KR20210108144A - Control method for lms system based on learning and simulation - Google Patents

Abstract

The present invention to provide an LMS simulation system that provides an LMS equipment operation state actually and visually in conjunction with an actual industrial controller. The present invention relates to an LMS system based on learning and simulation, comprising: a first step of collecting coil topology information that a plurality of coils constitute, from a servo drive; a second step of transmitting the coil topology information to a physical simulator; a third step of constructing, by the physical simulator, a virtual coil topology corresponding to the real environment in a simulation environment by using the coil topology information; a fourth step of performing a test drive by traversing a coil topology with a first mover; a fifth step of measuring a torque distribution value for each coil when the first mover passes each coil in the fourth step, and transmitting the measured torque distribution value together with a position value of the current mover to the physical simulator; a sixth step of generating, by the physical simulator, a torque profile for each coil by using the position value of the mover and the torque distribution value for each coil; and a seventh step of generating a convoluted model by using the generated torque profile for each coil as input data of a torque model generator having a built-in neural network machine learning algorithm.

Description

CONTROL METHOD FOR LMS SYSTEM BASED ON LEARNING AND SIMULATION

The present invention relates to a learning-based LMS (Linear Motion System, hereinafter referred to as LMS) system or a control method thereof, and specifically, a simulation capable of verifying the operation of the LMS controller software before hardware is manufactured and designing a simulation model It relates to a system for improving the control efficiency of an actual LMS system by applying the system and a generated torque model to an actual controller.

In addition, the LMS simulator of the present invention communicates with the actual LMS controller to describe the LMS operation by the controller command, and through a simulation based on the physics engine, it is possible to actually see the operation of the LMS in which the physical force acts. it is one system That is, the present invention relates to a method and system for controlling an LMS 　 system, and as a means for efficiently controlling the LMS, a synthetic torque model is generated through a learning process using a simulation model. It is intended to be applied to the control process of the LMS system.

In general, a linear motion system (LMS) is composed of a plurality of movers, a plurality of coils forming a topology, and a controller controlling the topology, and each coil has a structure in which electromagnetic force is generated by a servo drive device.

A permanent magnet is placed at the bottom of the mover, and the servo drive generates electromagnetic force by applying a current to the coil to move the mover located above the coil. It is a structure in which the mover moves by the repulsive force generated by the interaction of the generated magnetic force.

And, as a prior art for a process simulation apparatus or method, there is Korean Patent Publication No. 10-2017-0077888.

This prior art implements a process virtually by arranging virtual equipment and materials that perform the same functions as real equipment through 3D virtual space, and a plurality of equipment/material sets having preset shape information are stored. Part, parametric information storage in which parametric information about the shape of the material included in the specific equipment/material set changed in response to the change in the shape of the equipment included in the specific equipment/material set among the plurality of equipment/material sets is stored A layout order definition unit that selects a facility/material set included in the process to be simulated, and sets the arrangement order of each selected facility/material set, and a selected selected having shape information determined based on shape information and parametric information and a layout assembling defining unit for arranging each of the equipment/material sets according to the arrangement order, and assembling the arranged equipment/material sets in a preset assembling method, wherein the preset shape information includes shape information of each of a plurality of equipment and a plurality of equipment The shape information of the material corresponding to each consists of information that is matched with each other, and the shape information of each of a plurality of facilities and the shape information of the material corresponding to each of the plurality of facilities are matched to each other by constructing the shape information with a set of facilities/materials, A technology that can reduce the time required for separately simulating equipment and materials is disclosed.

However, the prior art has a limitation in that the process is simply converted into a simulation state and cannot be controlled through an actual industrial controller.

In order to drive the LMS in the industry, controller software that meets the needs of the industry along with the linear motor and driver technology must be developed, but there is no solution that can verify the control operation during the development process.

In addition, since the existing development process has a structure in which the LMS controller software development can be started after the LMS hardware configuration is completed whenever the process of the industry changes, there was a problem that it takes too much time to develop, and the conventional system uses the LMS There was a problem that the operation operation of the LMS hardware could not be shown realistically and visually by interacting with the controller software.

