CN112698568A - Filament constant tension control method based on optimized BP neural network - Google Patents
Filament constant tension control method based on optimized BP neural network Download PDFInfo
- Publication number
- CN112698568A CN112698568A CN202011147183.1A CN202011147183A CN112698568A CN 112698568 A CN112698568 A CN 112698568A CN 202011147183 A CN202011147183 A CN 202011147183A CN 112698568 A CN112698568 A CN 112698568A
- Authority
- CN
- China
- Prior art keywords
- neural network
- tension
- filament
- optimized
- value
- 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.)
- Granted
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 100
- 238000000034 method Methods 0.000 title claims abstract description 25
- 230000002068 genetic effect Effects 0.000 claims abstract description 35
- 230000035772 mutation Effects 0.000 claims abstract description 11
- 210000000349 chromosome Anatomy 0.000 claims abstract description 10
- 239000011159 matrix material Substances 0.000 claims abstract description 8
- 230000008030 elimination Effects 0.000 claims abstract description 4
- 238000003379 elimination reaction Methods 0.000 claims abstract description 4
- 238000004804 winding Methods 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 10
- 230000003044 adaptive effect Effects 0.000 claims description 8
- 230000004044 response Effects 0.000 claims description 8
- 210000002569 neuron Anatomy 0.000 claims description 6
- 108090000623 proteins and genes Proteins 0.000 claims description 5
- 206010064571 Gene mutation Diseases 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 4
- 238000012937 correction Methods 0.000 claims description 3
- 238000001514 detection method Methods 0.000 claims description 3
- 230000003993 interaction Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000001105 regulatory effect Effects 0.000 claims description 3
- 239000008266 hair spray Substances 0.000 abstract description 7
- 238000005457 optimization Methods 0.000 description 11
- 239000004753 textile Substances 0.000 description 11
- 238000004519 manufacturing process Methods 0.000 description 9
- 210000002268 wool Anatomy 0.000 description 9
- 230000006870 function Effects 0.000 description 7
- 238000004088 simulation Methods 0.000 description 7
- 238000005507 spraying Methods 0.000 description 7
- 230000008859 change Effects 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 4
- 230000001276 controlling effect Effects 0.000 description 4
- 238000013461 design Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 238000007664 blowing Methods 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 238000011217 control strategy Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 239000000835 fiber Substances 0.000 description 2
- 241001391944 Commicarpus scandens Species 0.000 description 1
- 230000005483 Hooke's law Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000003340 mental effect Effects 0.000 description 1
- 238000003825 pressing Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000036186 satiety Effects 0.000 description 1
- 235000019627 satiety Nutrition 0.000 description 1
- 238000009987 spinning Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
Images
Landscapes
- Tension Adjustment In Filamentary Materials (AREA)
- Spinning Or Twisting Of Yarns (AREA)
Abstract
The invention discloses a filament constant tension control method based on an optimized BP neural network, which comprises the steps of firstly, building a filament constant tension control hardware system based on the optimized BP neural network, determining a BP neural network structure, inputting sample data and an expected output value of the neural network, then initializing population of a weight and a threshold, calculating a population fitness value, carrying out selection, crossing and mutation operations on an initial value and the threshold of the neural network, carrying out win-win elimination according to the fitness value after genetic operation, finally describing a weight matrix and a threshold matrix in the optimized BP neural network structure into chromosome strings, judging whether the system reaches the required precision as a finishing condition, if so, finishing the algorithm, and if not, returning to continue the operation. The invention solves the problem of unstable tension in the hairspray tape yarn in the prior art.
Description
Technical Field
The invention belongs to the technical field of filament constant tension control, and particularly relates to a filament constant tension control method based on an optimized BP neural network.
Background
The textile industry is a representative of the development of the national light industry, is the basic industry of the country and is also the industry indispensable to the society. As a big country with historical culture, China is a symbol of China culture communication in the Tang dynasty, and silk represents advanced culture and textile technology of China, is famous in the world and is difficult to be obtained. With the continuous development of science and technology, people do not meet the requirements of satiety and pay more attention to the enjoyment of mental level. Therefore, the demands on the quality, softness and warmth of the fabric are increasing. With the continuous innovation and innovation of the technology, the equipment of textile enterprises in China is gradually lagged behind, the produced products are uneven, the production efficiency is low, and the like. The existing problem is that textile enterprises in China are not negligible. Due to the ever-increasing living requirements of people, the produced low-quality belt yarns cannot meet the pursuit of people for delicate life. In order to meet the requirement of people on high quality of the hairspray band yarn, researches show that the fluctuation of the filament tension of the hairspray band yarn is an important factor influencing the quality of the hairspray band yarn, the fluctuation exists in the whole textile process, and the filament tension needs to be kept constant for producing the high-quality hairspray band yarn. The filament tension is too large, and when the traction force applied to the filament is greater than the peak value of the elasticity of the filament, the filament is easy to break, so that the filament needs to be stopped and threaded again, and the production efficiency is seriously influenced; when the tension of the filament is too small, the produced tape yarn is lack of elasticity, meshes are different in size, and the forming is poor, so that the phenomenon of fiber leakage is easily caused.
