CN106444371A - AC servo control system based on migration neural network - Google Patents
AC servo control system based on migration neural network Download PDFInfo
- Publication number
- CN106444371A CN106444371A CN201610685405.2A CN201610685405A CN106444371A CN 106444371 A CN106444371 A CN 106444371A CN 201610685405 A CN201610685405 A CN 201610685405A CN 106444371 A CN106444371 A CN 106444371A
- Authority
- CN
- China
- Prior art keywords
- servo
- motor
- dsp controller
- servomotor
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses an AC servo control system based on a migration neural network. The system includes an upper computer, a motor control card, a DSP controller, a servo driver, a servo power supply, a Hall element, a servo motor and a photoelectric encoder. The upper computer is connected to the motor control card through a PCI bus. The motor control card is connected to the DSP controller through a serial port. The servo driver, the Hall element, and the servo motor are successively connected to the DSP controller. The photoelectric encoder is connected to the servo motor and the DSP controller. According to the invention, the system uses a DSP microprocessor to subtract a position feedback value from a position command value of the control card, and the difference value serves as control errors. The system also uses the deep neural network algorithm which is based on migration learning to generate a motor speed control signal, and effectively addresses the shortage of training samples required by the neural network control algorithm.
Description
Technical field
The present invention relates to AC servo control system, and in particular to a kind of AC Servo Control based on migration neutral net
System.
Background technology
Servomotor is also known as operating motor, and in automatic control system, its torque and rotating speed are controlled by signal voltage.
When the size of signal voltage and phase place change, the rotating speed of motor and rotation direction sensitively and accurately will be followed very much
Change.When blackout, rotor can be stalled in time.21 century is entered, AC servo is more and more ripe, and market is presented
Quick development in pluralism, lot of domestic and foreign brand comes into the market to compete.AC Servo Technology has become industrial automation at present
One of supportive technology.
Modern AC servosystem is applied to aerospace and military field, such as cannon, radar control earliest.Progress into
To industrial circle and civil area.Commercial Application mainly includes the numerical control machine of high-precision numerical control machine, robot and other broad sense
Tool, such as textile machine, printing machinery, package packing machine, armarium, semiconductor equipment, mailing machine, metallurgical machinery, automatically
Change streamline, various special equipments etc..Wherein servo consumption maximum industry be successively:Lathe, packaging for foodstuff, weaving, electronics
Quasiconductor, plastics, printing and rubber manufacturing machinery, add up to more than 75%.
The correlation technique of AC servo, constantly develops with the demand of user always.Motor, driving, sensing
With being continually changing, create various configurations of the corresponding technologies such as control technology.For motor, disc type can be adopted
Motor, iron-core less motor, linear electric motors, external rotor electric machine etc., driver can adopt various power electronic elements, sense and anti-
Feedback device can be different accuracy, the encoder of performance, rotation become and Hall element, control technology from the beginning of using single-chip microcomputer, one
Until using High Performance DSP and various programmable modules.
At present, the existing common AC servo control system of China mainly carries out servo using single-chip microcomputer as controller
Drive, small volume, good economy performance, but it is not good to calculate performance, it is difficult to be applied to computationally intensive modern control algorithms such as mould
The paste control strategy such as neutral net and neuron control.But Neural Network Control Algorithm needs to use markd training in a large number
Sample, and the neural network model of a servo-driver, it is impossible to be applied to another servo-driver.But in actual production
During, it is difficult to markd training sample in a large number is obtained, typically unmarked training sample in a large number, therefore neutral net mould
Type is difficult to popularization and application.
Inquired about by Patents, find there is following open source literature:
Patent " servo-control system based on neutral net and method " [application number CN200910236904.3] is disclosed
A kind of positional servosystem, using nerve network controller, for receiving model error, error differential output nerve network control
Device output processed carries out servo operation.The nerve network controller used by the patent need to rely on markd operation number in a large number
According to for training neutral net.
Patent " servo-control system based on RBF neural and method " [application number CN200910093591.0] is proposed
A kind of Neural Network Adaptive Control method for being applied to servosystem, including feedforward controller, PID controller, neutral net
Controller, robust item, adder, servo performs device, the program achieves the nonlinear compensation to servosystem and interference suppression
System, improves tracking accuracy and the robustness of servosystem.The patent needs also exist for the labelling that has of a large amount of servosystem and runs number
According to for training neutral net, to improve the control performance of neutral net.
Patent " a kind of embedded intelligent controller based on DSP " [application number CN200610008276.X] is main on hardware
Including:DSP processing unit, CPLD (CPLD), FLASH program storage, converter, DA is changed
Device, FIFO memory, CAN communication module.By multiple optimization designs, put forth effort to improve System Operation efficiency and process energy
Power, while with good autgmentability.In terms of control software, this controller has embedded multiple advanced controls such as complex neural network
Algorithm processed, can on-line tuning, optimal control parameter, improve real-time control performance.The composite nerve net used by the patent
Network, compared with traditional neural network, with convergence fast, calculate simple effect, but there is still a need for substantial amounts of have reference numerals
According to being trained network.