It is an object of the present invention to provide an LMS simulation system that provides an LMS equipment operation state in real and visually in conjunction with an actual industrial controller.

In addition, an object of the present invention is to develop a technology capable of developing motion applications in parallel with LMS hardware manufacturing and equipment construction, and verifying motion without actual hardware.

In addition, the present invention can verify motion without actual hardware, so it can shorten the delivery time for building LMS facilities, have flexibility in hardware installation and modification, and implement a control algorithm verified through simulation to increase stability. do.

Another object of the present invention is to provide an LMS system that enables distributed control and increases control efficiency by generating a torque distribution model.

The present invention, a plurality of coils forming a topology; a servo driver for generating electromagnetic force by applying a current to each of the coils; a plurality of movers positioned above the coil and provided with a permanent magnet to move by a repulsive force due to the mutual magnetic force of the electromagnetic force of the coil and the magnetic force of the permanent magnet; a controller controlling the topology; and a physics simulator having a torque model generator for generating a torque model by a machine learning algorithm; a learning and simulation-based LMS system comprising:

a first step of collecting coil topology information composed of the plurality of coils from the servo drive; a second step of transmitting the coil topology information to a physical simulator; a third step of configuring, in the physical simulator, a virtual coil topology corresponding to the real environment in a simulation environment using the coil topology information; a fourth step of performing a test drive by traversing the coil topology with one mover; a fifth step of measuring a torque distribution value for each coil and transmitting the measured torque distribution value together with the current mover position value to a physical simulator when the first mover passes each coil in the fourth step; A sixth step of generating a torque profile for each coil by using the physical simulator as a position value of a mover and a torque distribution value for each coil; and a seventh step of generating a convoluted model by using the generated torque profile for each coil as input data of a torque model generator with a built-in neural network machine learning algorithm; do.

In the fourth step, the test drive is repeated a plurality of times until the generated synthetic torque model converges, and the physical simulator transmits the generated synthetic torque model to the controller and the test drive is completed. further includes ;

The present invention may be a computer-readable recording medium in which a program for performing the above control method is recorded.

The learning and simulation-based LMS system according to the present invention enables the actual LMS controller to precisely control distributed control between the coils constituting the topology through a torque distribution learning process using a physics simulator, and through this, the stability of the system is improved and expected The effect of increasing the lifespan occurs.

1 is a graph showing the magnetic flux density distribution by position of a motor stator, which is a configuration of a learning and simulation-based LMS system of the present invention;
2 is a graph showing a torque profile when two consecutive coils control a mover in the learning and simulation-based LMS system of the present invention;
3 is a physics-based simulation software screen for learning the LMS torque model in the learning and simulation-based LMS system of the present invention;
4 is a block diagram of a learning and simulation-based LMS system of the present invention,
5 shows the operation algorithm of the learning and simulation-based LMS system according to the present invention.

The objects, specific advantages and novel features of the present invention will become more apparent from the following detailed description and preferred embodiments taken in conjunction with the accompanying drawings. In addition, the terms used are terms defined in consideration of functions in the present invention, which may vary according to the intention or custom of the user operator. Therefore, definitions of these terms should be made based on the content throughout this specification.

The present invention is a learning-based LMS simulation and control system for controlling an LMS (Linear Motion System), and the key is to learn a torque distribution generated by a coil topology in the LMS to generate a torque model.

The LMS consists of multiple movers, multiple coils forming a topology, and a controller that controls the topology. Each coil is a structure that generates electromagnetic force by a servo drive device, and a permanent magnet is placed under the mover. In order to move the mover located above the coil, the servo drive generates electromagnetic force by applying a current to the coil. The structure in which the mover moves by the repulsive force is as described above.

The controller identifies the mover position in the current coil topology based on the position of the mover detected by the coil, and the user can give a move command to each mover, and the controller interprets the mover movement command written by the user as a coil unit command. Thus, a torque can be applied to the mover by sending a command to the coil required for the mover to move.

When a mover moves by the electromagnetic force of the coil and goes out of the controllable range of the coil, the controller sends a command to the next coil positioned in the direction in which the mover is moving so that the mover can continuously perform the move command currently being performed.