In the prior art, "hangshuai, dirichun, popjunjie" embedded yarn tension control system design [ J ] mechanical design and manufacture, 2015(08): 140-. Fuzzy control is adopted as an algorithm of the controller, and PID parameters are subjected to online self-tuning through fuzzy reasoning, so that the change of the parameters in the control process can be automatically identified. Simulation results show that the overshoot of the fuzzy PID control is reduced by 21% compared with the overshoot of the conventional PID control, the adjustment time is shortened by 1s, and the fuzzy PID control effect is more stable and the adaptability is better when the input signal changes. On one hand: the prior art has the defect that the cost of hardware equipment of a tension control system is too high, and the tension control system is not suitable for small textile enterprises in China, so that the cost of the enterprises is increased. On the other hand: the algorithm generally adopts a classical PID control algorithm or a fuzzy control algorithm and other single algorithms, and the produced yarn can not meet the requirements of people in the face of systems with nonlinearity, multiple time-varying or complex characteristics and the like and can not achieve a good effect on the adjustment of the filament tension.
Disclosure of Invention
The invention aims to provide a filament constant tension control method based on an optimized BP neural network, which solves the problem of unstable tension in a hairspray belt yarn in the prior art.
The technical scheme adopted by the invention is that the filament constant tension control method based on the optimized BP neural network is implemented according to the following steps:
The present invention is also characterized in that,
the filament constant tension control hardware system based on the optimized BP neural network in the step 1 has the specific structure as follows: the tension detection module is a tension sensor, adopts a JZHL-T1 three-roller type, is arranged between a yarn drum filament and a tension adjusting device and is used for detecting an initial tension value of the filament, the tension adjusting module adopts a lever type, the detected tension value is compared with a set tension threshold value through a main control chip to act, and the motor driving module adopts a stepping motor to drive the adjusting device to move up and down; the man-machine interaction module is used for LCD display and LED buzzer alarm; the main control module adopts a single chip microcomputer of STC12C5A08S2 type and is used for comparing the signal detected by the tension sensor with the threshold value and processing the signal of the LCD display screen.
The optimization of the BP neural network structure in the step 2 specifically comprises the following steps:
comprises an input layer, a hidden layer and an output layer;
wherein, the input layer nodes are selected as follows: linear velocity v of winding mechanism2The friction force f between the filament and the pressure rod, the speed v of the pressure rod moving up and down0And the feed speed v of the thread1These four quantities serve as input layers for the neural network;
the hidden layer nodes are selected as follows: implicit toThe layer carries out threshold value filtering on the input signal and transmits the signal to the output layer, the output layer feeds the error signal back to the input layer through the hidden layer, and the error signal is obtained according to an empirical formula
n is the number of input layers, m is the number of output layers, a is a constant of 1 … 10, and the number of hidden layers is 5;
the output layer nodes are selected as follows: the neurons of the output layer correspond to three adjustable parameters k of the controllerp,ki,kd，kpThe proportion adjustment coefficient is used for accelerating the response speed of the system and improving the adjustment precision of the system; k is a radical ofiEliminating residual error for integral regulating coefficient; k is a radical ofdThe dynamic performance of the system is improved for differentiating the tuning coefficient.
The step 4 is as follows:
selecting a 4 x 5 x 3 neural network structure to adjust the linear velocity v of the winding mechanism2The friction force f between the filament and the pressure rod, the speed v of the pressure rod moving up and down0And the feed speed v of the thread1These four quantities serve as input samples for the neural network; motor control parameter k to be output desirablyp,ki,kdAs output, because the initial weight and threshold of the neural network are randomly generated, the randomly generated weight and threshold are sent to the genetic algorithm as the initial population of the genetic algorithm, and the genetic algorithm is based on the objective function of the neural network
Calculating fitness value of network genetic algorithm, Ep(t) the p-th group is output after t times of weight adjustment, a selection operator is acted on the population, the selection aims to directly inherit the optimized individual to the next generation or generate a new individual through pairing and crossing and then inherit the new individual to the next generation, the crossing operation is to mutually exchange partial genes of two paired chromosomes according to a certain mode so as to form two new individuals, the mutation operation is to simulate the gene mutation in the biological evolution process, and the gene mutation occurs at a certain part on the chromosome string to become the unique gene in the populationAnd (3) a gene type without two, wherein the termination condition is set as that the algorithm is ended when the evolution algebra is 100, if the termination condition is not met, the genetic operation is continued, and the parameters of the genetic algorithm are set as follows: determining the evolution algebra of the population as 100, the size scale of the population as 50, and setting the initial value of the adaptive parameter as the cross probability Pc0.6, probability of mutation PmDecoding the optimized initial value and the threshold value to be 0.05 and sending to a neural network; the neural network continuously compares the expected output correction weight value with the threshold value until k is metp,ki,kdThe requirements of (3).
