Free and Open Source Software for Industrial Process
Contents
1. Fig 2 Input Amplifier The contact resistance is very noisy typical moving contact noise The input amplifier Fig 3 has high input impedance and uses a low pass filter R and C that cut out part of the noise from this sensor We provides two time constants Sms and 0 5ms compatible with expected closed loop dynamic of the plant With the high impedance of the JFET operational amplifier LF356 if the ball loose the electrical contact with the rail s the input filter works as sample and hold the capacitor C remember the last valid voltage position value This non linear behavior of a fully linear filter and amplifier is an collateral effect of the non linear random proprieties of the moving contact resistance that represent the internal impedance of our position sensor To have two moving contact reference rail ball and ball sensing rail in series aggravate the final results A common cure to reduce the moving contact noise is to use carbon based electrical conducting grease We don t use it intentionally because we want to evaluate the noise rejecting proprieties of the controllers in worst cases situations B Actuator For actuator we use the LEGO Mindstorm standard motor model 71427 The specifications of this motor are available 14 from experimental tests The inductance of the motor s armature L4 2 07mH average value measured with an impedance bridge is not included2. Ball and beam Nonlinear model gt Ball_Position Joti of 1 z gt Fig 3 Scicos diagram bb a cos with complete continuous non linear model of the plant and the linear discrete controller Adjusting the R and Q parameters of the controller is possible to change the dynamics of the transient response Fig 3 show the results of the simulation using R and Q values found experimentally to stabilize the controller B The regulator The regulator implemented inside a PC is a sampled data system but the real plant is a continuous time system Various approach are possible we choose to transform the continuous model of the plant a a discrete one and design the controller in the sampled domain The first things to do is choose a sampling time Usually the sampling time is chosen looking at the closed loop expected performances If Tr is the desired 10 90 response time to a step of the reference Ts Tr 10 or less In other words at last ten sample in the rise time The sampling time should be high enough to cover within the Fs 2 Nyquist limit all the fast poles of the plant in order to avoid intersampling oscillation It is important to check also the authority capability to act with efficacy of the actuator present in the real system In our plant the Lego motor has a intrinsic bandwidth in the rang
3. 4V dynamic range of the USB DUX output voltage is enough to drive the motor the output amplifier works only as current booster with voltage gain set to one D Encoder Input Using the bidirectional counter input of the data acqui sition card USB DUX is possible to connect an encoder measuring the real angle of the beam We have not used this input in the experiment because we choose to use an observer to recreate the full state of the system With an observer it is possible to avoid the necessity of an expensive sensor to measure the position of the beam E Data Acquisition Card We use the USB DUX 13 interface We choose this device because it has sufficient resources to monitor and drive the plant USB DUX uses a Universal Serial Bus connection his allow to use any recent laptop USB DUX uses Comedi 8 device driver suite that allows to use different hardware without a coherent API In the spirit of free and open source development the USB DUX is fully documented hardware and software including schematics and source code of the firmware of the micro controller that runs the board The only real limit of this board is the USB bus unfortunately it is not yet possible to run it in true hard real time fashion USB interfaces because to do so it is necessary to rewrite the full USB Linux stack 15 Usually with recent 2 6 1x kernel this is not a serious limitations for applications with sampling time of 10ms or higher At pr
4. RTAI also in the supervisor PC that run RTAI Lab It is possible to avoid this requirement the RTAI XML J RTAI Lab 7 allow to use a general purpose Java enabled web browser to implement the same features of RTAI Lab but without any garantee about the latency of the data exchange D Comedi device drivers and libreries Comedi 8 is the standard way to use I O interface de vices with Linux The Comedi project develops open source drivers tools and libraries for data acquisition Comedi is a collection of drivers for a variety of common data acquisition plug in boards more than 100 devices are supported The drivers are implemented as a core Linux kernel module providing common functionality and individual low level driver modules To use the driver the application program should use Comedilib functions Comedilib is a user space library that provides a developer friendly interface to Comedi devices Included in the Comedilib distribution is documen tation configuration and calibration utilities and demon stration programs Kcomedilib is a Linux kernel module distributed with Comedi that provides the same interface as Comedilib in kernel space suitable for real time tasks It is effectively a kernel library for using Comedi from real time tasks This is a very important features because the hard real time RTAI tasks could not issues direct system call to uses E Linux and RTAI Linux kernels The Linux kernel is the found
