transactions of the institute of measurement and control · transactions of the institute of...

http://tim.sagepub.com

Measurement and Control Transactions of the Institute of

DOI: 10.1177/014233120002200103 2000; 22; 29 Transactions of the Institute of Measurement and Control

R. Isermann Mechatronic systems: concepts and applications

http://tim.sagepub.com/cgi/content/abstract/22/1/29 The online version of this article can be found at:

Published by:

http://www.sagepublications.com

On behalf of:

The Institute of Measurement and Control

can be found at:Transactions of the Institute of Measurement and Control Additional services and information for

http://tim.sagepub.com/cgi/alerts Email Alerts:

http://tim.sagepub.com/subscriptions Subscriptions:

http://www.sagepub.com/journalsReprints.navReprints:

http://www.sagepub.com/journalsPermissions.navPermissions:

© 2000 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution. at PENNSYLVANIA STATE UNIV on February 7, 2008 http://tim.sagepub.comDownloaded from

http://www.instmc.org.uk/

http://tim.sagepub.com/cgi/alerts

http://tim.sagepub.com/subscriptions

http://www.sagepub.com/journalsReprints.nav

http://www.sagepub.com/journalsPermissions.nav


Transactions of the Institute of Measurement and Control 22,1 (2000) pp. 29–55

Mechatronic systems: concepts andapplicationsR. IsermannInstitut fur Automatisierungstechnik, Technische Universitat Darmstadt, Darmstadt,Germany

The synergetic integration of mechanical processes, microelectronics and information pro-cessing opens new possibilities to the design of processes as well as for its automatic control.The solution of tasks within mechatronic systems is performed on the process side and thedigital–electronic side. As the interrelations during the design play an important role, simul-taneous engineering has to take place from the very beginning. Mechatronic systems aredeveloped for mechanical elements, machines, vehicles and precision mechanic devices. Theintegration of mechatronic systems can be performed by the components (hardwareintegration) and by information processing (software integration). The information processingconsists of low-level and high-level feedback control, supervision and diagnosis and generalprocess management. Special signal processing, model-based and adaptive methods areapplied. With the aid of a knowledge base and inference mechanisms, mechatronic systemswith increasing intelligence will be developed. The main goals are to increase systems perform-ance, reliability and economy, and decrease production costs.

Key words: adaptive control; electronics; fault diagnosis; fuzzy control; hardware-in-the-loopsimulation; information processing; integration; mechanics; modelling.

1. Introduction

Mechanical systems are increasingly integrated with actuators, sensors and digitalelectronics. Figure 1 shows a general scheme of a modern mechanical process like

Address for correspondence: R. Isermann, Institut fur Automatisierungstechnik, Technische Univer-sitat Darmstadt, Landgraf-Georg-Str. 4, D-64283 Darmstadt, Germany.E-mail: risermannKiat.tu-darmstadt.de

2000 The Institute of Measurement and Control 0142-3312(00)TM005OA



30 Mechatronic systems

a power-producing or a power-generating machine. A primary energy flows intothe machine and is then either directly used for the energy consumer in the caseof an energy transformer or converted into another energy form in the case of anenergy converter. The form of energy can, for example, be electrical, mechanical(potential or kinetic, hydraulic, pneumatic) chemical or thermal. The primaryenergy flow originates from an energy source or energy supply and can be manipu-lated by an actuator. The energy consumer is in many cases an energy sink or/andan energy storage. Machines are mostly characterized by a continuous or periodic(repetitive) energy flow. For other mechanical processes like mechanical elementsor precision mechanical devices, piecewise or intermittent energy flows are typical.

The energy flow is generally a product of a generalized flow and a potential(effort). Information on the state of the mechanical process can be obtained frommeasured generalized flows, like speed, volume or mass flow, electrical currentor potentials like force, pressure, temperature and voltage. These measured signalsare then the basis for information processing in a digital computer operating underreal-time conditions. Together with reference variables, the measured variablesare therefore the inputs for an information flow through the digital electronicsresulting in manipulated variables for the actuators or in monitored variables ona display. Hence, the mechanical–electronic system can be considered as consistingof a feedforward energy flow and a feedback information flow.

Presently a development is taking place to integrate the mechanical systems,the actuators, sensors and microelectronics, forming an inseparable overall system.

Figure 1 Energy and information flow within a mechatronicsystem



R. Isermann 31

These integrated mechanical–electronic systems are increasingly called mech-atronic systems.

Thus, MECHAnics and elecTRONICS are conjoined. The word ‘mechatronics’was probably first created by a Japanese engineer in 1969 (Kyura and Oho, 1996).A unified definition cannot be found (compare, e.g., Schweitzer, 1989; Ovaska,1992). In the IEEE/ASME Transactions on Mechatronics a preliminarily definition isgiven: ‘Mechatronics is the synergetic integration of mechanical engineering withelectronics and intelligent computer control in the design and manufacturing ofindustrial products and processes’ (Harashima et al., 1996).

All definitions agree that mechatronics is an interdisciplinary field, in whichthe following disciplines act together (Figure 2):

1) mechanical systems (mechanical elements, machines, precision mechanics);2) electronic systems (microelectronics, power electronics, sensor and actuator

technology);3) information technology (systems theory, automation, software engineering, arti-

ficial intelligence).

Descriptions of the developments until now can be seen in Schweitzer, 1992;Gansemeiser et al., 1985; Harashima et al., 1996; Isermann, 1996. An insight intogeneral aspects are given editorially in the journals Mechatronics (1991); Mechatron-ics System Engineering (1993); IEEE/ASME Transactions on Mechatronics (1996), theconference proceedings of, e.g., IEE, Mechatronics (1990); and in the following jour-nal articles Isermann (1993), M. Hiller (1995), Kaynak et al. (1995), Luckel (1995),and in the books of Kitaura (1987), Bradley et al. (1991), McConaill et al. (1991)and Isermann (1999a).

The solution of tasks for designing mechatronic systems is performed as wellon the mechanical and on the digital–electronic side. Herewith interrelations dur-ing the design play an important role, because the mechanical system influencesthe electronic system and vice versa, the electronic system also influences thedesign of the mechanical system. This means that simultaneous engineering has totake place with the goal of also creating synergetic effects.