The filament constant tension control method based on the optimized BP neural network has the advantages that the cost of hardware design can be reduced, the precision is relatively high, the structure is simple, the operation is convenient, the operation is reliable, the method is very suitable for small enterprises in the textile field of China, and the corresponding cost can be saved. The tension system is a nonlinear time-varying control system, and for the interference problem of tension fluctuation, the system is required to be free of overshoot and high in response speed. In order to meet the requirement of the yarn on tension control, a BP neural network controller is designed, and the controller can adjust the output parameters of the controller according to the running state of the system. The experimental simulation shows that the neural network algorithm meets the requirement of the rapidity of the system, but the random initial value has a certain influence on the early-stage stability of the tension system, the genetic algorithm is further adopted to optimize the BP neural network, the performance of the BP neural network optimized by the adaptive genetic algorithm is superior to that of the traditional BP neural network algorithm through comparison, the optimized algorithm eliminates the early-stage fluctuation of the filament tension system, the system is more stable, the time for the system to reach stability is shortened, and the performance of the system for quick response is better. The experimental and simulation results prove that the optimization algorithm has good control effect, quick response time and no overshoot. The requirements of the hairspray belt yarn on the rapid and stable control of the filament tension are met.
Drawings
FIG. 1 is an overall view of a tension control system;
FIG. 2 is a diagram of a filament tension control architecture;
FIG. 3(a) is a view of the passive unwinding of the filament;
FIG. 3(b) is a view of the active unwinding of the filament;
FIG. 4 is a PID control system based on a BP neural network;
FIG. 5 is a block diagram of a BP neural network according to the present invention;
FIG. 6 is a flow chart of adaptive genetic algorithm optimization of a BP neural network;
FIG. 7 is an input-output contrast curve before and after optimization;
FIG. 8 is an input-output contrast curve (10 times amplified) before and after optimization;
FIG. 9 is a JZHL-T1 three-roller tension sensor;
fig. 10 is a hardware block diagram of a filament constant tension control system.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The filament tension control in most enterprise wool blowing machines is of a passive unwinding type, and has many non-negligible factors, such as the friction force between a yarn drum and a yarn storage rack, the pulling force of a small needle cylinder on the yarn drum, and the like, which are important factors which are not negligible. Fancy yarns lacking tension control have certain disadvantages in competition. The method is applied to all textile enterprises in China, the quality of the fancy yarn is improved, and the international influence of the fancy yarn in China is further improved. Aiming at the uncontrollable and unstable filament tension in the wool blowing machine, a PID filament tension controller based on an optimized BP neural network is designed. Firstly, a winding mathematical model of a tension controller is established according to the system structure of the wool spraying machine. The control strategy of the system is that the BP neural network controls the rotation of the motor, but the BP neural network has larger fluctuation at the early stage through simulation discovery. In order to reduce the existing fluctuation of the system in the early stage, a Genetic Algorithm (GA) is introduced on the original basis to optimize the initial weight of the BP neural network. Simulation shows that the neural network can be optimized through a genetic algorithm, system fluctuation caused by random selection of initial values is eliminated, the stability of the system is improved, and the time for the system to reach stability is shortened. The research aims to solve the problem of controlling the filament tension of the wool spraying machine, fills the blank of small and medium-sized enterprises in China, and has great research significance and research prospect.
And (3) analyzing the structure of the control system:
in fancy spinning, factors affecting yarn tension are many, such as spindle speed, yarn number, and the like. The system architecture is shown in fig. 1. When the fluctuation moment of the filament tension exists in the process of producing the air-jet textured tape yarn, a tension sensor is needed to detect the tension value of the filament at any moment, the filament is wound on a measuring roller of a three-roller type tension sensor, the tension sensor amplifies and converts the detected tension value and sends the amplified and converted tension value to a controller, and the controller receives data and then displays the detected tension value on an LCD (liquid crystal display), and mainly displays the detected tension value, a set tension threshold value and the rotating speed of a motor. The general threshold value of the tension which can be set by the user is set to be 3.0, and if the user wants to adjust the threshold value of the tension, the adjustment can be carried out through an add-subtract key of the hardware platform; when the real-time detected tension value is smaller than the set tension threshold value, the motor of the pressure lever can drive the filaments to move downwards, and the tension of the filaments can be increased; when the tension value reaches the range of plus or minus 0.2 of the set threshold value, the motor stops rotating, and the tension value reaches a better state in the dynamic change process; when the tension value exceeds the positive 1 range of the set threshold value, the motor drives the pressure lever to move upwards, and the tension of the filament can be reduced. If the wire is broken, the stress of the tension sensor is 0, and the motor stops rotating the LED and the buzzer to give an alarm. The filament tension system belongs to closed-loop control, and the tension of the filaments can be continuously adjusted according to the production process in the production process. The control system has certain value on the production efficiency and the yarn quality of the tape yarns.
