging throughout the process. It can automatically optimize the algorithm according to the working conditions, reduce operation and maintenance costs, and meet the needs of industrial production for high-precision and high-reliability collaborative operation of multiple motors. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a flow chart of the multi-motor coordinated control method of the present invention. DETAILED DESCRIPTION

[0026] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0027] like Figure 1 As shown, the multi-motor coordinated control method specifically includes the following steps:

[0028] 1. Data collection and preparation stage

[0029] Equip each motor with a speed and torque sensor, press T s Collect data and build a database to enter rated parameters.

[0030] 2. Neural network construction and training stage

[0031] Build a multi-layer feedforward neural network with specific data as input and the predicted speed value at the next moment as output, and use multi-operating condition data for training to reduce the speed error.

[0032] 3. Error definition and control framework construction stage

[0033] Define the multi-motor speed and torque synchronization errors, construct the comprehensive error vector, and design the sliding film surface function and exponential reaching law.

[0034] 4. Real-time control execution phase

[0035] per ton s Data is collected periodically, filtered, calculated, and normalized before being input into the neural network. The error vector is calculated and substituted into the sliding film surface function to update the control quantity and drive the motor.

[0036] 5. Online parameter adaptive adjustment stage

[0037] Set ISE, E consumption Indicator, every mT s Genetic algorithm is used to adaptively fine-tune the neural network weights and synovial membrane structure parameters according to the indicators.

[0038] The specific embodiments in practical applications are divided into:

[0039] 1. Preliminary preparation stage

[0040] Hardware installation and parameter entry: A high-precision Hall effect speed sensor is precisely installed at the end of each motor shaft. Its resolution can reach 0.1 rpm, which can accurately capture the real-time speed of the motor; a current transformer torque sensor is placed at the motor winding to monitor the torque in real time. The sampling period T s Set it to 10 milliseconds. At the same time, build a motor parameter database and enter the rated parameters of each motor: such as the rated speed of motor No. 1 Moment of inertia J1 = 0.2 kg·m2; These parameters will serve as the basis for subsequent operations. The principle is that the initial characteristics of the motor determine its ideal operating state. Entering parameters facilitates the control system to compare the actual operating data and detect abnormalities in a timely manner.

[0041] Data acquisition system test: Start the equipment to run without load, and check the stability and accuracy of the sensor data collection. For example, observe the fluctuation of the speed data of motor No. 1, which should be within ±0.5 rpm, and the torque fluctuation should be less than 0.1 Nm; if it exceeds the range, fine-tune the sensor installation position or calibrate the sensitivity to ensure data reliability and provide a solid foundation for subsequent control.

[0042] 2. Neural Network Training Phase

[0043] Network construction and working condition simulation: A four-layer feedforward neural network is selected. The input layer contains 6 nodes, which correspond to the real-time speed, torque and speed change rate of the motor at the first three sampling moments; the output layer has 1 node, which outputs the predicted value of the motor speed at the next moment. After multiple tests, the number of nodes in the two middle hidden layers was set to 10 and 8 respectively, and the activation function used ReLU to speed up the network convergence. A large amount of working condition data was collected to simulate load mutations, resistance fluctuations (similar to deceleration conditions), frequent task switching (different speed requirements), and other scenarios, and the operating data of each motor under complex working conditions was obtained as a training set.

[0044] Training execution and optimization: The neural network is trained using the back propagation algorithm. The training goal is to minimize the sum of squares of the error between the predicted speed and the actual speed. Where N is the number of sampling points corresponding to the total duration of the training data. During training, the initial value of the learning rate is set to 0.01. If the error decreases slowly after every 100 rounds of training, the learning rate is reduced by a decay coefficient of 0.9. After 500 rounds of training, the average error of the predicted speed of each motor is controlled within 1 rpm, achieving applicable accuracy.

[0045] 3. Error Definition and Control Architecture Construction Phase

[0046] Definition of synchronization error: For a four-motor system, define the speed synchronization error, such as the speed synchronization error between motor 1 and motor 2. Between motor 3 and motor 4 The torque synchronization error is similar, like Integration to build a comprehensive synchronization error vector Comprehensively covers the coordinated relationship between each motor, which is conducive to the precise control of overall synchronization.

[0047] Synovial Surface Function and Reaching Law Design: Design Synovial Surface Function where c l The value is determined based on the importance of the motor and the sensitivity of the working condition. For example, a larger value is set for the coefficient c2 corresponding to the speed synchronization error of the second motor because it directly affects the accuracy. The exponential approach law is used. After multiple debugging, when η=5 and ∈=3, the system approaches the sliding surface quickly with small jitter, ensuring that the system converges stably to the synchronization state.

[0048] 4. Real-time control execution stage

[0049] Data collection and preprocessing (performed every 10 milliseconds): Each motor sensor collects speed and torque data, first passes through a Butterworth low-pass filter to remove high-frequency noise, and then calculates the speed change rate; the processed data is normalized to the [0,1] interval using a linear normalization method, and the formula is: For example, the speed of motor No. 1 is 1490 rpm, and the minimum and maximum speeds of the batch data are 1480 and 1510 rpm, respectively. The normalized value is Adapt to neural network input requirements.

[0050] Prediction and error calculation: Input the normalized data into the trained neural network to obtain the predicted speed value of each motor at the next moment Denormalize to restore the actual value; calculate the comprehensive synchronization error vector e(t) based on the real-time and predicted speed and torque data, and substitute it into the sliding surface function to get s(e(t)). Assuming that the speed synchronization error between motor 1 and motor 2 is 5 rpm at this moment, the system deviation degree can be accurately located through the sliding surface function calculation.