Analyzed by above-mentioned patent, find existing scheme using the Based Intelligent Control for needing to have flag data to be trained in a large number
Algorithm, but in practice, after a new servomotor needs operation for a long time, acquisition markd operation in a large number could be given
Data.
Content of the invention
Present invention aim to overcome that the deficiencies in the prior art, especially solving existing technical scheme shortage has labelling in a large number
The problems such as training data, governing speed are slow, dynamic response capability is poor.A kind of AC servo control system based on DSP is provided, should
Device uses dsp controller, in conjunction with migration Neural Network Control Algorithm, without the need for markd servosystem training data in a large number,
Only need to the control effect for having labelling training data, just can realizing high precision, good stability of a small amount of servosystem.
For solving above-mentioned technical problem, the present invention is adopted the following technical scheme that:
A kind of AC servo control system based on migration neutral net, including host computer, motor control card, DSP control
Device, servo-driver, servo power supply, Hall element, servomotor and photoelectric encoder, it is characterised in that wherein:
The host computer is connected to motor control card by pci bus, and the motor control card is by serial ports and DSP control
Device is connected, and wherein, the motor control card is instructed according to host computer and generates pulse train, pulse number, position, frequency and frequency
Rate of change, acceleration are by PC control;
The servo-driver, Hall element, servomotor are sequentially connected to dsp controller, the photoelectric encoder with
Servomotor is connected with dsp controller, wherein, controlling value and institute that the dsp controller is exported according to the servo-driver
State the angle feed-back value of photoelectric encoder, the speed feedback value of the Hall element and error signal is produced, using regulation algorithm pair
Error signal is carried out calculating and produces motor control signal;The Hall element is used for detecting the phase current of servomotor as speed
Value of feedback;The photoelectric encoder anglec of rotation of servomotor is converted to orthogonal electric impulse signal as angle feed-back
Value;
The deep neural network control method realized by a kind of AC servo, it is characterised in that including network structure
The steps such as design, data set pretreatment, unsupervised training, Training, specifically include:
S1, using the existing a large amount of service datas on certain servo devices, builds the depth of an AC servo motor
Degree neural network model.
S2, with the existing a small amount of service data of target servo motor, allows deep neural network carry out transfer learning, to adapt to
Operation variation tendency on target servo motor;
S3, the error amount amount of the being controlled output to target servo motor is calculated.
Further, the deep neural network control method, it is characterised in that the transfer learning that step S2 is adopted
Method be by being trained to existing deep neural network model again with the low volume data for obtaining on target servo motor, micro-
Adjust weights to adapt to target device.
Description of the drawings
Fig. 1 is the structural representation of a specific embodiment of the present invention.
Fig. 2 is the method flow diagram of constructing neural network in a specific embodiment of the present invention.
Specific embodiment
Below in conjunction with the accompanying drawings and specific embodiment the present invention is carried out in further detail with complete explanation.May be appreciated
It is that specific embodiment described herein is only used for explaining the present invention, rather than limitation of the invention.
Referring to Fig. 1, a kind of AC servo control system based on migration neutral net, including host computer 1, motor control card
2nd, dsp controller 3, servo-driver 4, servo power supply 5, Hall element 6, servomotor 7 and photoelectric encoder 8, wherein:
The host computer 1 is connected to motor control card 2 by pci bus, and the motor control card 2 is by serial ports and DSP
Controller 3 is connected, and wherein, the motor control card 2 is instructed according to host computer 1 and generates pulse train, pulse number, position, frequently
Rate and frequency change rate, acceleration are controlled by host computer 1;
The servo-driver 4, Hall element 6, servomotor 7 are sequentially connected to dsp controller 3, the photoelectric coding
Device 8 is connected with servomotor 7 and dsp controller 3, wherein, the control that the dsp controller 3 is exported according to the servo-driver 4
The angle feed-back value of value processed and the photoelectric encoder 8, the speed feedback value of the Hall element 6 produce error signal, use
Adjusting algorithm carries out calculating generation motor control signal to error signal;The Hall element 6 is used for detecting the phase of servomotor 7
Electric current is used as speed feedback value;The anglec of rotation of servomotor 7 is converted to orthogonal electric impulse signal by the photoelectric encoder 8
As angle feed-back value;
Preferably, the Hall element 6 detects the mutually electricity of servomotor 7 using electromagnetic isolation Hall element circuit
Stream, after carrying out A/D conversion, is transferred to dsp controller 3;On and off switch turns 3.3V power voltage supply using 5V voltage;PWM output is logical
When crossing optic coupling element and causing to transmit pwm control signal, control circuit and power circuit are separated;The dsp controller 3 can be external
Memory expansion, increases computational efficiency.