The invention discloses a filament constant tension control method based on an optimized BP neural network, which is implemented according to the following steps:
the filament constant tension control hardware system based on the optimized BP neural network in the step 1 has the specific structure as follows: the tension detection module is a tension sensor, adopts a JZHL-T1 three-roller type, is structurally shown in figure 9, is arranged between a yarn drum filament and a tension adjusting device and is used for detecting an initial tension value of the filament, the tension adjusting module adopts a lever type, the detected tension value is compared with a set tension threshold value through a main control chip to act, and the motor driving module adopts a stepping motor to drive the adjusting device to move up and down; the man-machine interaction module is used for LCD display and LED buzzer alarm; the main control module adopts a single chip microcomputer of STC12C5A08S2 type and is used for comparing the signal detected by the tension sensor with the threshold value and processing the signal of the LCD display screen.
the optimization of the BP neural network structure in the step 2 specifically comprises the following steps:
comprises an input layer, a hidden layer and an output layer;
wherein, the input layer nodes are selected as follows: linear velocity v of winding mechanism2The friction force f between the filament and the pressure rod, the speed v of the pressure rod moving up and down0And the feed speed v of the thread1These four quantities serve as input layers for the neural network;
the hidden layer nodes are selected as follows: the hidden layer carries out threshold value filtering on the input signal and transmits the signal to the output layer, the output layer feeds the error signal back to the input layer through the hidden layer, and the error signal is obtained according to an empirical formula
n is the number of input layers, m is the number of output layers, a is a constant of 1 … 10, and the number of hidden layers is 5;
the output layer nodes are selected as follows: the neurons of the output layer correspond to three adjustable parameters k of the controllerp,ki,kd，kpThe proportion adjustment coefficient is used for accelerating the response speed of the system and improving the adjustment precision of the system; k is a radical ofiEliminating residual error for integral regulating coefficient; k is a radical ofdThe dynamic performance of the system is improved for differentiating the tuning coefficient.
as shown in fig. 6, step 4, initializing a population of the weight and the threshold and calculating a population fitness value, performing selection, crossing and mutation operations on an initial value and the threshold of the neural network, performing win-win elimination according to the fitness value after genetic operations, determining whether the above steps are needed again according to a set genetic algorithm ending condition, returning if needed, and returning the optimized initial value and threshold to the neural network if not needed; through the algorithm principle of the neural network, information forward propagation and error backward propagation are carried out, weight and threshold values are continuously modified, a weight matrix and a threshold matrix in the optimized BP neural network structure are described into chromosome strings, whether the system reaches the required precision is judged as a finishing condition, if yes, the algorithm is finished, and if not, the algorithm is returned to continue.
The step 4 is as follows:
selecting a 4 x 5 x 3 neural network structure to adjust the linear velocity v of the winding mechanism2The friction force f between the filament and the pressure rod, the speed v of the pressure rod moving up and down0And the feed speed v of the thread1These four quantities serve as input samples for the neural network; motor control parameter k to be output desirablyp,ki,kdAs output, because the initial weight and threshold of the neural network are randomly generated, the randomly generated weight and threshold are sent to the genetic algorithm as the initial population of the genetic algorithm, and the genetic algorithm is based on the objective function of the neural network
Calculating fitness value of network genetic algorithm, Ep(t) the output of the network after t times of weight adjustment in the p-th group, acting a selection operator on the population, the purpose of selection is to directly inherit the optimized individuals to the next generation or generate new individuals by pairing and crossing and then inherit to the next generation, the crossing operation is to exchange partial genes of two paired chromosomes according to a certain mode, thereby forming two new individuals, the mutation operation is modularGene mutation occurs in a certain part of a chromosome string in the process of simulating biological evolution to become a unique genotype in a population, the termination condition is set as the algorithm is ended when the evolution algebra is 100, if the termination condition is not met, genetic operation is continued, and the parameters of the genetic algorithm are set as follows: determining the evolution algebra of the population as 100, the size scale of the population as 50, and setting the initial value of the adaptive parameter as the cross probability Pc0.6, probability of mutation PmDecoding the optimized initial value and the threshold value to be 0.05 and sending to a neural network; the neural network continuously compares the expected output correction weight value with the threshold value until k is metp,ki,kdThe requirements of (3).
And (3) tension system modeling analysis:
in the textile process, in order to ensure stable tension when producing the fuzzing tape yarn, the analysis and modeling are simplified into the figure 2, and the movement of the filament between the bobbin and the hollow spindle when producing the tape yarn can be analyzed as a winding structure. With reference to FIGS. 3(a) to 3(b), let T be the tension of the filament and V be the linear velocity of the filament traveling in the hollow ingot2The running speed of the tension adjusting unit is V0Herein, the winding speed will be defined as V2And is kept constant, V1Defined as the filament feed speed, V1To be connected with V0The influence of (c). If the tension is too small, the motor drives the pressure lever to move downwards, namely the tension is increased; if the tension is too large at the moment, the motor drives the pressing rod to move upwards, and the tension value is reduced at the moment. When the tension value is proper, the motor of the pressure lever stops running, and the tension at the moment is kept unchanged. The filament is then subjected to forces acting up and down the compression bar to change the tension.
