Research on the Control Strategy of Battery Roller Press Deflection Device by Introducing Genetic Algorithm to Optimize Integral Separation PID

At the site of continuous rolling of lithium battery electrode, most of its correction units use traditional PID control algorithms, coupled with the complex structure of the rolling equipment and the large number of control system actuators, result in the deviation of the pole piece position being controlled within 10mm only. In order to avoid the undesirable industrial problems caused by untimely correction and insufficient control accuracy, this paper proposes a control strategy for lithium battery roll press deflection device with the introduction of GA(Genetic Algorithm) optimized integral separation PID. For the problems of poor self-adaptive capability of deflection control and interference with equipment and environment during the rolling process. Analyze the causes of runout of the pole piece and the structure of the deflection correction system, establish a mathematical model and an integral separation fuzzy PID controller to realize the deflection control. Optimize the key parameters of the controller using GA to improve the control accuracy and performance of the deflection control. Simulations and experiments show that the deviation control strategy proposed in this paper can reduce the position deviation of the polar strip to within 4mm, which can effectively improve the deviation correction accuracy and the anti-interference capability of equipment operation, and also provide a good solution for the deviation control of other large and complex mechanical equipment.

rolling unit and other mechanical units of the roller press [1]. 23 The deviation correction system mainly has the following 24 functions in the smooth rolling of the pole piece [2], [3], [4]: 25 The associate editor coordinating the review of this manuscript and approving it for publication was Zhong Wu . 1. Avoid the deviation of the pole piece during the pro- 26 duction process and ensure the smooth running of the 27 pole piece on the main roll and drive roll. 3. Making the traction force of the pole piece parallel to 32 the center line of the driving roller, and ensuring uni-33 form tension of the winding and unwinding pole piece. 34 The current correction control strategy is more traditional, 35 not only the accuracy method is not high, but also the ability 36 of adaptive adjustment, and the anti-interference ability is 37 accuracy. Deviation between the selected initial parameters 2. In addition to the position of the pole piece belt, the 96 deviation correction part of the pole piece rolling mill 97 also has a certain coupling relationship with various 98 parameters such as guide rollers, tension and vibration. 99 The existing control method is difficult to solve the 100 interference of the equipment itself and the external 101 environment when the pole piece is rolled, and needs 102 to be further optimized and upgraded; 103 3. Most of the above methods are for the analysis and 104 research of strip steel. 105 The existing deviation control methods are robust control, 106 adaptive control, fuzzy control, neural network and PID con-107 trol. The PID control method is simple and widely used in 108 industry. With the fuzzy algorithm which does not need to 109 establish an accurate mathematical model and has excellent 110 anti-interference ability, it can be well applied to the tension 111 control process of rolling mill. Therefore, this paper studies 112 and analyzes the rectification mechanism and improvement 113 method of the pole piece in the pole piece rolling mill, and on 114 this basis, designs a kind of rectifying control method suitable 115 for the pole piece rolling mill, that is, the fuzzy PID control 116 method based on genetic algorithm(GA) optimization. On the 117 basis of integral separation PID control algorithm, fuzzy 118 algorithm is introduced to realize on-line adjustment of K P , 119 K I and K D parameters in PID control by using relevant expert 120 knowledge, which can effectively improve the accuracy and 121 anti-interference ability of the control link. In order to avoid 122 the problem that the traditional optimization method is easy 123 to fall into the local optimal solution, the GA is introduced to 124 optimize the PID initial parameters and fuzzy rules. 125 The robust control has good performance against the inter-126 ference of the internal and external environment of the sys-127 tem, but its control accuracy is slightly lower than that of other 128 methods. Adaptive control has good performance in dealing 129 with external random interference and lag response method 130 of mechanical equipment, but the lag of mill deviation control 131 is not high, and the interference is not particularly strong in 132 the process of stability control, so this method is not selected 133 in this paper. Neural network has strong performance in self-134 learning and self optimization, and has good adaptability in 135 rolling mill tension control, but it is not convenient to write 136 the algorithm of the control system. Compared with robust 137 control and adaptive control, the control strategy proposed in 138 this paper can guarantee the control accuracy while counter-139 ing the internal and environmental interference, and is more 140 suitable for the deviation control of the pole piece. 141 This article takes the pole piece rolling machine's deviation 142 correction system and deviation correction flow as the start-143 ing point, on this basis, it focuses on the genetically optimized 144 fuzzy PID control method [9]. 145 The novelty of this paper lies in the establishment of the 146 model of the deviation correction system and the optimization 147 of the initial parameters and fuzzy rules in the integral sepa-148 ration fuzzy PID. In the process of genetic optimization, the 149 initial parameters are optimized first. When the fuzzy rules improve the accuracy and anti-interference ability of the 171 control method [11]; Finally, the performance of the algo- However, in the actual rolling process, due to the com-188 plex structure of mechanical equipment, more actuators of In the process of polar strip conveying, the cylindrical guide 195 roll is commonly used, but in the actual production process, 196 the shape of the guide roll may not be standard. Or in the 197 long-term conveying process of the polar strip, the guide 198 roller is worn due to the friction with the polar strip. These 199 factors will lead to the change of the shape of the guide 200 roll, such as bending, bulging or the change of the roll body 201 structure into cone roll. Taking the cone roll as an example, 202 this paper analyzes the impression of the guide roll geometry 203 on the offset of the polar strip: assuming that the roll shape 204 is intact, the polar strip is only subject to the traction force 205 perpendicular to the center line of the guide roll. When the 206 roll shape is conical, as shown in Fig.2, due to the uneven 207 force on both sides of the polar strip, a transverse offset force 208 will be generated on the polar strip from the place with the 209 smaller roll diameter to the place with the larger roll diameter. 210 When the lateral offset force is greater than the maximum 211 static friction between the roll surface and the pole piece, the 212 pole piece will shift to the larger roll diameter.