Figure 2 Mechatronics: synergetic integration of differentdisciplines




Mechatronic systems show up in the development of mechanical elements,machines and vehicles and precision mechanic devices. Figure 3 gives someexamples. Hence, the integration with electronics comprise many classes of techni-cal systems. In several cases the mechanical part of the process is coupled withan electrical, thermodynamical, chemical or information processing part. The rea-son is that, e.g., machines are energy converters where in addition to the mechan-ical energy other kinds of energy appear. Therefore, mechatronic systems in awider sense comprise mechanical and also nonmechanical coupled processes.However, the mechanical part should be dominating.

Because an auxiliary energy is required to change the fixed properties of for-merly passive mechanical systems by feedforward or feedback control these mech-anical systems are also called active mechanical systems.

2. Functions of mechatronic systems

Mechatronic systems permit many improved and new functions. This will be dis-cussed by considering some examples.

2.1 Division of functions between mechanics and electronics

For designing mechatronic systems the interplay for the realization of functionsin the mechanical and electronic part is crucial. Compared to pure mechanical

Figure 3 Mechanical systems with integrated electronics:examples for mechanical elements, machines and precision mech-anics



R. Isermann 33

realizations, the use of amplifiers and actuators with electrical auxiliary energy hasalready led to considerable simplifications, as can be seen from watches, electricaltypewriters and cameras. A further considerable simplification in the mechanicsresulted from introducing microcomputers in connection with decentralized elec-trical drives, e.g., for electronic typewriters, sewing machines, multi-axis handlingsystems and automatic gears.

The design of lightweight constructions leads to elastic systems which areweakly damped through the material itself. An electronic damping through pos-ition, speed or vibration sensors and electronic feedback can be realized with theadditional advantage of an adjustable damping through the algorithms. Examplesare elastic drive chains of vehicles with damping algorithms in the engine elec-tronics, elastic robots, hydraulic systems, far-reaching cranes and space construc-tions (e.g., with flywheels).

The addition of closed loop control, e.g., for position, speed or force, does notonly result in a precise tracking of reference variables, but also an approximatelinear overall behaviour, though the mechanical systems show nonlinear behav-iour. By omitting the constraint of linearization on the mechanical side, the effortin construction and manufacturing may be reduced. Examples are simple mechan-ical pneumatic and electro-mechanical actuators and flow valves with electroniccontrol.

With the aid of freely programmable reference variable generation, the adap-tation of nonlinear mechanical systems to the operator can be improved. This isalready used for the driving pedal characteristics within the engine electronics forautomobiles, telemanipulation of vehicles and aircraft and in development forhydraulic actuated excavators and electric power steering.

However, with the increasing number of sensors, actuators, switches and controlunits, the number of cables and electrical connections also increase such thatreliability, cost, weight and the required space are major concerns. Therefore thedevelopment of suitable bus systems, plug systems, and redundant and recon-figurable electronic systems are challenges for the design.

2.2 Operating properties

By applying active feedback control the precision of, e.g., a position is reached bycomparison of a programmed reference variable, and a measured control variable,and not only through the high mechanical precision of a passively feedforward-controlled mechanical element. Therefore the mechanical precision in design andmanufacturing may be reduced somewhat and more simple constructions forbearings or slideways can be used. An important aspect is the compensation ofa larger and time-variant friction by adaptive friction compensation (Isermann etal., 1992; Isermann, 1999a). Then also a larger friction on the cost of backlash maybe intended (e.g., gears with pretension), because it is usually easier to compensatefor friction than for backlash.

Model-based and adaptive control further allow an operation in more operatingpoints (wide range operation), compared to fixed control with unsatisfactory per-formance (danger of instability or sluggish behaviour). A combination of robust




and adaptive control allows a wide range operation to become possible, e.g., forflow-, force-, speed-control and for processes like engines, vehicles and aircraft.A better control performance allows the reference variables to be moved closerto constraints with improved efficiencies and yields (e.g., higher temperatures,pressures for combustion engines and turbines, compressors at stalling limits,higher tensions and higher speed for paper machines and steel mills).

2.3 New functions

Mechatronic systems allow also functions which could not be performed withoutdigital electronics. First, nonmeasurable quantities can be calculated on the basisof measured signals and can be influenced by feedforward or feedback control.Examples of this are time-dependent variables like the slip for tyres, internal tensi-tions, temperatures, the slip angle and ground speed for steering control ofvehicles or parameters like damping and stiffness coefficients, resistances. Theadaptation of parameters like damping and stiffness for oscillating systems basedon measurements of displacements or accelerations is another example. Integratedsupervision and fault diagnosis becomes more and more important with increas-ing automatic functions, increasing complexity and higher demands on reliabilityand safety. Then the triggering of redundant components, a system reconfigur-ation, maintenance on request and any kind of teleservice make the system more‘intelligent’. Table 1 summarizes some properties of mechatronic systems com-pared to conventional electro-mechanical systems.

Table 1 Properties of conventional and mechatronic designed systems

Conventional design Mechatronic design

Added components Integration of components (hardware)

1 Bulky Compact2 Complex mechanisms Simple mechanisms3 Cable problems Bus or wireless communication4 Connnected components Autonomous units

Simple control Integration by information processing(software)

5 Stiff construction Elastic construction with damping byelectronic feedback

6 Feedforward control, linear (analogue) Programmable feedback (nonlinear) digitalcontrol control

7 Precision through narrow tolerances Precision through measurement andfeedback-control

8 Nonmeasurable quantities change Control of nonmeasurable estimatedarbitrarily quantities

9 Simple monitoring Supervision with fault diagnosis10 Fixed abilities Learning abilities



R. Isermann 35

3. Methods of integration

Figure 4 shows a general scheme of a classical mechanical–electronic system. Suchsystems resulted from adding available sensors and actuators and analogue ordigital controllers to the mechanical components. The limits of this approach werethe lack of suitable sensors and actuators, the unsatisfactory lifetime under roughoperating conditions (acceleration, temperature, contamination), large spacerequirements, the required cables and relatively slow data processing. Withincreasing improvements in the miniaturization, robustness and computing powerof microelectronic components, one can now try to put more weight on the elec-tronic side and to design the mechanical part from the beginning with a view toa mechatronic overall system. Then more autonomous systems can be envisaged,e.g., in the form of capsuled units with touchless signal transfer or bus connectionsand robust microelectronics.

The integration within a mechatronic system can be performed mainly in twoways – through the integration of components and through the integration byinformation processing.

