Robot kollarının adaptif kontrolü

Abstract

"Robot" kelimesi ilk olarak 1900'lerln başında bir tiyatro oyununda kullanılmıştır. Burada robotlar insanlar kadar zeki, insan tipinde ve yorulmak nedir bilmeyen maki- nalar olarak tasarlanmışlardır. II. Dünya savaşından sonra robot üzerinde ilk çalış malar başlamıştır. 1960 'lı ve 1970 'li yıllarda İse büyük bir ivme kazanmıştır. îlk başlarda PID türü basit kontro lörler kullanılırken, günümüzde robust (gürbüz) ve adaptif gibi karmaşık kontrol algoritmaları geliştirilmiştir ve ro bot kollarına uygulanmaktadır. Tezin birinci bölümünde robot kolları üzerinde yapı lan çalışmaların tarihçesi ve adaptif kontrol hakkında kısa bir bilgi verilmiştir. îkinci bölümde ise, robot kolu kon trolünde yararlandığımız, robot kolu hareket denklemlerinin özellikleri geniş bir biçimde tanıtılmıştır. üçüncü bölümde, robot kolu kontrolünde kullanılan klasik kontrol yöntemleri tanıtılmıştır. Bunun yanında, günümüzde kullanılan karmaşık kontrol algoritmalarına temel oluşturan yapılar anlatılmıştır. Dördüncü bölümde, geliş tirilen bazı adaptif kontrol algoritmaları verilmiştir. Beşinci bölümde, Cumhur Başpınar ile birlikte bir gürbüz -adapt if yöntem geliştirilmiş ve incelenmiştir. Altıncı bölümde, tez boyunca anlatılan bazı kontrol kurallarının simülasyon sonuçları gösterilmiştir.