215
During the installation of the guide roll, due to the operator's 216 error, the guide roll and the horizontal line may no longer be 217 parallel, as shown in Fig.3. In this case, compared with the 218 above analysis method of cone roll, a lateral offset force Fx 219 will be generated. The offset force points to the higher side 220 of both sides of the conductor roll, which will cause the pole 221 piece to offset to this side.   In addition, during the operation of the equipment, the 260 uneven tension distribution of the pole piece belt has an 261 adverse effect on the deviation correction control of the pole 262 piece belt to a certain extent, which is mostly caused by 263 mechanical structure problems. However, with the use of 264 equipment, the change of mechanical structure is inevitable, 265 so the influence of tension is not considered in the process of 266 rectification.

268
In order to avoid the above-mentioned adverse effects, and to 269 ensure that the pole piece strip can accurately pass through the 270 roll when unwinding, and can be neatly passed through the 271 rectification mechanism during rewinding, the edge correc-272 tion system is often used in pole piece rolling mills to achieve 273 the pole piece The correction is shown in Fig. 8.

274
The electrode strip rectification system is mainly com-275 posed of rectification detection device, main control system, 276 rectification control device and electric actuator [12]. The 277 photoelectric or ultrasonic sensor is installed on one side of 278 the electrode strip as the reference. When the polar strip is 279 offset, the detection device will measure the offset distance 280 and angle of the strip edge, determine the degree of the strip 281 offset, and transmit the collected information to the main con-282 trol system. After analysis and processing, the main control 283  The rectification process of pole piece is shown in Fig.9 292 FIGURE 9. Schematic diagram of pole piece correction process.  1. The detection and comparison link mainly involves 318 position detection. Because the detected signal is an analog 319 signal, the analog signal L and the corresponding position 320 information H L can be converted into a proportional link 321 through the proportional coefficient K L . The specific transfer 322 function is shown in (1). Then, by comparing the collected 323 position information L f with the set value L c and making a 324 difference, the position deviation information L e is obtained, 325 as shown in (2).