Let the filament elastic coefficient be k, and the tension be F (t), according to Hooke's law:
F(t)＝kx(t) (1-1)
in the formula: f (t) -real time filament tension, E-filament elastic modulus, t-time yarn travel in L range, cross-sectional area of S-filament
The deduction of the above formula shows that the filament tension as the adjusting object belongs to the integral link. Due to V2Is constant, thus changing V1The size is mainly determined by changing the operation state of the motor. When the tension is too large or too small, the motor should be adjusted to change V in time1The value of (c). When the tension value required by the process is reached, the filament is kept to stably run at the proper tension value.
The unwinding mode is improved:
the traditional wire feeding structure of the wool spraying machine adopts a passive unwinding mode, a filament passes through a crochet hook of a small needle cylinder, the up-and-down motion of the crochet hook passively pulls the filament to drive a winding drum to rotate, and the filament is passively fed. Thus, the frictional force has a great influence on the tension of the filament, resulting in unstable tension. A motor and a pressure lever are added on the passive unwinding structure, so that the tension of the filament can be indirectly controlled by controlling the pressure lever to move up and down.
In the invention, the controller adopts a single chip microcomputer with STC12C5A08S2 model, and the tension sensor is JZHL-T1 model and is the most widely applied tension sensor in the market. The selected range is 20N, the built-in transducer can output 0-5v voltage signals, and the LCD screen can display tension data, the rotating speed of the motor and a set tension value at the moment. The default tension threshold value of the invention is 3.0N, which represents the most appropriate tension in the production process, and the tension threshold value can be modified by a key according to the requirements of different types of yarns on tension. The control object of the invention is a motor of a filament compression bar; when the real-time detected tension value is smaller than the set tension threshold value, the motor of the pressure lever can drive the filaments to move downwards, and the tension of the filaments can be increased; when the tension value reaches the range of plus or minus 0.2 of the set threshold value, the motor stops rotating, and the tension value reaches a better state in the dynamic change process; when the tension value exceeds the positive 0.2 range of the set threshold value, the motor drives the pressure lever to move upwards, and the tension of the filament can be reduced. If the wire is broken, the motor can stop rotating the LED and the buzzer immediately to give an alarm.
The BP neural network controller structure is shown in FIG. 4, the system is composed of two parts, (1) the BP neural network adjusts the parameters of the controller according to the operation state of the system to achieve the optimization of a certain performance index, and the output of the neuron corresponds to the adjustable parameter k of the controllerp,ki,kd. (2) The classical PID control strategy performs closed-loop control on a control object.
The classical numerical PID algorithm is:
u(k)＝u(k-1)+Δu(k) (3-1)
Δu(k)＝kp(e(k)-e(k-1))+kie(k)+kd(e(k)-2e(k-1)+e(k-2)) (3-2)
in the formula kp,ki,kdThe control quantity at the sampling moment u (k) represents the proportional coefficient, the integral coefficient and the differential coefficient, and u (k-1) is the output at the k-1 moment; e (k) is the error input.
The BP neural network structure of the present invention is shown in fig. 5:
(1) selection of input layer nodes
The production process of the jet-textured tape yarn and the modeling analysis of the filament tension system for the reason of the fluctuation of the filament are clarified, and factors influencing the filament tension stability are used as input quantities in order to keep the filament tension stable. The quantities affecting the filament tension factor are used as training samples of the neural network, in which the number of nodes of the input layer corresponds. For a filament tension control system of a wool spraying machine, the tension of a filament is adjusted by mainly controlling a stepping motor to change the position of a pressure lever. The main factors influencing the filament tension by analysis are: linear velocity v of winding mechanism2The friction force f between the filament and the pressure rod, the speed v of the pressure rod moving up and down0And the feed speed v of the thread1. Therefore, these four quantities are used as the input layer of the present neural network.
(2) Selection of hidden layer nodes
The hidden layer plays a very important role in the neural network, and performs threshold filtering on an input signal and transmits the filtered signal to the output layer. The output layer feeds back the error signal to the input layer through the hidden layer. The three-layer neural network structure can approximate any function. The number of hidden layers is selected according to the complexity of the system, a BP neural network generally adopts a single hidden layer, the system structure is complex, multiple hidden layers can be selected, and although the multiple hidden layers can improve the precision of the system and reduce the error of the system, the training time can be influenced by more neural network layers. Therefore, the comprehensive consideration of the test selection precision, the error and the training time is required.
For a filament tension control system, MATLAB multiple simulation tests show that when the number of hidden layers is close to that of neurons of an input layer, the convergence rate is high, and the fluctuation peak value of the system is small. Thus according to empirical formulas
(n is the number of input layers, m is the number of output layers, a is a constant of 1 … 10), the number of hidden layers is 5.