Computer based automation is increasing due to its quality and productivity. Most automated manufac turing tasks are carried out by special -purpose machines which generally have high costs. Because of disadvan tages, automation machines called "ROBOT" was improved. Robots are capable of performing a variety of manufac turing functions in a more flexible working environment and at lower production costs. The word Robot originated from the Czech word robota, meaning work. Webster's dictionary defines robot as an automatic device that performs functions ordinarily ascribed to human beings. A defination used by the Robot Institute of America gives a more precise description of industrial robots: A robot is a reprog rammable multi-functional manipulator designed to move materials, parts, tools or specialized devices, through variable programmed motions for the performance of a variety of tasks. Briefly, a robot is a reprogrammable general -purpose manipulator with external sensors that can perform various assembly tasks. (See Chapter I) An industrial robot is a general -purpose, com puter controlled manipulator consisting of several rigit links connected in series by revolute or prismatic Join ts. One end of the chain is attached to a supporting base, while the other end is free and equipped with a tool to manipulate objects or perform assembly tasks. The motion of the joints result in relative motion of the links. Mechanically, a robot is composed of an arm (or mainframe) and a wrist subassembly plus a tool. It is designed to reach a workpiece located within its work volume. vi The word robot was introduced into the English language in 1921 by the play-wright Karel Capek in his satirical drama. In this work, robots are machines that resemble people, but work tirelessly. Early work lead ing to today's industrial robots can be traced to the period immediately following Wold War II. During the late 1940' s research programs were started at the Oak Ridge and Argonne National Laboratories to develop remotely controlled mechanical manipulators for handling radio-active materials. These systems were of the master-slave type, designed to reproduce faithfully hand and arm motions made by a human operator. Later, force feedback was added by mechanically coupling the motion of the master and slave units so that the operator could feel the forces as they developed between the slave manipuator and its environment. In the mid 1950' s the mechanical coupling was replaced by electric and hydrau lic power in manipulators. In the mid 1950' s George C. Devol developed a device whose operation could be pro grammed (and thus changed) and which could follow a sequence of motion steps determined by the Instructions in the program. Unlike hard automation machines, these robots could be reprogrammed and retooled at relative low cost to perform other jobs as manufacturing require ments changed. It became evident in the I960' s that the flexibility of these machines could be enhanced signifi cantly by the use of sensory feedback. H.A.Ernst [1962] reported the development of a computer-controlled mecha nical hand with tactile sensors. This device, called the MH-1, could feel blocks and use this information to control the hand so that it stacked the blocks without operator assistance. During the 1970' s a great deal of research work focused on the use of external sensors to facilitate manipulative operations. Bejczy [1974], at the Jet Propulsion Laboratory, implemented a computer- based torque control technique on his extended Stanford arm for space exploration projects. Since then, various control methods have been proposed for servoing mechani cal manipulators. One of these methods is adaptive control. Adaptive control is a fascinating field for research. It is also of increasing practical importance, since adaptive techniques are being used more and more in industrial control systems. However, the field is not mature, there are still many unsolved therotical and practical issues. vii In everyday language, to "adapt" means to change a behaviour to conform to new circumstances. An adaptive regulator is a regulator that can modify its behaviour in response to changes in the dynamics of the process and the disturbances. At one symposium in 1961, the following defination was suggested: " An adaptive system is any physical system that has been designed with an adaptive view point." Actually, adaptive con trol is a special type of nonlinear feedback control. In the early 1950' s there was extensive rese arch on adaptive control, in connection with the design of autopilots for high performance aircraft. Such air craft operate over a wide range of speeds and altitudes. It was found that ordinary constant-gain, linear feed back control could work well in one operating condition, but that changed operating conditions led to difficul ties. A more sophisticated regulator, which could work well over a wide range of operating conditions. This was adaptive regulator. In the I960' s many contributions to control theory were important for the development of adaptive control. Dynamic programming increased the understand ing of adaptive processes. There were also major devel opments in system identification and in parameter esti mation. There was a renaissance of adaptive control in the 1970's. In the late 1970' s and early 1980' s correct proofs for stability of adaptive systems appeared, albeit under very restrictive assumptions. There are strong ties to nonlinear system theory. Because adaptive systems are inherently nonlin ear. One way to look at adaptive systems is to view them as a combination of parameter estimation and cont rol. All adaptive systems are implemented using digital computers. Rapid and revolutionary progress in micro electronics has made it possible to implement adaptive regulators simply, and cheaply. There have been a number of applications of adaptive feedback control since the mid 1950 's. The early experiments were used with analog implementatio ns, but there were hardware problems. Systems implem ented using minicomputers appeared in the early 1970 's. The number of the applications has increased with the viil advent of the microprocessor. A great number of indus trial control loops are today under adaptive control. These include a wide range of applications in aerospace, process control, ship steering, robotics, and other industrial control systems. MRAC and STR theory are not applicable to plants which are nonlinear systems like robot manipu lators. These are unsuitable for direct application to the manipulator control problem. MRAC and STR theory are based on low parameter changes. However modern manipulators move sufficiently fast that, for example, the efficient joint may change by 300% in a friction of a second. Previous work on adaptive control of manip ulators almost completely ignores the issue of persis tent excitation. Previous work seems to have been moti vated by desire to avoid computation in the control computer. That is, in lieu of computing the complex dynamic model, it was desired to use some adaptive scheme to track changes in an assumed linear plant. This is not a reasonable objective. Because; 1) Resulting designs have no theoretical stability proof. Because, the underlying theory is valid only for constant unknown parameters. 2) Computing the dynamic model of a manipu lator as a part of the control law should no longer be regarded as prohibitive. Dubowsky and Des Forges were perhaps the first to propose the application of MRAC to the manip ulator control problem. Their adaptation scheme is based on linear decoupled models, one per joint. They state that the underlying theory is valid only if the manipulators changes configuration slowly relative to the adaptation rate. ix Horowitz and Tomizuka have proposed a scheme that takes some of the manipulators dynamics in to acco unt. They write the dynamics with portions depending only on manipulator position treated as unknown paramec ters that are adapt ively identified. The fate at which the manipulator changes configuration must be low compa red to the adaptation time constant in order for the theory to be valid. Koivo have used discrete linear time invariant decoupled models, one per joint, to model the dynamics of the manipulator. Based on this assumption, an adap tive controller is based on the STR theory. Another group of work, which sometimes goes under the name of adaptive control of manipulators, is based on sliding-mode or variable-structure system. In these schemes, parameters are not identified; so they may not be called adaptive control schemes. In this control theory, the switching signal is injected in to the control torque to achieve good tracing despite poor knowledge of parameters. Such chattering controls can not be implemented and the derivative of serious error goes to zero only in the mean. Such high frequency control action may also excite unmodeled dynamics. To date, the majority of work on the problem of adaptively controlling a mechanical manipulator has simply been the application to the manipulator system of methods that were developed for linear systems. Such methods do not lead to an overall adaptive scheme that can be rigorously proven stable. It has been the aim of many researchers to use adaptation as a means to simpli fy or otherwise avoid doing the nonlinear model computa tions in the computer control. As computing power beco mes more available and schemes to compute the dynamic model are improved, this motivation diminishes. Craig introduced a model based control of mechanical manipulators. Such controller suppresses disturbances and tracks desired trajectories uniformly in all configuration of the manipulator. This desi- reble performance is contingent on two assumptions. First, the dynamic model of the manipulator must be computed quicly enough so that discretization effects do not degrade performance relative to the continuous time, zero delay ideal. Second, the values of parameters appearing in the dynamic model in the control law must match the parameters of the actual system if the benefi cial decoupling and linearizing effects of computed torque servo are to be realized. (See Chapter IV) Slotlne introduced another parameter adaptive controller which has more adventageous than Craig's. Craig's method requires numerical differentiation of the joint velocities to obtain estimates of joint accelera tions and inversion of the estimated inertia matrix. Futhermore, Craig's control law must ensure that the matrix remains invertible during the adaptation pro cess. (For more detail, see Chapter IV) In Chapter V, Spong's Robust Adaptive Control Law and a Robust -Adapt ive control law are introduced. (For more detail, see Chapter V) In Chapter VI, there are simulation results of some control techniques introduced in this thesis.