OF DEVIATION CORRECTION SYSTEM
2. Motor drive link. In order to facilitate the subsequent 329 simulation model building and simulation analysis, this paper 330 simplifies the motor model and adopts a simple DC motor 331 model: In the equation, U a is the armature voltage, θ, L a and R a are 335 inductance and resistance of armature winding, E a is motor 336 back EMF, K e is motor back EMF coefficient, ω a is the motor 337 speed.

338
The electromagnetic torque of the motor is: In the Equation: K m is the torque coefficient of the motor. 341 The torque balance equation of the motor is as follows: In the Equation, J is the moment of inertia of the motor 344 shaft and f 0 is the damping coefficient.

345
In combination with the above (3) -(6), i a is eliminated by 346 transformation, and Laplace transformation is carried out to 347 obtain (7) In the Equation, The transfer function of voltage angle can be obtained by 352 simplification: Since the motor speed can be regarded as the differential 355 ω a = dθ dt of the motor angle, ω a (s) = sθ (s) can be obtained 356 by Laplace transform of the Equation. When y is brought into 357 the motor transfer function obtained above, the voltage speed 358 transfer function G v (s) as shown in (9) can be obtained. The 359 obtained transfer function is a typical second-order oscilla- we can get that: In the equation: τ 0 is the pure delay time.

378
By expanding e τ 0 s according to Taylor series, we can get 379 the following results: Becauseτ 0 is smaller, so only the first two terms are 382 retained for subsequent calculation, that is e τ 0 s ≈ 1 + τ 0 s.

383
In this case, combined with the proportional coefficient K s , 384 the simplified transfer function G s (s) of the whole mechanical 385 transmission link can be obtained as follows: The closed-loop transfer function of the rectifying system 388 of the lithium battery plate rolling mill can be obtained by 389 combining the above three parts of the transfer function: integral and differential calculation of the deviation signal to 406 output the control signal to control the motor drive device to 407 complete the position correction of the pole piece [13]. When 408 setting the three parameters of the PID controller K P , K I , K D , 409 a lot of relevant expert knowledge is needed, but most of these 410 parameter setting knowledge is a vague experience, which is 411 not conducive to the formation of clear parameters. The fuzzy 412 control is to express the experience of the expert's ambiguity 413 through the computer, which can effectively optimize the 414 three parameters in the PID controller [14].

415
In order to avoid the phenomenon that the position devia-416 tion signal of the pole piece belt is too large during the start, 417 stable operation and stop of the pole piece rolling mill, it will 418 have a greater impact on the integral link of the PID control 419 and produce an integral accumulation phenomenon. This arti-420 cle introduces the integral separation control method in the 421 PID control process. When the position deviation signal of 422 the pole piece is large, the stability of the system is improved 423 by removing the integral link in the PID control; When the 424 deviation signal decreases and is closer to the set value, the 425 integration link is re-introduced to eliminate the static error 426 in the correction control. At this time, the overall flow of the 427 integral separation fuzzy PID algorithm as shown in Fig.10 is 428 obtained [15].

430
During the start-up, stable operation and stopping of the 431 lithium mill, the position information of the pole piece is not 432 stable and constant, and there may be excessive deviation 433 of the pole piece position signal. At this time, the position 434 signal of the polar strip will produce a large change in a short 435 period of time, in the process of PID control, this situation 436 will have a large impact on the integration link, resulting in 437 the accumulation of integration phenomenon. Without chang-438 ing the control strategy, it will cause the corrective control 439 accuracy and speed to be reduced, and even the execution 440 equipment may oscillate under the influence of the output 441 signal, which in turn threatens the mechanical structure of the 442 mill. In order to avoid the possible bad consequences of the 443 above problems, this paper introduces the integral separation 444 control method in the PID control process. Improving the 445 stability of the system by removing the integral link in the PID 446 control when the deviation signal of the pole piece position 447 is large, When the deviation signal decreases and is closer to 448 the set value, the integration link is reintroduced to eliminate 449 the static error in the correction control. The operation flow 450 is shown in Fig.11.