(3) Selection of output layer nodes
The design is based on a BP neural network PID controller. The neural network on-line learns to continuously adjust the parameters of the controller according to the running state of the system so as to achieve the actual expected output value, and the output of the neuron corresponds to three adjustable parameters k of the controllerp,ki,kd。
(4) Given a sample pair of input and output, the output of the network is calculated. Let the input layer input be I1(i) The output is O1(i) With weight from input layer to hidden layer of wijWith hidden layer input of I2(i) The output is O2(i) The weight from hidden layer to output layer is vijThe input of the output layer is I3(i) The output of the output layer is O3(i) In that respect Hidden layer and transportExcitation function selection S-function for layer
1) Calculating input and output of an input layer:
2) calculating the input and output of the hidden layer:
computing input and output of an output layer
(5) Calculating an objective function J
Suppose EpIs an objective function at the pth group of inputs, then
In the formula: y iskp(t) is the output of the network after t times of weight adjustment when the p group of samples are input; k is the kth node of the output layer.
The objective function J (t) is used as an evaluation of the system on the network learning condition.
And (3) simulation comparison:
as can be seen from the input and output curves in fig. 7 and 8, the unoptimized BP neural network has a large overshoot at the early stage, the time for reaching the stability is 1.85s, and the tension fluctuation of the system before reaching the stability is large, so that the requirement of the system on tension control is not met; after GA optimization, the overshoot of the BP neural network is reduced compared with that of the neural network, the time for achieving stability is also reduced to 0.85s, system fluctuation is obviously reduced, and the overshoot still exists; and the BP neural network after the AGA optimization eliminates the overshoot of the system, namely eliminates the early fluctuation of the tension system, and the time for reaching the stability is 0.66 s. Compared with the BP neural network and GA optimized algorithm, the adaptive genetic algorithm can better optimize the BP neural network, the optimization algorithm can eliminate overshoot of the system, the convergence speed of the system is improved, and the quick response and the stability of the system are further improved.
The STC12 chip can be replaced by an ATmega16 type single chip microcomputer, the chip is provided with 8 paths of 10-bit successive approximation ADC conversion, the structure of hardware can be optimized to a certain extent, and the cost is reduced. The optimized BP neural network algorithm can be replaced by a fuzzy neural network algorithm, and the same effect can be achieved. The invention is not only suitable for the wool spraying machine, but also can be applied to textile machines such as a bobbin winder and the like. The device not only can be used for adjusting the tension of filament, also can be used for adjusting the tension of yarn, fibre etc. through the different range of changing tension sensor, alright in order to satisfy actual production demand.
The filament constant tension control system is low in cost, simple in structure and stable in operation. The tension value of the filament can be detected in real time, different tension threshold values can be set for different types of filaments, the problem of controlling the filament tension of the wool spraying machine is solved, the wool spraying machine is popularized and applied to small and medium-sized textile enterprises, and the enterprise cost is saved. The tension control method adopts the BP neural network algorithm optimized by the adaptive genetic algorithm, and the adaptive genetic algorithm has the advantages of not needing to repeatedly test genetic parameters and saving a large amount of experimental time. The genetic algorithm optimization is mainly the randomness of initial value selection of the neural network, so that the condition that the output of the neural network falls into a local extremum can be avoided. The optimized algorithm can eliminate the overshoot of the system, improve the convergence speed of the system and further improve the quick response and stability of the system.
Claims (4)
1. The filament constant tension control method based on the optimized BP neural network is characterized by comprising the following steps:
step 1, constructing a filament constant tension control hardware system based on an optimized BP neural network;
step 2, determining a BP neural network structure according to the complexity of the system;
step 3, inputting sample data and an expected output value of the neural network;
step 4, initializing a population of the weight and the threshold and calculating a population fitness value, performing selection, crossing and mutation operations on an initial value and the threshold of the neural network, performing preferential elimination according to the fitness value after genetic operation, judging whether the steps need to be performed again according to a set genetic algorithm finishing condition, returning if necessary, and returning the optimized initial value and threshold to the neural network if not necessary; through the algorithm principle of the neural network, information forward propagation and error backward propagation are carried out, weight and threshold values are continuously modified, a weight matrix and a threshold matrix in the optimized BP neural network structure are described into chromosome strings, whether the system reaches the required precision is judged as a finishing condition, if yes, the algorithm is finished, and if not, the algorithm is returned to continue.
2. The filament constant tension control method based on the optimized BP neural network according to claim 1, wherein the filament constant tension control hardware system based on the optimized BP neural network in the step 1 has a specific structure as follows: the tension detection module is a tension sensor, adopts a JZHL-T1 three-roller type, is arranged between a yarn drum filament and a tension adjusting device and is used for detecting an initial tension value of the filament, the tension adjusting module adopts a lever type, the detected tension value is compared with a set tension threshold value through a main control chip to act, and the motor driving module adopts a stepping motor to drive the adjusting device to move up and down; the man-machine interaction module is used for LCD display and LED buzzer alarm; the main control module adopts a single chip microcomputer of STC12C5A08S2 type and is used for comparing the signal detected by the tension sensor with the threshold value and processing the signal of the LCD display screen.