5. EFERENCES 1 Scilab Online Available http www scilab org 2 S L Campbell J Ph Chancelier et R Nikoukhah Modeling and Simulation in Scilab Scicos Springer 2005 Scicos Online Available http www scicos org 4 2 cap 12 Code Generation 5 Syndex Online Available http vww rocq inria fr syndex 3 fate 6 RTAI Online Available http vww rtai org Available http vww dti supsi ch bucher scilab howto pdf 7 RTAI XML Project Online Available http artist dsi unifi it rtaixml 8 Comedi Online Available http vww comedi org 9 Links to companies Online Available http www linuxdevices cony 10 Xenomai Online Available http vww xenomai org 11 Quanser Online Available http Avww quanser com TQ Education and Training Limited Online Available http Avww tg com 12 LEGO Mindstorm Online Available http mindstorms lego com 13 USB DUX Online Available http Avww linux usb daq co uk 14 LEGO motor parameters Online Available http vww philohome com motors motorcomp htm 15 USB RTAI stack works in progress Online Available http developer berlios de projects usb4rt 16 PC CARD DAS16 16 AO Measurement Computing Online Available www measurementcomputing com 17 2 cap 9 Scicos Blocks 18 Comedi Comedilib Tutorial and User Manual comedilib pdf Available http vww
6. Free and Open Source Software for Industrial Process Control Systems Simone Mannori Ramine Nikoukhah Serge Steer INRIA Rocquencourt Domaine de Voluceau 78153 Le Chesnay Cedex France simone mannori inria fr vamine nikoukhah inria fr Abstract Scilab and Scicos are open source and free soft ware packages for design simulation and realization of in dustrial process control systems They can be used as the center of an integrated platform for the complete development process including running controller with real plant Scicos HIL Hardware In the Loop and automatic code generation for real time embedded platforms Linux RTAI RTAI Lab RTAI XML J RTAI Lab These tools are mature working alterna tives to closed source proprietary solutions for educational academic research and industrial applications We present using a working example a complete development chain from the design tools to the automatic code generation of stand alone embedded control and user interface program Keywords Real time control dynamical systems simu lation simulation software hybrid systems Scilab Scicos RTAI Comedi I INTRODUCTION We show using a real example a complete development chain that allow to e construct a mathematical model of the plant e validate the model of the plant through simulation and open loop measurements e design a suitable controller e simulate the full system and optimize controller param et
7. ation over all this building is constructed The Linux kernel provides a level of flexibility and reliability simply impossible to achieve with any other free or commercial operating system For these reasons many companies in the field of real time and embedded operating systems 9 offer support for Linux The real time performances of the latest 2 6 1x Linux kernel are very close to real time for sampling time in the milliseconds range but the vanilla standard unmodified Linux is NOT a real time operating system because with the standard Linux kernel is not possible give guarantee about the absolute maximum latency of task activation There are many project that modify patch the Linux kernel in order to create a parallel hard real time kernel and user space where the tasks have deterministic activation timing while leave the other normal tasks running in the standard Linux user s space We use RTAI because offer direct support for Scilab Scicos but is also possible to use Xenomai 10 III CONSTRUCTION OF THE PLANT The fastest way to have a plant is to buy such ready made There are many companies on the educational mar ket that offer such products 11 We choose the classical ball and beam experiment Fig 1 a because its intrinsic instability and non linearity represents a recognized control system benchmark The construction is possible using readily available cheap parts For most of the me
8. chanical parts we use LEGO Technique Mindstorm 11 components less than 200 euro for a complete kit Fig 1 Complete system From the bottom a ball and beam b amplifiers and power supplies c USB DUX data acquisition card The ball is made of stainless steel The ball rolls on two brass rails salvaged from a train model and glued on the Lego bricks Fig 1 a We build using a single experimental printed circuit board Fig 1 b the input amplifier the power output amplifier and the relatives power supplies As data acquisition cards we use the USB DUX Fig 1 c 13 A Ball position sensor The ball position is measured using one of the rails as a variable voltage source Fig 3 A floating 5 Volts voltage and a series resistor supply a constant 700 mA current to the reference rail This arrangements is simple but inefficient almost all the electrical power is dissipated in the linear regulator and in the external 6 8 ohm 5 Watt power resistor To avoid this waste of electrical power we plan to use a constant voltage 100mV low noise switching power supply The metallic stainless steel ball works as a sliding contact the second rail closes the circuit sending the low level signal 0 40 mV equivalent to 0 30 cm to the input amplifier LF 356 12K 120 R R E Wi S68 Ro 0 2Volt O 47uF AS C i E heva 330K4270KS RI y 0 40 mV IT L_beam 30 cm Vv 10K R2 LN Vv KA