The integration of components (hardware integration) results from designingthe mechatronic system as an overall system and imbedding the sensors, actuatorsand microcomputers into the mechanical process (Figure 5). This spatial inte-gration may be limited to the process and sensor, or the process and actuator.The microcomputers can be integrated with the actuator, the process or sensor,or be arranged at several places.

Integrated sensors and microcomputers lead to smart sensors and integratedactuators, and microcomputers develop to smart actuators. For larger systems busconnections will replace the many cables. Hence, there are several possibilities forbuilding an integrated overall system by proper integration of the hardware.

The integration by information processing (software integration) is mostlybased on advanced control functions. Besides a basic feedforward and feedbackcontrol, an additional influence may take place through process knowledge andcorresponding on-line information processing (Figure 5). This means a processingof the available signals at higher levels (discussed in the next section). Thisincludes the solution of tasks like supervision with fault diagnosis, optimizationand general process management. The respective problem solutions result in real-time algorithms which must be adapted to the mechanical process properties, forexample expressed by mathematical models in the form of static characteristics,differential equations, etc. Therefore, a knowledge base is required, comprisingmethods for design and information gain, process models and performance cri-teria. In this way the mechanical parts are governed in various ways through

Figure 4 General scheme of a (classical) mechanical–electronicsystem




Figure 5 Ways of integration with mechatronic systems

higher level information processing with intelligent properties, possibly includinglearning, thus forming an integration by process-adapted software.

In the following section, the integration through information processing will beconsidered further.

4. Information processing methods

The governing of mechanical systems is usually performed through actuators forchanging of positions, speeds, flows, forces or torques, and voltages. The directlymeasurable output quantities are frequently positions, speeds, accelerations orforces and currents.



R. Isermann 37

4.1 Multi-level control systems

The information processing of direct measurable input and output signals can beorganized into several levels (Figure 6):

level 1: low level control (feedforward, feedback for damping, stabilization,linearization);

level 2: high level control (advanced feedback control strategies);level 3: supervision, incl. fault diagnosis;level 4: optimization, co-ordination (of processes);level 5: general process management.

Figure 6 Advanced intelligent automatic system with multi-control levels, knowledge base, inference mechanisms and inter-faces




Recent approaches for mechatronic systems mostly use signal processing in thelower levels, for example damping or control of motions or simple supervision.Digital information processing, however, allows the solutions of many more tasks,such as adaptive control, learning control, supervision with fault diagnosis,decisions for maintenance or even redundancy actions, economic optimization andcoordination. The tasks in the higher levels are sometimes summarized as ‘pro-cess management’.

4.2 Special signal processing

The methods described above are partially also applicable for nonmeasurablequantities which are reconstructed by using mathematical process models. In thisway it is possible to control, e.g., damping ratios, material and heat stress andslip, or to supervise quantities like resistances, capacitances, temperatures withincomponents, or parameters of wear and contamination. This signal processingmay require special filters to determine amplitudes or frequencies of vibrations,to determine derivated or integrated quantities, or state variable observers.

4.3 Model-based and adaptive systems

The information processing is, at least in the lower levels, performed by simplealgorithms or software-modules under real-time conditions. These algorithms con-tain free adjustable parameters, which have to be adapted to the static anddynamic behaviour of the process. In contrast to manual tuning by trial and errorthe use of mathematical models allows precise and fast automatic adaptation.

The mathematical models can be obtained by identification and parameter esti-mation, which use the measured and sampled input and output signals. Thesemethods are not restricted to linear models, but also allow the identification ofseveral classes of nonlinear systems. If the parameter estimation methods are com-bined with appropriate control algorithm design methods, adaptive control sys-tems result, which can be used for precise controller tuning permanently or onlyfor commissioning (Isermann et al., 1992).

4.4 Intelligent systems

The information processing within mechatronic systems may range between sim-ple control functions and intelligent control. Various definitions of intelligent con-trol systems do exist (see, e.g., Saridis, 1977; Saridis and Valavanis, 1988; Åstrom,1991; White and Sofge, 1992; Antaklis, 1994; Harris, 1994; Gupta and Sinha, 1996).An intelligent control system may be organized as an on-line expert systemaccording to Figure 6 and comprises:

1) multi control functions (executive functions);2) knowledge base;



R. Isermann 39

3) inference mechanisms;4) communication interfaces.

The on-line control functions are usually organized in multi-levels, as alreadydescribed. The knowledge base contains quantitative and qualitative knowledge.The quantitative part operates with analytical (mathematical) process models,parameter and state estimation methods, analytical design methods (e.g., for con-trol and fault detection), and quantitative optimization methods. Similar moduleshold for the qualitative knowledge, e.g., in form of rules (fuzzy and softcomputing). Further knowledge is the past history in the memory and the possi-bility of predicting the behaviour. Finally, tasks or schedules may be included.

The inference mechanism draws conclusions either by quantitative reasoning(e.g., Boolean methods) or by qualitative reasoning (e.g., possibilistic methods)and takes decisions for the executive functions.

Finally, communication between the different modules, an information manage-ment database and the man–machine interaction has to be organized.

Based on these functions of an on-line expert system an intelligent system can bebuilt up, with the ability ‘to model, reason and learn the process and its automaticfunctions within a given frame and to govern it towards a certain goal’. Hence,intelligent mechatronic systems can be developed, ranging from ‘low-degree intel-ligent’ (Isermann, 1999a), such as intelligent actuators, to ‘fairly intelligent sys-tems’, such as self-navigating automatic guided vehicles.

An intelligent mechatronic system can adapt the controller to the mostly nonlin-ear behaviour (adaptation) and store its controller parameters dependent on theposition and load (learning), supervise all relevant elements and perform a faultdiagnosis (supervision) to request for maintenance or if a failure occurs (decisionson actions). In the case of multiple components, supervision may help to switchoff the faulty component and perform a reconfiguration of the controlled process.

5. Modelling of mechatronic systems

Mathematical process models for the static and dynamic behaviour are requiredfor various steps in the design of mechatronic systems, such as, e.g., simulation,control design, reconstruction of variables (section 8). There are mainly two waysto obtain these models, the theoretical modelling based on first (physical) prin-ciples and the experimental modelling (identification) with measured input andoutput variables. A basic problem in theoretical modelling of mechatronic systemsis that the components originate from different domains. There exists a welldeveloped domain-specific knowledge for the modelling of, e.g., electrical circuits,multibody mechanical systems or hydraulic systems and also correspondingsoftware packages. However, a computer-assisted general methodology for themodelling and simulation of components from different domains is still missing(Otter and Elmqvist, 1997).