3. The filament constant tension control method based on the optimized BP neural network according to claim 2, wherein the optimized BP neural network structure in the step 2 is specifically as follows:
comprises an input layer, a hidden layer and an output layer;
wherein, the input layer nodes are selected as follows: linear velocity v of winding mechanism2The friction force f between the filament and the pressure rod, the speed v of the pressure rod moving up and down0And the feed speed v of the thread1These four quantities serve as input layers for the neural network;
the hidden layer nodes are selected as follows: the hidden layer carries out threshold value filtering on the input signal and transmits the signal to the output layer, the output layer feeds the error signal back to the input layer through the hidden layer, and the error signal is obtained according to an empirical formula
n is the number of input layers, m is the number of output layers, a is a constant of 1 … 10, and the number of hidden layers is 5;
the output layer nodes are selected as follows: the neurons of the output layer correspond to three adjustable parameters k of the controllerp,ki,kd，kpThe proportion adjustment coefficient is used for accelerating the response speed of the system and improving the adjustment precision of the system; k is a radical ofiEliminating residual error for integral regulating coefficient; k is a radical ofdThe dynamic performance of the system is improved for differentiating the tuning coefficient.
4. A filament constant tension control method based on an optimized BP neural network according to claim 3, wherein the step 4 is as follows:
selecting a 4 x 5 x 3 neural network structure to adjust the linear velocity v of the winding mechanism2The friction force f between the filament and the pressure rod, the speed v of the pressure rod moving up and down0And the feed speed v of the thread1These four quantities serve as input samples for the neural network; motor control parameter k to be output desirablyp,ki,kdAs output, because the initial weight and threshold of the neural network are randomly generated, the randomly generated weight and threshold are sent to the genetic algorithm as the initial population of the genetic algorithm, and the genetic algorithm is based on the objective function of the neural network
Calculating fitness value of network genetic algorithm, Ep(t) the p-th group is output after t times of weight adjustment, a selection operator is acted on a population, the selection aims to directly inherit an optimized individual to the next generation or generate a new individual through pairing and crossing and then inherit the new individual to the next generation, the crossing operation is to mutually exchange partial genes of two paired chromosomes according to a certain mode so as to form two new individuals, the mutation operation is to simulate the gene mutation in the biological evolution process, the gene mutation occurs at a certain part on a chromosome string and becomes a unique genotype in the population, the termination condition is set as the termination algorithm when the evolution algebra is 100, if the termination condition is not met, the genetic operation is continued, and the parameters of the genetic algorithm are set: determining the evolution algebra of the population as 100, the size scale of the population as 50, and setting the initial value of the adaptive parameter as the cross probability Pc0.6, probability of mutation PmDecoding the optimized initial value and the threshold value to be 0.05 and sending to a neural network; the neural network continuously compares the expected output correction weight value with the threshold value until k is metp,ki,kdThe requirements of (3).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011147183.1A CN112698568B (en) | 2020-10-23 | 2020-10-23 | Filament constant tension control method based on optimized BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011147183.1A CN112698568B (en) | 2020-10-23 | 2020-10-23 | Filament constant tension control method based on optimized BP neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112698568A true CN112698568A (en) | 2021-04-23 |
| CN112698568B CN112698568B (en) | 2022-11-25 |
Family
ID=75506392
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011147183.1A Expired - Fee Related CN112698568B (en) | 2020-10-23 | 2020-10-23 | Filament constant tension control method based on optimized BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112698568B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113848706A (en) * | 2021-09-13 | 2021-12-28 | 无锡宏源机电科技股份有限公司 | Silk thread tension detection method, control method and control device |
| CN114355848A (en) * | 2021-12-20 | 2022-04-15 | 淮阴工学院 | Tension detection and intelligent control system |
| CN115438592A (en) * | 2022-11-08 | 2022-12-06 | 成都中科合迅科技有限公司 | Industrial research and development design data modeling method based on system engineering |
| CN116281422A (en) * | 2023-05-18 | 2023-06-23 | 青岛宏大纺织机械有限责任公司 | Control method and system of tension controller of bobbin winder |
| CN118439460A (en) * | 2024-07-04 | 2024-08-06 | 浙江恒逸石化有限公司 | Doffing path planning method, device, equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101308363A (en) * | 2008-07-07 | 2008-11-19 | 浙江大学 | Beam dyeing machine cloth velocity-tension coordinating control method |