9. comedi org ANNEXE MATHEMATICAL MODEL OF THE PLANT The mathematical model of the mechanical system has three internal state the ball posizion x the ball s rotation angle and the beam angle O The equation relative to the ball dynamics is mg sin m J r where r is the ball radius meter m the ball mass Kg et Jy 2mr 5 the rotational moment of inertia Using the relation x or whit the hypothesis that the ball roll without slide the dynamics of the ball is simplified to 5gsin 0 7 A second equation model the dynamics of the beam T mg cos 0 x Jp mx76 where T represent the torque on the beam s axis Using the simplified model of the motor that don t consider the inductance Ty Kria where Ty is the motor torque Nm Kr the torque constant Nm A and i4 the motor current A Introducing the cinematic ratio of the gears that couple the motor to the axis of the bean Kc 40 8 y T Tu where Wy is the rotational speed of the motor axis 6 and the motor equation is Um Raia KyOy we obtain KcKria mgcos 0 x J 6 mx76 where _ Uy Kyomu ia _ Ra Substituting x par v we obtain the complete model Ky KeK KcKrU Gre ee cos gt ip a Jp mx Ra Jp mx Ra Jp mx 6 o0 5 j xgsin d v z8sin 0 x Vv The state of this system are 0 90 v0 x0 0 0 0 where is an arbit
10. d programmable logic architectures use a specific code generator for Linux RTAI This solution merge the hard real time capability of Linux RTAI kernel 6 with the creation of a stand alone program from Scicos computational functions ideal for embedded systems with limited hardware resources The stand alone program that implement the Scicos diagram con exchange data using the RTAI Lab user interface or the J RTAI Lab user interfaces The former needs RTAI installed in the host system the latter uses Java applets that runs on any Java enabled browser on any OS C RTAI Lab user interface RTAI Lab is a graphics environment that allow to display in real time the internal variables of a RTAI running controller using scopes and other virtual instruments It is also possible modify the internal parameter in order to tune the system RTAI tasks and RTAI Lab exchange data using IP address and NET RPC protocol The NET RPC uses Ethernet or other compatible media interfaces in real time using custom UDP data packet This protocol support an arbitrary number to tasks controller connected with a single supervisor This is the ideal environment for distributed control applications With some limitations on the network load the same media can support real time and standard TCP IP connection simultaneously RTAI Lab uses RTAI API to synchronize the communication and update the display with real time priority This imply the presence of an Linux
11. e 10 40 Hz function of the load inertia We choose Ts 10 ms because is a good compromise between this requirement and is realizable with the available hardware and software components in soft and hard real time modes Input amplifier data acquisition and output amplifier can be considered fully transparent With the full linearized model of the plant sys syslin c A B C D and the sampling time Ts the equivalent discrete time plant is computed with sys_d dscr sys Ts and splitted in the four elements with Ad Bd Cd Dd abcd sys_d The classic approach at the design of the state variable regulator is to compute using the expected controlled sys tem the position of the closed loop poles then and compute a feedback matrix F to do that Scilab has a function that do all the work F ppol A B poles This kind of approach has several drawbacks A better solution is to use an LQR designed feedback matrix that is more robust to parameters variations and allow to find a good compromise between the time reponse and the effort on the actuator In order to improve the steady state accuracy of the system an external discrete integrator is added in the loop The matrix of our new augmented feedback system are F F Ki Ad kA 3 Bd E With this arrangement is possible to compute the gain of the integrator Ki and the stabilizing feedback matrix F at the same time The value of R and Q are found ex
12. ers e run the controller inside the standard simulator while connected to the real plant in soft real time mode SCICOS HIL Scicos Hardware In the Loop e use an automatic code generator to create stand alone controller program that could run stand alone in hard real time mode Linux RTAI Code Generator e implement graphics user interfaces RTAI Lab J RTAI Lab to remotely interact with the controller using network connections We provide all the references to the documentation available thought Internet that describe all the details of installation configuration and utilization of the software packages involved Our aim is just to show what is possible to realize within a very limited budget we try to limit serge steer inria fr the cost of the parts using standard low cost devices and stimulate development of solutions that use imagination instead of money Just as an example the cost of all the software used is zero II SOFTWARE PACKAGES OVERVIEW One if not the most important critical issue for the developing of modern digital control systems are the software tools used for the design and validation A Scilab We use Scilab Scicos as main development environment Scilab 1 is a scientific software package for numerical computations providing a powerful open computing envi ronment for engineering and scientific applications Scilab has a complete functions toolbox for modeling simulating and aidi