The basic principles of theoretical modelling for systems with energy flow areknown and can be unified for components from different domains as electrical,mechanical and thermal (see Paynter, 1961; MacFarlane, 1964; Wellstead, 1979;




Karnopp et al., 1990; Cellier, 1991 and Gawthrop and Smith, 1996). The modellingmethodology becomes more involved if material flows are also incorporated asfor fluids, thermodynamics and chemical processes.

A general procedure for theoretical modelling of lumped parameter processescan be sketched as follows (Isermann, 1999a):

1) definition of flowsa) energy flow (electrical, mechanical, thermal conductance processes);b) energy and material flow (fluidic, thermal transfer, thermodynamics,

chemical processes);2) definition of process elements: flow diagrams

a) sources, sinks (dissipative processes);b) storages, transformers, converters;

3) graphical representation of the process modela) multi-port diagrams (terminals, flows and potentials or across and

through variables);b) block diagrams for signal flow;c) bond graphs for energy flow;

4) statement of equations for all process elementsa) Balance equations for storages (mass, energy, momentum);b) Constitutive equations for process elements (sources, transformers,

converters);c) Phenomenological laws for irreversible processes (dissipative systems:

sinks);5) interconnection equations for the process elements

a) continuity equations for parallel connections (node law: S through-vari-ables = 0);

b) compatibility equations for serial connections (closed circuit law: S acrossvariables = 0);

6) overall process model calculationa) establishment of input and output variables;b) state space representation;c) input/output models (e.g. differential equations, transfer functions).

An example for steps 1–3 is shown in Figure 7 for a drive-by-wire vehicle.Transformers, converters, storages and sinks usually have no definite causality.However, sources show a definite causality either as potential or flow sources.Together with actuators they usually impress a potential or flow variable on thefollowing process elements and force a definite causality of the overall system.

A unified approach for processes with energy flow is known for electrical, mech-anical and hydraulic processes with incompressible fluids. Generalized throughand across variables can be defined, as shown in Table 2.

In these cases the product of the through and across variables is power. Thisunification enabled the formulation of the standard bond graph model (Karnopp etal., 1990). Also for hydraulic processes with compressible fluids and thermal pro-cesses these variables can be defined to result in powers (Table 2). However, thisis not engineering practice where mass flows and heat flows are used. If thesevariables are used, so-called pseudo bond graphs with special laws result, leaving



R. Isermann 41

Figure 7 Different schemes for an automobile (as required fordrive-by-wire-longitudinal control). (a) Scheme of the compo-nents (construction map); (b) energy flow diagram; (c) multi-portdiagram with flows and potentials; (d) signal flow diagram formulti-port

Table 2 Generalized across and through variables for processes with energy flow

System Through variables Across variables

Electrical Electric current I Electric voltage UMagnetic Magnetic flow f Magnetic flow uMechanical

Translation Force F Velocity wRotation Torque M Rotational speed v

Hydraulic Volume flow V· Pressure pThermodynamical Entropy flow s· Temperature T




the simplicity of standard bond graphs. Bond graphs lead to a high level abstrac-tion, have less flexibility and need additional effort to generate simulation algor-ithms. Therefore, they are not the ideal tool for mechatronic systems in general(Otter and Elmqvist, 1997). Also the tedious work to establish block diagramswith an early definition of causal input/output blocks is not very suitable.

As shown above, all process elements except the sources have no definite caus-ality, i.e., it is not defined what the input variables are and what the output vari-able are before connecting a special source, which can be either an effort sourceor a flow source. Therefore, in order to stay enough flexible during physical mod-elling, the models of the elements should stay noncausal as long as possible(except the sources). An appropriate approach for modelling is to have a softwarelanguage available with following properties:

1) basis process elements are described by their physical laws (differential andalgebraic equations without definite causality);

2) connection of process elements via terminals or ports, like the real system (i.e.,the models possess ‘cuts’ as interfaces);

3) inclusion of other models, e.g., obtained from experiments for elements whichcannot be modelled physically (i.e., models from identification, especially non-linear models, also in form of look-up-tables and then with specializedcausality);

4) easy user handling by adding or neglecting elements (because modelling is fre-quently an iterative procedure) and by a graphical interface.

These properties can be met by object-oriented modelling environments, likedescribed in Birtwistle et al. (1973), Elmqvist (1978), Elmqvist and Mattson (1989),and realized in modelling languages SIMULA and DYMOLA. These modellinglanguages use object-oriented programming, as applied for other complexsoftware (Wegner, 1990), characterized by objects, classes and inheritance. Withregard to object-oriented modelling, the following definitions are used (Fritzsonand Engelson, 1998):

1) objects are a collection of equations, functions with variables and parameters.They may share states which are instance variables;

2) classes are templates from which objects or subclasses can be created;3) inheritance allows us to reuse equations, functions of a class, when defining

new objects and classes.

Based on such a framework, unified object-oriented languages for physical sys-tems modelling in different domains are developed presently, like MODELICA,OMOLA and VHDL-AMS.

The following features are then implemented: encapsulation of knowledge inan object with well-developed interfaces, topological interconnection ability byplugging process model elements together, hierarchical modelling by declaringinterconnected models as new objects, object instantiation by describing genericobject classes to instantiate new objects, class inheritance to reuse already encapsu-lated knowledge, a generalized networking capability by interconnecting modelsthrough nodes, using across and through variables, and a graphical interface.



R. Isermann 43

Recent publications on these developments are Åstrom et al. (1998), Cellier et al.(1996) and Isermann (1997a).

Hence, the development of theoretical modelling of mechatronic systems with aunified transparent and flexible procedure from the basic components of differentdomains to simulation are a challenge for further development. Many componentsshow nonlinear behaviour and nondifferentiable nonlinearities like different typesof friction and backlash. For more complex process parts multidimensional map-pings (e.g., combustion engines, tyre behaviour) must also be integrated.