Referring to Fig. 2, the deep neural network control method realized by a kind of AC servo, it is characterised in that include
The steps such as network structure design, data set pretreatment, unsupervised training, Training, specifically include:
One of beneficial effect of the present invention program be using existing unmarked in a large number service data, to realize neutral net
Training, solving existing nerve network controller needs markd this defect of service data of a large amount of target servo motors.Its
Principle is expressed as follows:The existing unmarked in a large number service data on certain servomotor is obtained, the present embodiment uses existing
Another servomotor service data, for build a deep neural network, specifically include procedure below:
(1) in the present embodiment, deep neural network uses denoising autocoder per the connected mode of two interlayers
(DAE), its working method is as follows:
1), the vector of prefilter layer neuron composition is set as X=(x1, x2..., xn), x ∈ [0,1]d;Subsequent layer neuron
The vector of composition is Y=(y1, y2..., yn), y ∈ [0,1]d′.Prefilter layer node is connected entirely with subsequent layer node.Each is automatic
Encoder will be input into x by parameter θ={ W, the b } for determining and be mapped to a hidden layer and represent y:
Y=fθ(x)=s (Wx+b) (1)
Wherein b is amount of bias, and W is weight matrix of the size for d × d ', and s represents a nonlinear activation function, this
Sigmoid function used in example.
2), after, acquisition hidden layer represents y, the decoding of autocoder is carried out, i.e., by another group of parameter θ '={ W ', b ' }
Y is mapped, generates the reconstruct z of x:
Z=gθ(y)=s (W ' y+b ') (2)
When the error minimum of reconstruct, parameter (W, b, W ', b ') reaches optimum:
L is an error function, adopts mean square difference function in this example.
3), in actual applications, due to workshop environment and sensor accuracy problem, the servomotor operation of acquisition
Substantial amounts of noise is entrained with data.In order to suppress impact of the noise to model accuracy, need input vector x to be carried out partly break
Bad, generate vectorFormula (1) is substituted into afterwards:
Neutral net can be so made to have higher robustness.
(3) before network is trained, need first to carry out pretreatment to data set.Pretreatment is comprised the following steps:
1), x-th sample (0 <=x in sequence is represented to training sample s, a s (x) with m consecutive sample values
< m, is arranged from front to back by sample time order).Set step-length n (n < m), then first for generating list entries s1
For { s (0), s (1) ..., s (m-1) }, second list entries s of generation2So that 0,0 ..., 0, s (n), s (n+1) ... s
(m-1) }, the 3rd list entries s of generation3For 0,0 ... 0, s (2n), s (2n+1) ... s (m-1) }, by that analogy.?
In controlled quentity controlled variable prediction, from predicted point more close to numerical value, the weights for accounting in prediction are heavier, on the other hand, from predicted point
Numerical value farther out can also reflect the potential trend of controlled quentity controlled variable change.
2), sample data is normalized, interval is [0,1]:
(4) the unsupervised training stage, by carrying out weights beginningization with a large amount of unlabeled exemplars to each hidden layer of network, net is allowed
Network has the ability of the feature for extracting training sample.
(5) the Training stage, by markd sample set, the error in network is modified, finely tunes each nerve
Connection weight between unit, the ability for making network really change with prediction of energy consumption and export.
S2:With target servo motor some service datas existing, deep neural network is allowed to carry out transfer learning, to adapt to
Operation variation tendency on target servo motor, specifically includes following steps:
(1) the same service data for obtaining target servo motor, in the present embodiment unlike, to target servo electricity
Machine, only obtains the service data of a day.
(2) by same preprocessing means, the markd sample set towards the servomotor is obtained.
(3) original train neural network model on the basis of, with the markd sample set of target servo motor again
Secondary carry out Training, network is finely adjusted, with allow forecast model adapt to new equipment operation conditions.
S3:The error amount amount of being controlled output to target servo motor is calculated, and its method is as described below:
When being controlled to servomotor using migration neutral net, according to controlling value and the photoelectricity of servo-driver output
The angle feed-back value of encoder, the speed feedback value of Hall element produce error signal, error signal are input into neutral net, i.e.,
Control signal is obtained.
It is above-mentioned that but embodiments of the present invention are not limited by the above for the present invention preferably embodiment, its
His any spirit without departing from the present invention and the change that is made under principle, modification, replacement, combine, simplify, all should be
The substitute mode of effect, is included within protection scope of the present invention.
