Design of Pretension Tubular Rope Machine Control System Based on RBFNN Tuning PID

Transcription

1 06 International Conference on Coputer, Mechatronics and Electronic Engineering (CMEE 06) ISBN: Design of Pretension ubular Rope Machine Control Syste Based on RBFNN uning PID Hui LI and Ning ZHENG School of Autoation & Electric Engineering, ianin University of echnology and Education, ianin 300 ianin Sino-Geran University of Applied Sciences, ianin , China Keywords: Pretension, PLC, SM3, RBF, PID. Abstract. he pretension process is a coplicated echanical and electrical control syste, and the tension is the ain factor that affects the quality of the rope. Aiing at the proble that the traditional PID controller is difficult to achieve the ideal control effect, the RBF neural network PID self-tuning control odel is proposed. It realizes the on-line autoatic adustent of the controller paraeters, which ensures the constant tension operation effect of the pretension process. Introduction With the uch stronger quality requireents for steel wire rope fro custoers, steel wire rope anufactures are iproving their production equipent, under this background, the corresponding anufacturing and control technology has advanced. Pretension tubular rope achine is iproved, which eets the wire rope structure and specifications. Pretension processing functions is added to the original equipent, coplete the pay-off, stranding, rope laying, pretension, and spooling wire, and other aor process, and then once fored rope aking can be achieved online. In the new equipent, the intelligent tension control syste plays a key role in the iproveent of wire rope processing quality. General Structure of the Unit he general structure of the pretension tubular rope achine is as shown in Fig.. he unit consists of one or ore rotors, pay-off reels, pressed wire ould, lay pitch plate (traction gearbo), the post-deforing device, double traction wheels and, tension detectors, oil spray devices, devices, spooling devices, the drives syste, s and electrical control syste ain ;-rotors;3-payoff reel;4-drive shaft;5-noseplate;6-die;7-lay pitch plate;8-double traction wheel ;9-tension control ;0-pretension detection;-double traction wheel ;-ill spray;3-wire ;4-wire ; 5-wire spooling ;6-spooling Figure. Unit of tubular closer for pretension wire rope aking. Electric Control Syste he electric control syste of the pretension tubular rope achine is as shown in Fig., which

2 consists of the PLC, frequency, secondary instruent and so on, and establishes the HMI outside the control cabinet by using a touch screen to facilitate user s anageent and operation. Four variable frequency s, i.e. the ain, stretching, and spooling, are used in wire rope aking process, which are respectively controlled by a frequency converter, and the control odes and speed are deterined according to the speed of the working parts of the strander or the relationship between the working parts and the relative technical data of the wire ropes to be processed. In the control syste, hardware selection depends on the control accuracy, counication speed, response tie, perforance price ratio and reliability. A Sieens S7-00 PLC (CPU4XP) is used as the control center [], and the frequency is a YASKAWA vector control A000 [], and the touch screen is P70 series of Sieens. Besides ipleenting conventional logical control function, such as starting, braking, resetting, forward or reversely rotating, ogging, interlock protection, fault display and so on, the syste counicates inforation with frequency s and SM3 by eans of RS485 interface in accordance with MODBUS protocols in the network counication interface [3]. he PLC sends control or query coands to s, and the s answer. Considering the PLC ipleentation of neural network calculation is ore difficult, so the neural network algorith is realized by ARM SM3, and RBF network tuning PID closed-loop control of tension can be perfored by using its A/D, D/A and calculation function. torque preset tension transducer AQ AI DO 0 DO 3 transitter touch screen RS-485 CPU4XP EM3 EM DI 4 DI 3 DI 6 D/A A/D SM3 RS-485 length-easured pulse signal ain ain stretching stretching speed preset spooling spooling torque preset spooling speed output display Figure. Control syste structure. By using a touch screen, users can input and set the speed, diaeter, lay pitch, tension and length, and can display the running frequency, current, speed, tension and real length. he length-easured device, which easures the strands length, sends a pulse signal one half eter. he ain, stretching and spooling need rotary encoders, and the type is E6B-CWZ6C-000P (OMRON). Pretension Control Pretension is to apply a certain tensile load on the wire rope in order to partly or copletely eliinate the etra structural elongation caused by defects of lay or twisting and frequent load of the rope, and to iprove the strand force unifor, and to iprove the odulus of elasticity and fatigue life of the rope. he stretching part is coprised of the stretching variable frequency, stretching frequency