13. esent time the only way that allow a full hard real time guarantee behavior is the use of ISA PCI PC CARD I O board 7 IV MODELLING AND SIMULATING THE SYSTEM A very useful way to evaluate all system constants is to write all the equations inside a single Scilab script bb _10 sce see appendix A In the first part of the script all the system constants are defined then these constants are used to calculate the coefficients of the linearized model of the system A Modelling the plant Supposing that the linear model of the plant is represented by the time continuous state matrix A B C et D the discrete sampled time model is obtained using the Scilab instruction sys_d dscr sys Ts where sys is the Scilab object that represent the linear system sys syslin c A B C D The A matrix has dimension 4 x 4 the model has four state see the annexe the matrix B is 4 x there is only an input the motor voltage C is 1 x 4 the position of the ball is the only output and D is null The eigenvalue of the the A matrix show that the lin earized model is unstable in the equilibrium point there is an eigenvalue with positive real part A full state feedback approach is used to stabilize the system The feedback matrix F is calculated using LQR method using Q and R values tuned experimentally bb_a O El i Eoo o Motor_In num z X EE Z tY ee S H SUM E OssPosError
14. ethod was effective also to design the observer C Kalman observer At not satisfied with the performances of the LQR ob server we had tried to use a steady state optimal Kalman observer The computation of the feedback matrix L is more involved because the Scilab lqe function require the preparation of an equivalent system that include input state output and cross input output correlation noise sources The performances of the Kalman observer is equivalent to the LQR observer VI Scicos RTAI CODE GENERATOR There are many situation where the performance of the Scicos HIL solution are not sufficient if the plant is un stable with catastrophic consequence for fast high accuracy systems that require sampling time lower than 4 ms for embedded system where is not possible install the full Scilab Scicos package but a small stand alone program that implement the controller is the only option The previous diagram can reused but but all the I O block must be from the specific RTAI Scicos palette 0 10 N 0 05 0 004 a SO cee a N 0 05 j Aopo T T T T T T T 0 0 15 30 45 6 0 15 90 105 120 135 150 Fig 5 Ball position m time s 1 254 1 254 2 50 T T T T T T Fig 6 Motor voltage Volt time s VII CONCLUSION In this paper we have presented a full working system that will be used for the teaching of the industrial control R
15. in the model because it introduces a time constant Te La Ra 0 83ms which can be neglected The limited electromechanical conversion efficiency of the motors appears as different Kr torque constant and Ky speed constant These two constants are equals in the case of perfect 100 efficient motor where all the electrical power is converted in mechanical work The constants are different also because inside the Lego motor there is a re duction gear that lower the efficiency but increase the torque reducing the maximum speed and improve the linearity of the DC motor in the low speed region The estimated internal motor plus reduction gear equivalent moment inertia are negligible respect the the moment of inertia of ball and beam In the model we neglect all the other possible sources of friction The motor has more than sufficient speed and torque to move the ball and the beam directly We introduce an external reduction gear to utilize better the motor speed range C Power Amplifier The output of an operational amplifier is boosted with a complementary pair Darlington transistors This linear power amplifier has more than sufficient bandwidth 10kHz voltage 15V and current 1 5A capability to drive the loaded motor The output voltage is limited to 9V to avoid motor overload that will permanently damage the motor We choose to set Ay 1 for the power amplifier because we found that the
16. ng the design of hybrid control systems Since 1994 it has been distributed freely along with the source code via the Internet It is currently used in educational and industrial environments around the world Scilab includes hundreds of mathematical functions with the possibility to add inter actively programs from various languages FORTRAN C C JAVA It has sophisticated data structures including lists polynomials rational functions linear systems an interpreter and a high level programming language Scilab Language B Scicos Writing the simulation of an hybrid dynamical system as scripts using the powerful functions of the Scilab language is possible 2 but it is time consuming and it is very easy to insert bugs during the manual coding To simplify this job Scilab include Scicos 3 a graphical hybrid dy namical system modeler and simulator toolbox Scicos is used for applications in control system communication signal processing queuing systems and to study physical and biological systems Within Scicos graphical editor it is possible to place configure and connect blocks in order to create diagrams to model hybrid dynamical systems and simulate it Most of the Scicos graphical user interface is written in Scilab language for complete integration with Scilab easy customization and maximum flexibility Each blocks is rapresented by two functions e the interfacing function Written in Scilab language define