[0051] Control quantity generation and motor drive: Calculate the control input quantity increment Δu according to the exponential reaching law i (t), combined with the motor's current control input u i (t), update the control quantity The control quantity is converted into a suitable voltage and frequency signal by the inverter to act on the motor drive circuit to adjust the motor voltage and current. If the No. 1 motor needs to increase speed to match the demand, the voltage is increased in time and the current is increased to increase the motor speed and correct the synchronization deviation.

[0052] 5. Online parameter adaptive adjustment stage

[0053] Performance indicator monitoring: Introducing synchronization error integral indicator The energy consumption index is used as a performance evaluation index. Real-time monitoring shows that during continuous operation and when the equipment load increases, the ISE value increases, indicating that the synchronization error accumulates; consumption A sharp increase indicates that the control energy consumption is too high.

[0054] Genetic algorithm parameter adjustment (executed every 5 seconds): Based on the current performance indicators, the genetic algorithm is used to fine-tune the neural network weights, synovial variable structure parameters η, ∈, c l The population size is set to 30, the initial crossover probability is 0.8, and the mutation probability is 0.1. Depending on the degree of system fluctuation, the mutation probability is increased to 0.2 when the fluctuation is large to help the optimization jump out of the local optimum. After adjusting the parameters, the system quickly adapts to the working conditions, the synchronization error returns to a low level, and the energy consumption is reasonably reduced.

[0055] During the entire operation cycle of the automation equipment, through this multi-motor collaborative control method, the speed synchronization error of each motor is stably controlled within ±2 rpm, and the torque fluctuation is less than 0.5 Nm, which significantly improves the work quality and reduces the scrap rate by about 30%; in the face of ±15% load mutation and ±5% grid voltage fluctuation, it can restore synchronization within 20 milliseconds, and the number of equipment shutdowns is reduced by 8 times per month on average, efficiently meeting production needs, demonstrating the excellent practicality and stability of this method in complex industrial scenarios.

[0056] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A multi-motor coordinated control method, characterized in that: The following steps are involved: S1. Each motor is equipped with a high-precision speed sensor and torque sensor to collect the motor speed n in real time. i (t) and torque T i (t) information, where i = 1, 2, ..., n, n is the total number of motors, and the sampling period is set to T s ; Build a motor parameter database and enter the rated speed of each motor Rated torque Moment of inertia J i ; S2, select a multi-layer feedforward neural network to calculate the real-time speed, torque and speed change rate Δn of the motor at the previous k sampling moments i (tj), where j = 0, 1, ..., k-1, as the input layer node information, the output layer node is set to the motor speed prediction value at the next moment The ReLU activation function and back propagation algorithm are used to train the neural network using the motor's operating data under steady-state, acceleration, deceleration, and load mutation conditions. The training goal is to minimize the sum of squares of the error between the predicted speed and the actual speed. Where N is the number of sampling points corresponding to the total duration of the training data; S3. For multi-motor systems, the speed synchronization error is defined as Where i≠j; torque synchronization error Integration to build a comprehensive synchronization error vector Design of synovial surface function Where L is the dimension of the comprehensive synchronization error vector, c l is a positive constant; using the exponential approach law Where η, ∈ are positive parameters, and sign(·) is the sign function.

2. The multi-motor coordinated control method according to claim 1, characterized in that: Also includes real-time control steps: ① Each sampling period T s Perform data acquisition and preprocessing operations. Each motor sensor collects speed and torque data, removes noise interference through filtering, and calculates the speed change rate; The processed data is normalized to the interval [0,1] using the linear normalization method. The linear normalization formula is: Where x is the original data, x min 、x max They are the minimum and maximum values ​​of the sample data where the original data is located, respectively, to adapt to the neural network input requirements; ② Input the normalized data into the trained neural network to obtain the predicted speed value of each motor at the next moment Then reverse normalization is performed to obtain the predicted value of the actual physical quantity; based on the real-time and predicted speed and torque data, the comprehensive synchronization error vector e(t) is calculated and substituted into the sliding film surface function to obtain s(e(t)); ③ Calculate the control input increment Δu according to the exponential reaching law i (t), combined with the motor's current control input u i (t), update the control quantity u i (t+T s )=u i (t)+Δu i (t), so that the control quantity acts on the motor drive circuit, adjusts the motor voltage and current, and changes the motor speed and torque.

3. The multi-motor coordinated control method according to claim 2, characterized in that: There are online parameter adaptive adjustment steps: ①Introduction of synchronization error integral index Control energy consumption indicators As a performance evaluation indicator; ② Every certain time mT s , where m is a positive integer. Based on the current performance indicators, the genetic algorithm is used to fine-tune the neural network weights, the synovial variable structure parameters η, ∈, c l ; The crossover and mutation probabilities of the genetic algorithm are adaptively adjusted according to the real-time fluctuation level of the system.

4. The multi-motor coordinated control method according to claim 1, characterized in that: The number of hidden layer nodes of the multi-layer feedforward neural network is determined based on empirical formulas and multiple experiments to adapt to the nonlinear mapping requirements of the motor operating state.

5. The multi-motor coordinated control method according to claim 1, characterized in that: The motor parameter database can be updated in real time. When the motor undergoes maintenance, replacement of key components or long-term operation causing parameters to change, new parameters are entered to ensure that the control algorithm operates according to accurate motor characteristics.

6. The multi-motor coordinated control method according to claim 2, characterized in that: The filtering process uses a combination of low-pass filtering and Kalman filtering to remove high-frequency noise and system random noise.

Images