| CN101338473A (en) * | 2008-08-07 | 2009-01-07 | 浙江大学 | Warp tension control method of loom |
| CN103076743A (en) * | 2012-12-28 | 2013-05-01 | 西安建筑科技大学 | Tension fuzzy PID (Proportion Integration Differentiation) control method for recoiling machine |
| CN203259897U (en) * | 2013-04-26 | 2013-10-30 | 上海工程技术大学 | Tension force control system of wire feeding machine |
| CN105404153A (en) * | 2015-12-17 | 2016-03-16 | 吉林大学 | Coil winding machine control method based on BP nerve network, and coil winding machine |
-
2020
- 2020-10-23 CN CN202011147183.1A patent/CN112698568B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101308363A (en) * | 2008-07-07 | 2008-11-19 | 浙江大学 | Beam dyeing machine cloth velocity-tension coordinating control method |
| CN101338473A (en) * | 2008-08-07 | 2009-01-07 | 浙江大学 | Warp tension control method of loom |
| CN103076743A (en) * | 2012-12-28 | 2013-05-01 | 西安建筑科技大学 | Tension fuzzy PID (Proportion Integration Differentiation) control method for recoiling machine |
| CN203259897U (en) * | 2013-04-26 | 2013-10-30 | 上海工程技术大学 | Tension force control system of wire feeding machine |
| CN105404153A (en) * | 2015-12-17 | 2016-03-16 | 吉林大学 | Coil winding machine control method based on BP nerve network, and coil winding machine |
Non-Patent Citations (1)
| Title |
|---|
| 朱耀麟 等: "模糊神经网络的长丝恒张力控制系统设计", 《单片机与嵌入式系统应用》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113848706A (en) * | 2021-09-13 | 2021-12-28 | 无锡宏源机电科技股份有限公司 | Silk thread tension detection method, control method and control device |
| CN113848706B (en) * | 2021-09-13 | 2024-01-12 | 无锡宏源机电科技股份有限公司 | Silk thread tension detection method, control method and control device |
| CN114355848A (en) * | 2021-12-20 | 2022-04-15 | 淮阴工学院 | Tension detection and intelligent control system |
| CN115438592A (en) * | 2022-11-08 | 2022-12-06 | 成都中科合迅科技有限公司 | Industrial research and development design data modeling method based on system engineering |
| CN115438592B (en) * | 2022-11-08 | 2023-01-24 | 成都中科合迅科技有限公司 | Industrial research and development design data modeling method based on system engineering |
| CN116281422A (en) * | 2023-05-18 | 2023-06-23 | 青岛宏大纺织机械有限责任公司 | Control method and system of tension controller of bobbin winder |
| CN116281422B (en) * | 2023-05-18 | 2023-08-18 | 青岛宏大纺织机械有限责任公司 | Control method and system of tension controller of bobbin winder |
| CN118439460A (en) * | 2024-07-04 | 2024-08-06 | 浙江恒逸石化有限公司 | Doffing path planning method, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112698568B (en) | 2022-11-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112698568B (en) | Filament constant tension control method based on optimized BP neural network | |
| CN105404153B (en) | A kind of coil winding machine control method and coil winding machine based on BP neural network | |
| CN109766989B (en) | Intelligent configuration method for technological parameters in polyester fiber production process | |
| CN110531727B (en) | Multi-stage optimization control method for polyester fiber production and manufacturing whole process | |
| CN101338473B (en) | Warp tension control method of loom | |
| CN112130453B (en) | Control method and system for improving MCS production stability based on machine learning | |
| CN109898200A (en) | A kind of ring throstle intelligent control method and control system | |
| CN112651928B (en) | Polyester filament yarn uniformity online detection system based on dynamic convolution neural network | |
| CN115328062B (en) | Intelligent control system for spunlace production line | |
| CN106868660A (en) | A kind of roller draft analogy method arranged in sliver based on fiber | |
| CN113848706B (en) | Silk thread tension detection method, control method and control device | |
| CN113755978B (en) | Automatic joint robot compliance device based on tension feedback | |
| Xiao et al. | Research on integral separation control of warp tension based on fuzzy parameter optimization | |
| CN106599431A (en) | FDY (fully drawn yarn) spinning technology for optimizing multi-model method based on mixture Gaussian weighting function | |
| CN117000780A (en) | Self-adaptive cold-rolled strip multichannel shape forecasting method | |
| CN115933564B (en) | Textile yarn production process monitoring regulation and control system | |
| CN102011220B (en) | Fuzzy-controller-based autolevelling control system and control method | |
| CN112831873B (en) | Independent process independent control system and method based on ring spinning yarn quality | |
| CN104529149B (en) | A kind of system and method for the intelligent on-line automatic regulation rubbing patterns of FSU | |
| CN115034470A (en) | Textile scheduling method based on evolutionary neural network | |
| CN212293896U (en) | Monitoring leveling device of combing machine | |
| CN203451691U (en) | Roving testing machine | |
| CN108107855B (en) | Class brain is layered the Intelligentized control method that fiber under memory mechanism stretches link | |
| CN102877170B (en) | Spinning tension adjustment curve learning device and leaning method | |
| CN115367549A (en) | SSA-SQP-based chemical fiber winding constant residual tension control method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20221125 |