5.1 Identification of mechatronic systems

For the verification of theoretical models, several well known identificationmethods can be used, like correlation analysis and frequency response measure-ment, Fourier and spectral analysis. As some parameters are frequently unknownor change with time, parameter estimation methods can be applied for modelswith continuous time or discrete time, especially if the models are linear in theirparameters (Eykhoff, 1974; Isermann, 1992). For the identification and approxi-mation of nonlinear, multidimensional characteristics, artificial neural networks(multi-layer perceptrons or radial-basis functions) are flexible methods, which canbe expanded for nonlinear dynamic processes (Isermann et al., 1997, 1999b).

6. Advanced control methods for mechanical systems

Because of the integration of various functions, the use of modern tools plays animportant rule for the design of the control system if higher performances arerequired. It is proposed to consider the basic control as a knowledge-based multi-level feedback control system which is shown in Figure 8.

Figure 8 Knowledge-based multilevel feedback control formechatronic systems




The knowledge-based multilevel feedback control system is a part of the intelli-gent system of Figure 6. The knowledge base consists of mathematical processmodels, parameter estimation and controller design methods, and control per-formance criteria. The feedback control is organized in lower level and higherlevel controllers, a reference value generation module and controller parameteradaptation. With this structure the main control functions of mechatronic systemscan be organized. In this section some control principles will be considered briefly.

6.1 Lower level feedback control

The goal of the lower level feedback is to provide a certain dynamic behaviour(e.g., enforcement of damping), to compensate for nonlinearities like friction, andto reduce parameter sensitivity. Some examples are:

1) damping of high frequent oscillations – weakly damped higher frequent oscil-lations appear, e.g., in multimass drive chains. The damping can generally beimproved by high pass filtering the outputs and using a state variable feedbackor PD (proportional-derivative) feedback. Figure 9 shows a scheme of anadaptive damping feedback with parameter estimation.

2) compensation of nonlinear static characteristics – nonlinear static characteristics arepresent in many subsystems of mechanical processes. Figure 10 shows as anexample the position control for a nonlinear actuator. Frequently a first nonlin-earity appears in the force or torque-generating part, like an electromagnet ora pneumatic or hydraulic actuator, where, e.g., the force FD = f(U) follows anonlinear static characteristic. This nonlinearity can now be compensated forby an inverse characteristic U = f−1(U9) such that the I/O-behaviour FD = f(U9) becomes approximately linear (Isermann and Raab, 1993),and a linear (PID-type) controller GC1 can be applied.

3) friction compensation – for many mechanical systems the friction can bedescribed approximately by

FF±(t) = fFC± sign Y· (t) + fFy± Y· (t) uY· (t)u . 0

Figure 9 Adaptive damping feedback of drive chain oscillations(usually a selection of vM, vG and vD is fed back)



R. Isermann 45

Figure 10 Adaptive position control of a nonlinear actuator

where fFC is the Coulomb friction and fFy the linear viscous friction coefficientwhich may be dependent on the motion direction, indicated by + or −.

The Coulomb friction has an especially strongly negative effect on the controlperformance, if a high positioning accuracy is required, because it leads to a hys-teresis effect. When the process stops within the hysteresis width before the setpoint is reached, only the integral part of the position controller can compensatefor the offset. This may yield a significant loss of control performance and accu-racy, especially during small position changes.

The basic idea of friction compensation is to compensate the relay function ofthe Coulomb friction by adding an adequate compensation signal UFC to thenormal control action. Different methods such as dithering, feedforward compen-sation and adaptive friction compensation are alternatives (Isermann, 1999a; Iser-mann and Raab, 1993; Tomizuka, 1995).

After compensating for the nonlinear friction, the position controller GC2 can bedesigned to control the underlying linearized force control and the remainingmechanical process without friction (Figure 10). In simple cases, a linear controllerof PID-type or a state controller is sufficient.

An alternative for position control of nonliner actuators is the use of a slidingmode controller. It consists of a nominal part for feedback linearization and anadditional feedback to compensate for model uncertainties (Utkin, 1977; Slotineand Weiping, 1991). The resulting chattering near the included switching functiongenerates a dither signal. A comparison of a fixed PID-controller with frictioncompensation and a sliding mode controller for an electromagnetic actuator is,e.g., shown by Pfeufer et al. (1995). The sliding mode controller resulted in a goodrobustness in response to changing process parameters on the cost of higherdesign and computational effort.




6.2 Higher level feedback control

The task of the higher level controller is to generate a good overall dynamic behav-iour with regard to the servo dynamics function due to changes of the positionreference W(t), and to compensate for disturbances stemming from, e.g., variationsof the load forces FL(t) (see Figure 10). This high level controller may be realizedas a parameter-optimized controller of PID-type or a state controller with or with-out state observer. A state observer is required if only the position Y(t) is measur-able. If both, Y(t), and Y·(t) can be measured Y(t) can be obtained by differen-tiation of Y·(t) (if required at all ) such that no state observer is needed (seeIsermann et al., 1995).

The control scheme may be expanded by additional feedback controllers froma load or working process which is coupled with the mechanical process shownin Figure 7, resulting in a multiple cascaded control system.

6.3 Adaptive control

A prerequisite for the application of advanced control algorithms is the use ofwell adapted process models. This then leads to self-tuning or adaptive controlsystems:

1) Parameter scheduling – parameter scheduling based on the measurement of vary-ing operation conditions is an effective method for dealing with known andapproximately time-invariant process nonlinearities. Supposing auxiliary vari-ables V are measurable, and correlate well with the process changes, the adap-tation of the controller parameters G is performed as a function of V(parameter schedules).

2) Parameter-adaptive control systems – parameter-adaptive control systems arecharacterized by using identification methods for parametric process models.This is indicated in the adaptation level of Figure 8. Parameter estimation hasproven to be an appropriate basis for the adaptive control of mechanical pro-cesses, including the adaptation to nonlinear characteristics, Coulomb friction,and unknown parameters such as masses, stiffness and damping (see Isermannand Raab, 1993; Isermann et al., 1995). These digital adaptive control systemswork well if the assumptions for their design and convergence are satisfied.This includes, e.g., proper excitation of the process dynamics. For the caseswhere the assumptions are violated, a supervision level is required, whichtakes appropriate action.

6.4 Fuzzy control

The development of fuzzy logic theory (Zadeh, 1972) stimulated alternative waysto solve automatic control problems. Based on these basic ideas, fuzzy controllerswere proposed (Mamdani and Assilian, 1975) that describe human control inlinguistic form. As fuzzy logic provides a systematic framework with which to



R. Isermann 47

treat vague variables and information, it should be applied primarily if sensorsyield imprecise outputs, the process behaviour is only qualitatively known or theautomation functions cannot be described by equations or Boolean logic. As dis-cussed in Isermann (1997b, 1998), the potentials of fuzzy logic approaches in gen-eral increase with higher automation levels, because the degree of the qualitativeknowledge and the required intelligence grow with the hierarchical level.