Claims (2)
1. a kind of based on migration neutral net AC servo control system, including host computer, motor control card, dsp controller,
Servo-driver, servo power supply, Hall element, servomotor and photoelectric encoder, it is characterised in that wherein:
The host computer is connected to motor control card by pci bus, and the motor control card is by serial ports and dsp controller phase
Even, wherein, the motor control card is instructed according to host computer and generates pulse train, pulse number, position, frequency and frequency change
Rate, acceleration are by PC control;
The servo-driver, Hall element, servomotor are sequentially connected to dsp controller, the photoelectric encoder and servo
Motor is connected with dsp controller, wherein, controlling value and the light that the dsp controller is exported according to the servo-driver
The angle feed-back value of photoelectric coder, the speed feedback value of the Hall element produce error signal, using regulation algorithm to error
Signal is carried out calculating and produces motor control signal;The Hall element is used for detecting the phase current of servomotor as velocity feedback
Value;The photoelectric encoder anglec of rotation of servomotor is converted to orthogonal electric impulse signal as angle feed-back value.
2. a kind of migration nerve net realized by AC servo based on migration neutral net according to claim 1
Network control method, it is characterised in that walk including network structure design, data set pretreatment, unsupervised training, Training etc.
Suddenly, specifically include:
S1, using the existing a large amount of service datas on certain servo devices, builds the depth god of an AC servo motor
Through network model.
S2, with the existing a small amount of service data of target servo motor, allows deep neural network carry out transfer learning, to adapt to target
Operation variation tendency on servomotor;
S3, the error amount amount of the being controlled output to target servo motor is calculated.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610685405.2A CN106444371A (en) | 2016-08-12 | 2016-08-12 | AC servo control system based on migration neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610685405.2A CN106444371A (en) | 2016-08-12 | 2016-08-12 | AC servo control system based on migration neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106444371A true CN106444371A (en) | 2017-02-22 |
Family
ID=58181166
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610685405.2A Pending CN106444371A (en) | 2016-08-12 | 2016-08-12 | AC servo control system based on migration neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106444371A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109245665A (en) * | 2018-11-22 | 2019-01-18 | 哈尔滨工业大学 | A kind of motor servo control method based on data study |
-
2016
- 2016-08-12 CN CN201610685405.2A patent/CN106444371A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109245665A (en) * | 2018-11-22 | 2019-01-18 | 哈尔滨工业大学 | A kind of motor servo control method based on data study |
CN109245665B (en) * | 2018-11-22 | 2020-08-07 | 哈尔滨工业大学 | Motor servo control method based on data learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106406162A (en) | Alternating current servo control system based on transfer neural network | |
CN104876145B (en) | Electronic active compensation hoister system | |
CN103034127B (en) | A kind of axial magnetic bearing control system | |
CN102497156B (en) | Neural-network self-correcting control method of permanent magnet synchronous motor speed loop | |
CN103231798A (en) | Control device of digital electric steering engine and control method | |
CN101917150A (en) | Robust controller of permanent magnet synchronous motor based on fuzzy-neural network generalized inverse and construction method thereof | |
CN105892412B (en) | Multi-shaft motion control system hardware structure based on self-defined bus | |
CN104533716B (en) | A kind of independent pitch load controlled method based on Kalman filtering | |
CN105242587A (en) | Double-drive hybrid direct-current and low-voltage intelligent servo driver circuit | |
CN103208958A (en) | DC (direct control) servo drive control system | |
CN105888970B (en) | The adaptive inner mould vibration control method that intelligent blower blade is optimized based on grey information | |
CN104834219A (en) | PMLSM driven XY platform control method based on empirical mode decomposition, and system thereof | |
CN108923711A (en) | A kind of absolute value encoder adjusting zero method of servo-system zero point check and correction | |
CN106444371A (en) | AC servo control system based on migration neural network | |
CN104476744B (en) | A kind of injection machine energy-saving servo-control system | |
CN109639200A (en) | A kind of rotary inertia on-line identification method based on electric motor load torque detection | |
CN105099263B (en) | Supersonic motor Position And Velocity control system based on generalized regression nerve networks identifier | |
CN106338911B (en) | A kind of expert PID control method applied to rotary electromechanical actuator servo-system | |
CN117047774A (en) | Multi-joint hydraulic mechanical arm servo control method and system based on sliding film self-adaption | |
CN116614030A (en) | Permanent magnet servo motor composite model predictive control method based on GPC and FCS-MPC | |
CN104238624B (en) | A kind of tracking of maximum power point and device | |
CN104155909B (en) | Stability in finite time control system based on the tape speed observer of PC+FPGA | |
CN107786130B (en) | Linear switched reluctance motor control system and its control method | |
CN108155836A (en) | Based on global online heuristic dynamic programming permanent magnet synchronous motor vector control method | |
CN109510551A (en) | The implementation method and device of essential safe type servo-system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170222 |
|
WD01 | Invention patent application deemed withdrawn after publication |