1 is a graph showing the magnetic flux density distribution for each position of a motor stator, which is a component of the learning-based LMS control system of the present invention, and the electromagnetic field of the coil is generated in a complicated manner depending on the shape of the coil and the design of the mechanical part constituting the coil. In addition, the magnetic field formed by the permanent magnet of the mover is also affected by various factors such as size, shape, and arrangement. When a constant current flows due to the irregular magnetic field formation of the coil and the mover, the torque applied to the mover shows a non-linear distribution according to the position of the mover. In order to generate a model from such a torque distribution, a model function should be selected based on values measured through nonlinear regression analysis, but it was difficult to determine the suitability of the model function to be used.

Recently, general nonlinear model generation has been facilitated by neural network theories such as multi-layer perceptron structures, and implementation has become possible as neural networks are used in various fields such as deep learning.

2 is a graph showing a torque profile when two consecutive coils control a mover in the learning-based LMS control system of the present invention. It can be seen from the graph of FIG. 2 that, when the mover is located between the two coils in the controllable range, the peak current of the coil can be lowered by appropriately distributing and applying the torque to the two coils in order to apply a specific torque to the mover.

The technology of generating a torque model by learning the torque distribution generated by the coil topology in the LMS is the key. In particular, this patent provides a physics-based simulation environment for learning the torque model. The effect of being able to test the movement characteristics of the mover in a simulation environment using the model occurs. 3 is a physical-based simulation software screen for learning the LMS torque model in the learning-based LMS simulation and control system of the present invention.

4 is a block diagram of a learning-based LMS simulation and control system of the present invention, and the learning-based LMS control system includes a coil topology composed of a plurality of contiguous coils, a servo drive controlling each coil, and torque to the drive It can be seen that it is composed of a motion controller that transmits commands, a physics simulator including a torque model generator that generates a torque model by a machine learning algorithm, and a simulation interface for data sharing between the physics simulator and the motion controller.

5 shows an operation algorithm of a learning-based LMS control system according to the present invention. The operation control sequence of the learning-based LMS control system is as follows.

1) First, when the booting of the controller in the system according to the present invention starts, the current real-time network-connected servo drive is scanned and data is collected from the corresponding servo drive to configure the topology information currently configured by the coils. Coil topology information can be manually input by a user or can be automatically searched for by using one mover.

2) By the above operation, when the topology configuration is completed, the coil topology configuration of the current network is transferred to the physical simulator. The physics simulator constructs a virtual coil topology corresponding to the real environment in the simulation environment by using the configured topology information.

3) The controller places one mover at the starting point of the coil topology to perform a test drive for model generation, and the controller performs a test drive. When the test drive starts, the controller moves the mover and sends a control command to traverse the coil topology configured in the system. through the physics simulator.

4) The physics simulator creates a torque profile for each coil using the position of the mover and the measured torque distribution for each coil. A convoluted model is generated by using the generated torque profile for each coil as input data of a torque model generator having a built-in machine learning algorithm such as a neural network. The generated torque model represents the distribution of the combined torque between adjacent coils according to a specific amount of current for each coil topology.

The above 'torque model generator' is a newly developed software module for LMS control, and it is based on neural network learning by taking the torque profile collected through trial run as input. Neural network base, machine, learning, and algorithm adjust the weights of the nodes that make up the neural network by using multiple profiles and inputs to generate predictions and models. And, as a result of the algorithm execution, each coil is a topology, the position of the mover, and the output according to the current and the output torque according to the current are predicted by the current-torque and the output distribution factor.