The static and dynamic behaviour of most mechanical systems can be ratherprecisely described by mathematical process models obtained through theoreticalmodelling and identification methods. Hence, there is in many cases no need toapply fuzzy concepts for the control of mechanical systems in the lower levels.However, fuzzy control concepts may be of interest for:

1) fuzzy-tuning and adaptation of classical controllers;2) fuzzy quality and comfort control;3) fuzzy control for special (abnormal) operating conditions.

Particularly for reference value generation of underlying (classical) control sys-tems (Figure 8), where the quality or comfort and therefore the human receptionplays a role, fuzzy rule-based methods offer interesting possibilities, i.e., for thehigher control levels. Examples for such mechatronic systems are:

1) the comfort control of suspensions in passenger cars;2) the comfort of start-up of automobiles with clutch manipulation, gear shifting

and automatic transmission;3) distance and velocity control of automobiles and elevators.

For a more detailed description see Isermann (1997b, 1998).

7. Supervision and fault detection

With the increasing number of automatic functions (autonomy) inlcuding elec-tronic components, sensors and actuators, increasing complexity and increasingdemands on reliability and safety, integrated supervision with fault diagnosisbecomes more and more important. This is therefore a significant natural featureof an intelligent mechatronic system. Figure 11 shows a process influenced byfaults. These faults indicate unpermitted deviations from normal states and canbe generated either externally or internally. External faults are for example causedby the power supply, contamination or collision, internal faults by wear, missinglubrication, actuator or sensor faults. The classical method for fault detection isthe limit value checking of a few measurable variables. However, incipient andintermittant faults cannot usually be detected and an in-depth fault diagnosis isnot possible with this simple approach. Therefore, model-based fault detectionand diagnosis methods were developed in recent years, allowing early detectionof small faults with normally measured signals, and also in closed loops(Isermann, 1997b). Based on measured input signals U(t) and output signals Y(t),and process models, features are generated by, e.g., parameter estimation, stateand output observers, and parity equations (Figure 11).

These residuals are then compared with the residuals for normal behaviour and,




Figure 11 Scheme for a model-based fault detection

with change detection methods, analytical symptoms are obtained. Then a faultdiagnosis is performed via methods of classification or reasoning.

A considerable advantage is that the same process model can be used for boththe (adaptive) controller design and the fault detection. In general, continuoustime models are preferred if fault detection is based on parameter estimation orparity equations. For fault detection with state estimation and also parity equa-tions, discrete-time models can be used.

Advanced supervision and fault diagnosis is a basis for improving reliabilityand safety, state dependent maintenance, triggering of redundancies and recon-figuration.

8. Design procedure for mechatronic systems

The design of mechatronic systems requires a systematic development and use ofmodern design tools.

Table 3 shows five important development steps for mechatronic systems, start-ing from a purely mechanical system and resulting in a fully integrated mech-atronic system. Depending on the type of the mechanical system, the intensity ofthe single development steps is different. For precision mechanical devices fairlyintegrated mechatronic systems already exist. The influence of the electronics onmechanical elements may be considerable, as shown by adaptive dampers, anti-lock system brakes and automatic gears. However, complete machines andvehicles show first a mechatronic design of their elements and then slowly a



R. Isermann 49

Table 3 Steps in the design of mechatronic systems

redesign of parts of the overall structure as can be observed in the developmentof machine tools, robots and vehicle bodies.

8.1 Software tools for the design

The computer aided development of mechatronic systems comprises:

1) constructive specification in the engineering development stage using CAD-and CAE-tools;

2) model building for obtaining static and dynamic process models;3) transformation into computer codes for system simulation;




4) programming and implementation of the final mechatronic software.

Some software tools are described for example in Otter and Gruebel (1993).A broad range of CAD/CAE tools is available for two- and three-dimensionalmechanical design, such as Auto CAD with a direct link to CAM (computer-aidedmanufacturing); PADS for multi-layer printed-circuit board layout, etc. However,computer-aided modelling is not as advanced (see Section 5). Object-oriented lan-guages such as DYMOLA and MOBILE for modelling of large combined systemsare described in Otter and Gruebel (1993), Elmqvist (1993) and Hiller (1995).These packages are based on specified ordinary differential equations, algebraicequations and discontinuities. A recent description of computer-aided control sys-tem design can be found in James et al. (1995). For system simulation (and control-ler design) a variety of program systems exist, like ACSL, SIMPACK,MATLAB/SIMULINK, MATRIX-X. These simulation techniques are valuabletools for design, as they allow the study of the interaction of components, andthe variations of design parameters before manufacturing. However, they are notgenerally suitable for real-time simulation.

8.2 Real-time simulation

For the design of mechatronic systems, real-time simulation is increasingly applied.This is particulary required if the process, the hardware and the software aredeveloped simultaneously in order to minimize iterative development cycles andto meet a short time to market. With regard to the required speed of the compu-tation, simulation methods can be subdivided into:

1) simulation without (hard) time limitation;2) real-time simulation;3) simulation faster than real-time.

Some application examples are given in Figure 12. Herewith, real-time simul-ation means that the simulation of a component is performed such that the inputand output signals show the same time-dependent values as the real dynamicallyoperating component. This becomes a computational problem for processes whichhave fast dynamics compared with the required algorithms and calculation speed.

Different types of real-time simulation methods are shown in Figure 13. Thereason for the real-time requirement is mostly that one part of the investigatedsystem is not simulated, but real. Three cases can be distinguished:

1) the real process can be operated together with the simulated control by usinganother hardware than the final hardware. This is also called ‘control prototyp-ing’;

2) the simulated process can be operated with the real control hardware, which iscalled ‘hardware-in-the-loop simulation’;

3) the simulated process is run with the simulated control in real-time. This may berequired if the final hardware is not available or if a design step before thehardware-in-the-loop simulation is considered.



R. Isermann 51

Figure 12 Classification of simulation methods with regard tothe speed and application examples

Figure 13 Classification of real-time simulation

In the following, the hardware-in-the-loop simulation and control prototypingis considered.