17. ntee that the sampling time will be respected The latest 2 6 1x Linux kernel with the low latency features active and with the internal timer set to the maximum resolution lms 1000Hz could be considered real time at 99 for a sampling time of 10ms This performance is more than enough for laboratory applications with electromechanical systems where sampling time in the range of 4 20 milliseconds The respect of the sampling time is a very critical issue for a digital controller because the position of poles and zeros are function of sampling time Ts If Ts change the position of pole zero change with potential dangerous consequence in the transient steady state and stability proprieties of the closed loop hybrid system It is also important to limit the number of Scicos scopes and traces they uses CPU resources Increasing the scope s internal buffers reduces the CPU loads at the expense of the real time update of the display A Compensation of the mechanical non linearity The very cheap Lego parts create a non linear sticky friction sticktion that block the movement of the beam for very low values of the motor voltage As result of this dead zone effect the motor is totally incapable of making the small movement to place the ball in the perfect equilibrium position with zero ball speed We use a custom Scicos block to create an artificial non linearity that try to compensate the dead zone effect using a fixed
18. perimentally using R 1 and Q 1 1 1 1 1 as starting point F_lqr_i bb_dlqr Adi Bdi Qi R p F F_lgqr_i 1 4 i Ki F_lqr_i 5 C The observer Instead that fill the plant with sensors using a Luenberg observer The added integrator does not change the design of the observer because it is external at the real plant then we use the non augmented matrix We design the observer using manual placement of real poles in order to obtain an observer faster than the plant V TESTING THE CONTROLLER WITH THE REAL PLANT ScICOS HARDWARE IN THE LOOP With some limitations it is possible to use Scicos to control a real plant The first step is to create input blocks to acquire the feedback signal and output blocks to drive the system Using the examples found in the Scicos 17 and Comedi 18 documentation we have developed the interfacing and computational functions that realize the T O using Scicos blocks Within Scicos it is possible to run the simulation in real time using the menu option Simulate gt Setup and setting Realtime scaling equal 1 With this parameter set Scicos wait for the right time to read the inputs make the calculations and update the outputs It is clear that the time required to complete all the actions should be less than the sampling time Normally this is not a problems with the modern PC Unfortunately Linux is not an hard real time OS also with free CPU time there is no guara
19. rary value It is possible to stabilize the ball in a generic position We examine in detail the case x9 0 Linearizing the model around the equilibrium state 0 we obtain an approximate linear model AE Bu y where Ky K2K g KcKr z Te 0 0 7 Ra Jp t 0j 4 0 0 0jp 0 i 0 32 0 0 0 x 0 0 1 0 0 and u Uy The matrix C is 0 0 0 1 This linear model is used for the design of the regulator and the observer
20. the graphical representation the input output and control ports and signals and the user configurable parameters the computational function that is the real code used in the simulations The computational function can be written in Scilab Language for easy development or in C for maximum efficiency and speed The C com putational function can be pre compiled or compiled before simulation in order to allow the user to ccus tomize without leave the Scicos environment Scicos simulations could be used to interact with real system in many ways e Scicos HIL Hardware In the Loop With some limi tation is possible use the Scicos simulation to control a real plant in real time the internal general purpose C generator The internal Code Generator menu Object Code generation 4 translate the Scicos symbolic diagram representa tion in stand alone general purpose C code This code uses the standard I O terminal as interface It is left to the user to customize the C code in order to interact with other input output devices e g data acquisition cards and resolves the real time and synchronism issues This basic but very flexible solution is designed to work with hand coding or with external software tools as Syndex 5 For the next release of Scilab vers 5 the internal code generator will be fully integrated with Syndex to allow the physical implementation of Scicos diagram on multiprocessors an
21. voltage step to the motor in the correct direction The dead zone canceling non linearty is realized inside the block Mathematical Expression that uses the equation sign ul atul The parameter a defined in the script or in the Context represent the equivalent motor voltage of the the dead zone the minimum voltage required to move the beam The full regulator is show in Fig 4 We used a non linear block Saturation to limit the motor voltage during the starting phase during the normal working operation the voltage computed form the controller is ever below the saturation level 2V rA ii Motor Comedi D A rl Comedi D A CH 0 num z ig e z den z De si i Comedi A D A Comedi A D CH 0 SUM SUM La 097 lt e a Lo SUM fe 3 te Fig 4 Scicos Hardware In the Loop B The observer The fists attempt to manually place the poles of the observer was unsuccessful The position signal is noisy because the rail ball rail contact is not perfect This noise only partially attenuated by the single pole low pass filter of the input amplifier create stability problems in the observer After several trial and error iterations with different strategies to place the poles of the observer real poles complex conjugate poles deadbeat we found that the LQR m
Download Pdf Manuals
Related Search