451
When making a judgment on whether to introduce the 452 principle of integral separation, it is first necessary to set the 453 deviation threshold ω according to the actual control object. 454 When the controlled quantity exceeds the set value too much 455 and is greater than the set ω, the integral link in the PID 456 control is eliminated, and conversely when the controlled 457 quantity is close to the set value and is less than the set ω, 458 the integral link is reintroduced to complete the proportional 459 integral differential control. By bringing the parameter β 460   The method is as in (14) (15).  12,12], so its basic domain is determined to be [ −12,12]. 506 Increasing the number of elements in the universe can 507 improve the control accuracy, but it increases the amount of 508 calculation, and the improvement of fuzzy control effect is 509 not obvious. When the above fuzzy subset is selected, the 510 universe of e and e c can be selected as {−6, −4, −2,0,2,4,6}. 511 In actual control, the values of input and output are gener-512 ally not elements in the universe. In this case, it is necessary to 513 transform the universe through quantization factor and scale 514 factor. Generally speaking, the basic universe of the system 515 input is: the error e(t) is [−e, e], the error change rate e c (t) is 516 [−e c , e c ], and the basic universe of the output variable is [−u, 517 u]. According to the basic domain and calculation formula of 518 input and output, the quantization factors k e , k ec and scale 519 factor k u can be obtained: where:  Table 1. K P should be selected so that the deviation correction 560 system can respond quickly; at this time, if the rate of 561 change of the deviation value e c is too large, the K D 562 is appropriately reduced to weaken the influence of the 563 differential link [19]. 564 2. When the pole piece deviation e is centered, the value 565 of K P can be increased appropriately to increase the 566 effect of the integration link; the value of K D should 567 be appropriately reduced to avoid overshoot and oscil-568 lation; K I should be a small value to avoid excessive 569 static errors. 3. When the pole piece deviation e is small, in order to 571 avoid the overshoot and oscillation effect caused by 572 the integration link, the value of K P should be appro-573 priately reduced. At the same time, the value of K I 574 should be appropriately increased to eliminate static 575 errors; at this time, if the error change rate e c of the 576 correction system is small, in order to reach a stable 577 state as soon as possible, the value of the integration 578 coefficient K I should also be appropriately reduced. 579 If the value of e c is medium or large, K D takes a 580 moderate value to stabilize the differential link. 581 VOLUME 10, 2022 the fuzzy rules shown in Table 2 can be established.  Combined with the transfer function model of the correc-609 tion structure obtained by the analysis, the integral separation 610 fuzzy PID simulation model shown in Fig.15 is constructed 611 in Simulink. In the figure, the simulated step signal is used as 612 the input, and the performance of the controller is analyzed by 613 analyzing the response parameters such as the response rate 614 and overshoot of the output waveform.

617
The previous research on the correction control algorithm 618 involved the selection of PID initial parameters and the 619 construction of fuzzy rules. However, in the selection pro-620 cess of these parameters, most of the application is expert 621 experience combined with simulation debugging, PID ini-622 tial parameters and fuzzy rules may have certain deviations. 623 Genetic algorithms do not rely on specific inference func-624 tions, but rather improve the adaptive capacity of their own 625 populations from the perspective of probabilistic and genetic 626 properties by simulating the evolutionary alternation of natu-627 ral populations [20]. Therefore, in this paper, GA is added 628 to optimize a large number of the above parameters and 629 improve the selection accuracy of PID initial parameters and 630 fuzzy rules.

632
In the application process of GA, first of all, we must deter-633 mine the problem to be solved, and then build a basic popula-634 tion by randomly generating individuals; after that, the fitness 635 function is constructed to analyze and evaluate the number 636 of codes in the population, to identify excellent individuals 637 and eliminate the inferior individuals with insufficient fitness; 638 finally, by selecting, crossing and mutating the remaining 639 In the Equation: u(t) is the output of the closed-loop control 703 model in PID control method, e(t) is the closed-loop control 704 error, tu is the rise time, ω1, ω2 and ω3 is the weight coeffi-705 cient.