8.2.1 Hardware-in-the-loop simulation: The hardware-in-the-loop simulation(HIL) is characterized by operating real components in connection with real-timesimulated components. Usually, the control system hardware and software is thereal system, as used for series production. The controlled process, consisting ofactuators, physical processes and sensors, can then either be simulated or parts




of it may be real components (Figure 14a). In general, mixtures of the cases shownare realized. Frequently some actuators are real, and the process and the sensorsare simulated. The reason is that actuators and the control hardware very oftenform one integrated subsystem or that actuators are difficult to model preciselyand to simulate in real time. (The use of real sensors together with a simulatedprocess may require considerable realization efforts, because the physical sensorinput does not exist and must be generated artificially). In order to change orredesign some functions of the control hardware or software a bypass unit canbe connected to the basic control hardware. Hence, hardware-in-the-loop simu-lators may also contain partially simulated (emulated) control functions.

The advantages of the hardware-in-the-loop simulation are generally:

1) design and testing of the control hardware and software without operating areal process (moving the process field into the laboratory);

2) testing of the control hardware and software under extreme environmentalconditions in the laboratory (e.g., high/low temperature, high accelerationsand mechanical shocks, aggressive media, electromagnetic compatibility);

3) testing of the effects of faults and failures of actuators, sensors and computerson the overall system;

4) operating and testing of extreme and dangerous operating conditions;5) reproducible experiments, frequently repeatable;6) easy operation with different man–machine interfaces (cockpit design and

training of operators);7) saving of cost and development time.

Figure 14 Realtime simulation: hybrid structures. (a) Hardware-in-the-loop simulation; (b) control prototyping



R. Isermann 53

8.2.2 Control prototyping: For the design and testing of complex control sys-tems and their algorithms under real-time constraints, a real-time controller simul-ation (emulation) with hardware (e.g., an off-the-shelf signal processor) other thanthe final series production hardware (e.g., special ASICS) may be performed. Theprocess, the actuators and sensors can then be real. This is called control prototyp-ing (Figure 14b). However, in this case also, parts of the process or actuators maybe simulated, resulting in a mixture of HIL-simulation and control prototyping.The advantages are mainly:

1) early development of signal processing methods, process models and controlsystem structure including algorithms with high level software and high per-formance off-the-shelf hardware;

2) testing of signal processing and control systems together with other design ofactuators, process parts and sensor technology, in order to create synergeticeffects;

3) reduction of models and algorithms to meet the requirements of cheaper massproduction hardware;

4) defining the specifications for final hardware and software.

Some of the advantages of HIL-simulation also hold for control prototyping.Some references for real-time simulation are Hanselmann (1993) and Isermann etal. (1999).

9. Conclusions

This contribution has shown that mechatronic systems cover a great variety oftechnical processes, and require the integration of components and informationprocessing during the design. Several disciplines like mechanics, electronics andinformation technology are merged together to create synergetic effects. Severalof the discussed developments can be observed in existing products of precisionengineering and show up presently with increasing tendency for mechanicalcomponents, machines, automobiles and aircraft.

The intensive use of computer-aided design, theoretical and experimental mod-elling, digital simulation, hardware-in-the-loop simulation, control systemsdesign, verification, software engineering, etc., and the interrelations with thebasic process design require methods and tools developed in the control com-munity. Therefore, the development of mechatronic systems is a real challengefor control engineering.

References

Antaklis, P. 1994: Defining intelligent control.IEEE Control Systems, June, 4–66.

Åstrom, K.J. 1991: Intelligent control. Proceed-ings of the European Control Conference, Greno-ble.

Åstrom, K.J., Elmqvist, H. and Mattson, S.E.1998: Evolution of continuous-time modeling

and simulation. 12th European SimulationMulticonference, ESM ’98, 16–19 June 1998,Manchester, UK.

Birtwistle, G.B., Dahl, O.J., Myhrhaug, B. andNygaar, K. 1973: Simula begin. AuerbachPublishers.

Bradley, D.A., Dawson, D., Burd, D. and




Loader, A.J. 1991: Mechatronics – electronics inproducts and processes. London: Chapmanand Hall.

Cellier, F.E. 1991: Continuous system modelling.Berlin: Springer.

Cellier, F.E., Elmqvist, H. and Otter, M. 1996:Modeling from physical principles. In Lev-ine, W.S., editor, The control handbook. BocaRaton: CRC Press, 9–107.

Elmqvist, H. 1978: A structured model langu-age for large continuous systems, Ph.D. dis-sertation. Report CODEN: LUTFD2/(TFRT-1015) Department of Automatic Control,Lund Institute of Technology, Sweden.

—— 1993: Object-oriented modeling and auto-matic formula manipulation in Dymola.Scandinavian Simulation Society SIMs, June,Kongsberg.

Elmqvist, H. and Mattson, S.E. 1989: Simulatorfor dynamical systems using graphics andequations for modeling. IEEE Control SystemsMagazine 9, 53–58.

Eykhoff, P. 1974: System identification. London:John Wiley.

Fritzon, P. and Engelson, V. 1998: Modelica –a unified object-oriented language for systemmodeling and simulation. 12th European Con-ference on Object-Oriented Programming(ECOOP ’98), 20–24 July, Brussels, Belgium.

Gansemeiser, J., Brexel, D., Frank, T. andHumpert, A. 1995: Integrated product devel-opment. Third Conference on Mechatronics andRobotics, Paderborn, Germany, 4–6 October,Teubner, Stuttgart.

Gawthrop, F.E. and Smith, L. 1996: Metamodel-ling: bond graphs and dynamic systems. Lon-don: Prentice Hall.

Gupta, M.M. and Sinha, N.K. 1996: Intelligentcontrol systems. New York: IEEE Press.

Hanselmann, H. 1993: Hardware in-the-loopsimulation as a standard approach for devel-opment, customization, and production test.SAE 930207, 1–5 March.

Harashima, F., Tomizuka, M. and Fukuda, T.1996: Mechatronics – What is ist, why andhow? An editorial, IEEE/ASME Transactionson Mechatronics 1, 1–4.

Harris, C.J., editor. 1994: Advances in intelligentcontrol. London: Taylor & Francis.

Hiller, M. 1995: Modelling, simulation andcontrol design for large and heavy manipu-lators. Proceedings of the International Confer-ence on Recent Advances in Mechatronics, 14–16 August, Istanbul, Turkey.