3 , rotary encoder, double traction wheel and, tension detector and controller. he spun strands are drawn forwards by the traction devices. Double traction wheel is driven by the stretching, and its rotation speed is a little higher than double traction wheel in order to perfor the pretension operation [4]. In this production line, each operating conditions are different. If using the conventional fied paraeter PID control, the control effect is not ideal and the control quality will decline, it is difficult to eet the requireents of high precision achining [5]. However, Based on the radial basis function (RBF) neural network PID paraeters on-line adustent can obtain satisfactory constant tension control effect [6]. RBF Neural Network uning PID Algorith Structure of RBF Neural Network he structure of RBF neural network (RBFNN) is regularly forulated with three-layer feed-forward network [7], which are the input, hidden and output layers. Fig. 3 shows the structure of the general RBFNN. h w h w y n w h Figure 3. Structure of RBFNN. In the RBF network structure, X,, n is the network input vector, H h, h, h is the radial basis vector, where h is Gaussian function and can be generated by: X C h ep b (=,,,) () where b is baseband paraeter, C is the center vector of th node in hidden layer. he weight vector for the network is W w, w, w. he network output y can be defined as: y k wh. () he perforance inde function of RBF is J I y k yk. (3) According to gradient descent algorith, weights iterative algorith, node center and radial paraeters are as follows: w k w k y k y k h w k w k (4) b y k y k w h X C b 3 (5)

4 b k b k b b k b k (6) c y k y k w i c c k c k c c k c k i i i i i b i where η is the learning rate and α is the oentu factor, the network structure 3-6- is used. Jacobian atri (he sensitivity of obect output to controlled input) algorith can be shown as (7) (8) y k y k ci wh where u k u k b u k. (9) RBF Neural Network uning PID Controller he PID control based on the RBF network is coposed of two parts of the RBFNN identifier and the PID controller. he RBFNN identifier identifies the approiate odel of the controlled obect by the input and output data of the controlled obect, and then replaces the input and output relationship of the controlled obect. he paraeters of PID controller are realized by RBFNN identifier. RBFNN tuning PID control block diagra shown in figure 4. r + -e PID u Plant y + RBF - y Figure 4. Structure of the RBFNN tuning PID controller. Deviations fro the desired output can be epressed as: ek r k y k. (0) where r(k) is the syste reference input, y(k) is the syste output and e(k) is the syste error. he three inputs of the PID are:. () k e k e k k e k. () 3 k ek ek ek. (3) PID algorith is generally given as: u k k p k ki k k p3 k. (4) Moreover, k P, k I and k D are paraeters of PID controller. he error inde function is defined by

5 E k e k. (5) he three paraeters of the PID are odulated by gradient descent algorith E E y u y k e k k P I kp y u kp u (6) E E y u y k e k k D ki y u ki u. (7) E E y u y k e k k y u 3 kd y u kd u. (8) can be gained fro Jacobian atri. Stretching-copression force easureent transducers DMGZ300 is used to easure the strands tension, and the tension transitter produces electrical signals corresponding to the tension (0~ 50kN), and the current signal (4~0A) fro the transitter is fed to the AI channel of the PLC and SM3. Using the tension reference input and tension easureent and RBF neural network PID algorith, SM3 adusts the PID paraeters, and PID output is sent to the AI of the PLC via D/A, the PLC transits the torque reference value to the, whose control type is closed-loop vector control with speed transducers, and by speed liiting, controls the output frequency of the via Modbus counication to ake the strands oving speed between the two double traction wheels steady, and then the purpose of constant tension is achieved. In the process, selector switch is used to perfor tension or speed control which is only (anually) used when starting and reeving the strands, but tension control is used when noral autoatic running. Conclusions he control syste of the tubular closer adopts PLC, SM3, frequency converter, AC, tension detector, encoder and counication technology to realize closed loop vector control. In the syste, all the oving parts can coordinate the linkage operation, especially on the RBFNN self-tuning PID algorith, which ensures the control precision, thus greatly iproving the processing quality of the steel wire rope. References [] Sieens AG. S7-00 PLC Syste Manual. 008(8), [] Yaskawa Electric Corporation. Yaskawa AC Drive-A000 High Perforance Vector Control Drive echnical Manual. 04(), [3] Li, H., Wu, X., Master-slave Counication between S7-00 PLC and MCU Based on Modbus Protocol. Research and Eploration in Laboratory, 0,3(4): 8-84,69. [4] He, G., Wang, J., Electric control of pretension tubular strander. Metal Products, 0,37(): [5] Astro, K. and. Hagglund (995). PID Controllers: heory, Design, and uning. Instruent Society of Aerica. Research riangle Park, USA. [6] Cheng, L., Zhang, G., Wan, B., Hao, L., Qi, H., Ming, D., Radial Basis Function Neural Network-based PID Model for Functional Electrical Stiulation Syste Control,3st Annual International Conference of the IEEE EMBS Minneapolis, Minnesota, USA, Septeber -6, [7] Chen, Y., RBF based PID control and siulation, Coput. Siul. (0) -5.