In addition, the 'convoluted model' may be a kind of 'current-torque 　 conversion 　 coefficient 　 distribution 　 model'. The 'current-torque 　 conversion 　 coefficient 　 distribution 　 model' shows the correlation between the input current amount and the output torque according to the coil topology and the position of the mover in the form of a distribution function. This current-torque conversion coefficient distribution function enables more precise LMS dispersion control. When controlling the motor, the controller generates an amount of current with a control command to apply the desired torque. In the case of a general circular motor, the amount of current for applying the torque is distributed almost uniformly depending on the position of the rotating shaft, so the torque and the amount of current can be expressed in a constant constant multiple relationship. can However, in the case of LMS, since the torque applied to the mover by the amount of current varies depending on the relative position of the coil and the mover (ie, the coil topology and the position of the mover), it is difficult to express the torque and the amount of current as a constant multiple. If the 'current-torque conversion coefficient distribution model' is reflected in the controller, the correlation between the input current amount and the output torque can be expressed according to the position of the mover, so more precise control is possible.

5) Repeat each procedure of the test drive until the generated synthetic torque model for each topology converges sufficiently. The physics simulator sends the synthesized torque model generated by the learning module to the controller through the simulation interface and completes the test drive. When this step is completed, the controller will have a synthetic torque model for each topology.

In the present invention, the condition for the end of the algorithm is that the convergence of the torque is not, but the current-torque, the conversion, the coefficient, the distribution, and the model converge. The torque model generator of the algorithm is a module that repeatedly receives the torque profile as input and generates the current-torque conversion coefficient and distribution model for each position. At the beginning of the algorithm iteration, the input torque and the number of samples of the profile are small, so the distribution generated at every iteration increases as the number of samples increases by one day or more. Here, in the algorithm, 'torque = convergence of the model' means that the torque, the model, the output from the generator, every iteration, the distribution, and the model are distributed within a constant deviation.

6) During this operation, the controller distributes the torque command to each coil using the synthesized torque model for each topology. By sending commands over a real-time network, the actual control system can be driven.

On the other hand, in the present invention, it is possible to implement a program in which the simulation system is recorded as computer-readable codes on a computer-readable recording medium. The computer-readable recording medium includes all types of recording devices in which data readable by a computer system is stored. Examples of the computer-readable recording medium include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage device.

In addition, the computer-readable recording medium may be distributed in a network-connected computer system, and the computer-readable code may be stored and executed in a distributed manner. And functional programs, codes, and code segments for implementing the present invention can be easily inferred by programmers in the technical field to which the present invention pertains.

Although the preferred embodiment of the present invention has been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements by those skilled in the art using the basic concept of the present invention as defined in the following claims are also provided. is within the scope of the

Claims (5)

a plurality of coils forming a topology; a servo driver for generating electromagnetic force by applying a current to each of the coils; a plurality of movers positioned above the coil and provided with a permanent magnet to move by a repulsive force due to the mutual magnetic force of the electromagnetic force of the coil and the magnetic force of the permanent magnet; a controller controlling the topology; and a physics simulator having a torque model generator for generating a torque model by a machine learning algorithm, comprising: an LMS system comprising:
a first step of collecting coil topology information composed of the plurality of coils from the servo drive;
a second step of transmitting the coil topology information to a physical simulator;
a third step of configuring, in the physical simulator, a virtual coil topology corresponding to the real environment in a simulation environment using the coil topology information;
a fourth step of performing a test drive by traversing the coil topology with one mover;
a fifth step of measuring a torque distribution value for each coil and transmitting the measured torque distribution value together with the current mover position value to a physical simulator when the first mover passes each coil in the fourth step;
A sixth step of generating a torque profile for each coil by using the physical simulator as a position value of a mover and a torque distribution value for each coil; and
A method of controlling a learning and simulation-based LMS system comprising: a seventh step of generating a convoluted model by using the generated torque profile for each coil as input data of a torque model generator with a built-in neural network machine learning algorithm. According to claim 1,
The test driving of the fourth step is a control method of a learning and simulation-based LMS system, characterized in that repeating the test driving a plurality of times until the generated synthetic torque model converges. According to claim 1,
The physical simulator transmits the synthesized torque model generated to the controller and completes the test drive; an eighth step of controlling the learning and simulation-based LMS system, characterized in that it further comprises. 4. The method of claim 3,
The control method of a learning and simulation-based LMS system, characterized in that the controller holds the synthesized torque model for each coil topology by the above process. A computer-readable recording medium in which a program for performing the control method according to any one of claims 1 to 4 is recorded.

Images