706
The fitness function of the above Equation is programmed, 707 in which the weight coefficient ω1, ω2 and ω3 is set to 0.999, 708 0.001 and 2 respectively. The selection operation is performed by ranking the values of 711 fitness, retaining the individuals with higher fitness among 712 them, and introducing new individuals for generating new 713 populations for subsequent genetic operations, which are 714 performed here using the betting roulette method.

715
Before performing the crossover and variation opera-716 tions, the crossover probability P c and variation probability 717 VOLUME 10, 2022  The specific calculation method is as follows.
where: f max is the maximum value of population fitness; f avg 726 is the mean value of population fitness, f is the larger value 727 of fitness for the two individuals in the crossover, f' is the 728 value of fitness for the individual performing the mutation 729 operation, k 1 , k 2 , k 3 and k 4 are constants, and k 2 >k 1 and 730 k 4 >k 3 are guaranteed. We set the parameters to k 1 = 0.66, 731 k 2 = 0.99, k 3 = 0.001, k 4 = 0.1. In MATLAB, the genetic operation is carried out through the 751 operation flow described above. In the course of 100 genetic 752 iterations, the change curve of the optimal individual fitness 753 value of each generation is shown in Fig.17. From the figure we can see that the fitness value reaches 755 stability around 30 generations. Therefore, in order to ensure 756 the output of the optimal PID initial parameters, the optimal 757 TABLE 3. The optimal solution of PID parameters for pole rectification after 100 genetic iterations.
individual after the population is inherited for 100 generations 758 is selected as the initial PID parameter. The initial parameters 759 of PID optimized by GA are accurate to 3 decimal places, 760 so there will be some differences in the initial parameters 761 after the optimization is completed. In this paper, 10 different 762 populations are constructed. After multiple iterations of opti-763 mization, 10 sets of K P , K I , and K D parameters are obtained.

764
The specific values are shown in Table 3.

765
By averaging the values in Table 3, the initial PID parame-  Fig.18(a). The obtained optimal 773 individual genome is extracted to obtain the optimized fuzzy 774 rule code, and Fig.18(b) is a histogram representation of the 775 optimal fuzzy rule code. By extracting the genes in Fig.18(b), it is divided into  Table 4 can be obtained. Before the simulation, we need to confirm the initial param-787 eter settings of PID control. In this paper, the three initial 788 parameters of PID, K P , K I and K D are obtained by test and 789 comparison. They are 7, 1 and 2 respectively. The parameters 790 of the PID controller were obtained using an intelligent soft 791 computing technique, but are omitted in the paper for typo-792 graphical and other reasons.In the simulation analysis, the 793 parameters of the PID controller were determined based on 794 the Ziegler-Nichols critical scale factor method using the root 795 trajectory function rlocus as well as the rlocfind command on 796 the basis of the MATLAB platform.Based on this, we carry 797 out the comparative study of the three methods. Taking step 798 signal as input signal, sinusoidal signal and smaller step 799 signal as interference, and setting threshold value to 5, the 800 integral part in PID is eliminated.

801
In this paper, the traditional PID, integral separation PID 802 and integral separation fuzzy PID control methods are com-803 pared and simulated, as shown in Fig.19. In the simulation 804 comparison of the different control methods, the three initial 805 parameters of the PID are set to be the same. The input signal, 806 the value of the disturbance signal and the time of joining 807 are all the same. Simulation analysis is performed under the 808 premise of ensuring the same conditions. And there is no 809 environmental interference in the simulation analysis, which 810 can ensure the fairness of simulation analysis comparison. 811 At the same time, in the simulation process, a small step 812 signal is introduced at 10ms to simulate the small interfer-813 ence in the process of deviation correction control. In the 814 actual correction process, large signal interference may occur, 815 so this paper introduces a sinusoidal signal to simulate the 816 large interference that may occur in the correction control 817 process. At this time, Fig.20 is obtained.