IEE, Mechatronics: Design intelligentmachines. 1990: Proceedings of the IEE Inter-

national Conference 12–13 September, Univer-sity of Cambridge.

IEEE/ASME Transactions on Mechatronics.1996: Piscataway: IEEE Press.

Isermann, R. 1992: Identifikation dynamischerSysteme, 2nd edition, Vols 1 and 2. Berlin:Springer.

—— 1993: Towards intelligent control of mech-anical processes. Control Engineering Practice1, 232–52.

—— 1996: Modeling and design methodologyfor mechatronic systems. IEEE/ASME Trans-actions on Mechatronics 1, 16–28.

——, editor. 1997a: Workshop RechnergestutzteModellbildung dynamischer Systeme, 29October, Techn. Univ. Darmstadt.

—— 1997b: Supervision, fault detection andfault diagnosis methods – an introduction.Special section on supervission, fault detec-tion and diagnosis. Control Engineering Prac-tice.

—— 1998: On fuzzy logic applications for auto-matic control, supervision, and fault diag-nosis. IEEE Transactions on System, Men, andCybernetics – Part A 28, 221–35.

—— 1999a: Mechatronische Systeme. Berlin:Springer.

—— 1999b: Modeling, identification and simul-ation of mechatronic systems. IFAC-Congress, Beijing. Amsterdam: Elsevier.

Isermann, R. and Raab, U. 1993: Intelligentactuators. Automatica 29, 1315–31.

Iserman, R., Ernst, S. and Nelles, O. 1997:Identification with dynamic neural net-works – architectures, comparisons, appli-cations. Plenary, IFAC Symposium on SystemIdentification (SYSID), Fukuoka, Japan.

Isermann, R., Keller, H. and Raab, U. 1995:Intelligent actuators, Chapter 21. In: Gupta,N.N. and Sinha, N.K., editors, Intelligent con-trol systems. Piscataway: IEEE Press.

Isermann, R., Lachmann, K.-H. and Matko, D.1992: Adaptive control systems. London: Pren-tice Hall.

Isermann, R., Schaffnit, J. and Sinsel, S. 1999:Hardware-in-the-loop simulation for thedesign and testing of engine control systems.Control Engineering Practice 7, 643–53.

James, J., Cellier, F., Pang, G., Gray, J. andMattson, S.E. 1995: The state of computer-aided control system design (CACSD). IEEETransactions on Control Systems, special issue,6–7 April.

Karnopp, D.C., Margolis, D.L. and Rosenberg,R.C. 1990: System dynamics: a unified approach.New York: Wiley.



R. Isermann 55

Kaynak, O., Ozkan, M., Bekiroglu, N. andTunay, I. editors. 1995: Recent advances inmechatronics, Proceedings of the InternationalConference on Recent Advances in Mechatronics,14–16 August, Istanbul, Turkey.

Kitaura, K. 1987: Industrial mechatronics, NewEast Business Ltd., in Japanese.

Kyura, N. and Oho, H. 1996: Mechatronics – anindustrial perspective. IEEE/ASME Trans-actions on Mechatronics 1, 10–15.

Luckel, J., editor. 1995: Third Conference onMechatronics and Robotics, Paderborn, Ger-many, 4–6 October, Stuttgart: Teubner.

MacFarlane, A.G.J. 1964: Engineering systemsanalysis. Cambridge, MA: G.G. Harrop.

Mamdani, E.H. and Assilian, S. 1995: Anexperiment in linguistic synthesis with afuzzy logic controller. International JournalMan–Machine Studies 7, 1–13.

McConaill, P.A., Drews, P. and Robrock, K.H.1991: Mechatronics and robotics I. Amsterdam:IOS Press.

Mechatronics: An International Journal. Aimsand scope. 1991. Oxford: Pergamon Press.

Mechatronics System Engineering: Inter-national Journal on Design and Application ofIntegrated Electromechanical Systems. 1993.Netherlands: Kluwer.

Otter, M., and Elmqvist, H. 1997: Energy flowmodeling of mechatronic systems via objectdiagrams. Proceedings of the 2nd MATHMOD,Vienna, 705–10.

Otter, M. and Gruebel, G. 1993: Direct physi-cal modeling and automatic code generationfor mechatronics simulation. 2nd Conferenceon Mechatronics and Robotics. Duisburg, 27–29September, IMECH, Moers.

Ovaska, S.J. 1992: Electronics and informationtechnology in high range elevator systems.Mechatronics 2, 89–99.

Paynter, H.M. 1961: Analysis and design ofengineering systems. Cambridge, MA: MITPress.

Pfeufer, T., Landsiedel, T. and Isermann, R.1995: Identification and model-based nonlin-ear control of electromechanical actuatorswith friction. IFAC-Workshop “Motion Con-trol”, October, Munchen, Germany.

Saridis, G.N. 1977: Self-organizing control ofstochastic systems. New York: Marcel Dekker.

Saridis, G.N. and Valavanis, K.P. 1988: Ana-lytical design of intelligent machines. Auto-matica 24: 123–33.

Schweitzer, G. 1989: Mechatronik-Aufgaben undLosungen, VDI-Berichte No. 787. Dusseldorf:VDI-Verlag.

—— 1992: Mechatronics – a concept withexamples in active magnetic bearings. Mech-atronics 2, 65–74.

Slotine, J.J.E. and Weiping, L. 1991: Appliednonlinear control. Chapter 7: Sliding Control,London: Prentice Hall.

Tomizuka, M. 1995: Robust digital motion con-troller for mechanical systems. Proceedings ofInternational Conference on Recent Advances inMechatronics, 14–16 August, Istanbul, Tur-key.

Utkin, V.I. 1977: Variable structure systemswith sliding mode: a survey. IEEE Trans-actions on Automatic Control 22, 212–22.

Wegener, P. 1990: Concepts and paradigma ofobject-oriented programming. OOPS Messen-ger 1, 9–87.

Wellstead, P.E. 1979: Introduction to physicalsystem modelling. London: Academic Press.

White, D.A. and Sofge, D.A., editors. 1992:Handbook of intelligent control. New York: VanNorstrand Reinhold.

Zadeh, L. 1972: A rationale for a fuzzy control.Journal of Dynamic Systems, Measurement andControl 94, Series G, 3–4.



transactions of the institute of measurement and control · transactions of the institute of...

Documents