818
By comparing and analyzing Fig.19 and Fig.20, we can 819 see that under the same control signal, the response rates of 820 the three methods are relatively close. However, by analyzing 821 the subsequent overshoot oscillations and the time to finally 822 reach steady state, we can see that the effect of integral 823   to the initial step signal will be faster, at the same time, its 842 overshoot is much smaller than the curve corresponding to the 843 original setting parameters, and it also reaches a stable state 844 more quickly; after the disturbance is introduced, the curve 845 obtained after optimization can also return to the set step sig-846 nal value more quickly, which has stronger anti-interference 847 ability.

848
In order to verify the performance of optimizing fuzzy 849 rules, the old and new fuzzy rules are substituted into the pole 850 piece correction simulation model, respectively, and Fig.22 is 851 obtained. Fig.22(a) is the response curve corresponding to the 852 two rules under the original PID parameters, and Fig.22(b) 853 is the response curve under the optimized PID parameters. 854 It can be seen from the information in the figure that the initial 855 response rates of the two curves are approximately the same, 856 but in the process of subsequent curves returning to stability, 857 the response curve corresponding to the optimization rule 858 has a smaller overshoot and the curve state is stable faster. 859 Under the two PID parameters, the simulation performance 860 of the optimized fuzzy rules is better than the original fuzzy 861 rules. It can be seen that the optimized fuzzy rules have good 862 adaptability to the changes of PID parameters. The experimental platform adopts the embedded integrated 867 measurement and control system, uses the self-designed 868 embedded control motherboard as the main controller, and 869

889
When the main roll speed is set at 30m/min, the experimental 890 equipment can run stably. When the roll speed is more than 891 40 m/min, the electrode strip will break and shake due to 892 the excessive tension, which makes it impossible to control 893 the tension accurately. When the roll speed is lower than 894 10m/min, the strip will be relaxed and wrinkled due to the 895 low tension, and the tension cannot be controlled accurately. 896 Now in the actual industrial environment, the rolling speed 897 of lithium battery is 15∼30m/min, which can fully meet the 898 experimental requirements. In the experimental verification, the unwinding correction 902 structure was taken as an example for analysis. In the exper-903 iment, the position of the correction sensor was aligned, and 904 the alignment position at this time was set as the standard 905 value of the pole piece position, corresponding to the value 906 0. When the position of the pole piece belt changes, the cor-907 responding standard value is taken to be positive or negative. 908 After the equipment runs stably, collect the edge information 909 of the pole piece in the rectification control and obtain the 910 traditional PID control algorithm. The position information 911 of the unwinding pole piece is shown in Fig.25(a). After intro-912 ducing this algorithm, the pole piece position information is 913 shown in Fig.25(b).  The initial reference value is also set to 0, and the range of 946 lithium battery pole piece correction is set within±50mm. 947 Under the two control algorithms, 10 groups of tension values 948 are collected respectively, and the measured value of pole 949 position and its error comparison are shown in Fig.27. Compared with the collected position information of the 951 winding pole piece, it can be seen that the position of the pole 952 piece belt fluctuates greatly in the traditional PID algorithm, 953 so the pole piece position may shift too much in the operation 954 process, thus causing an alarm; In contrast, the position of 955 the pole piece corresponding to this algorithm fluctuates less, 956 there is no sampling point approaching the limit value of the 957 rectification range, and the average position deviation is less 958 than that of the traditional PID algorithm, so the stability and 959 control accuracy are significantly higher.

961
The genetic optimization integral separation fuzzy PID 962 control strategy can greatly improve the stability and 963 anti-interference ability of the system. Compared with the 964 dual chip architecture of 'ARM+DSP', the control strategy 965 can achieve the requirements of high accuracy on the premise 966 of using only one main control board for control through 967 the optimization design of the control algorithm, and save 968 more economic costs. While the deviation prediction error 969 proposed by G. Wang was basically maintained at ± 15mm. 970 In contrast, the control strategy proposed in this paper can 971 control the deviation within ± 4mm. Therefore, it has signif-972 icant advantages in control cost and control accuracy.